ref #13: in-memory metadata repositories (S02)

Normalize chat ids to a single UUID form so locks, repository keys, and mount paths cannot diverge through path-like aliases.

.gitignore

@@ -14,3 +14,5 @@ wheels/
 !docs
 !AGENTS.md
 !tasks.md
+
+opencode.json

AGENTS.md

@@ -41,16 +41,16 @@
 - Keep HTTP models and middleware inside `adapter/http/fastapi/`
 
 ## Workflow
-- Use `tasks.md` for planning
+- Use dir `tasks/` for planning
 - Do not use Beads
 - Do not use `bd`
 - Use `uv` for Python commands and dependency management
-- Do not create commits on your own
-- Work on one task at a time
 - Prefer delegation for implementation
-- Delegate only one task at a time
+- After implementation, run `Code-Reviewer` agent
-- After one task return to the user with result verification and next options
+- Pass errors to `test-engineer` agent to capture
-- Wait for the user before the next task commit or fix
+- Delegate `Feature-Developer` agent fix the errors
+- Repeat the cycle until no errors remain
+- Ensure all tests pass
 
 ## Makefile
 - `make install` install deps with `uv`
@@ -83,3 +83,6 @@
 - Prefer explicit wiring over magic
 - Do not expand scope without user approval
 - Do not `from __future__ import annotations`
+- Do not `from importlib import import_module`
+- Do not `importlib`
+- Do not `cast`

Makefile

@@ -31,7 +31,7 @@ test:
 	uv run pytest -v
 
 lint:
-	uv run ruff check .
+	uv run ruff check . --fix
 
 typecheck:
 	uv run mypy .

README.md

@@ -1,36 +1,268 @@
-Это шаблон Python-сервиса на чистой архитектуре с заменяемым web-слоем, типизированным конфигом, явным dependency wiring и observability через порты.
+# master-service
 
-## Что это за проект
+`master-service` — это control-plane сервис для sandbox-контейнеров с AI-агентом.
+Он поднимает и переиспользует sandbox на чат, подключает рабочие volume, восстанавливает state после рестарта и отдает наружу минимальный HTTP API под `/api/v1`.
 
-- Небольшой референсный сервис со слоями `domain/`, `usecase/`, `repository/` и `adapter/`
+Важно: в локальном `config/app.yaml` исторически еще стоят template-имена `web-python-skelet`.
-- Шаблон для сервисов на FastAPI, где FastAPI остается только во внешнем HTTP adapter
+Если хочешь, чтобы `/health` и OTel service name локально тоже показывали `master-service`, переопредели:
-- Проект, где конфиг собирается из `config/app.yaml`, `.env` и env vars в одно дерево dataclass-конфигов
+- `APP_NAME=master-service`
-- Проект, где repository и usecase создаются один раз на старте приложения в composition root
+- `APP_OTEL_SERVICE_NAME=master-service`
-- Проект, где логи, метрики и трейсы скрыты за интерфейсами и могут работать через `stdout`, файл или OpenTelemetry runtime
 
-## Основные идеи
+Сервис реализован на Python с Clean Architecture:
+- `domain/` — сущности и доменные ошибки
+- `usecase/` — сценарии приложения и порты
+- `repository/` — реализации repository
+- `adapter/` — HTTP, config, DI, Docker runtime и observability
 
-- Clean Architecture и границы SOLID
+## Что умеет сейчас
-- Направление зависимостей только внутрь
+
-- Тонкие adapter-слои и явная сборка зависимостей
+Текущий sandbox MVP покрывает:
-- Заменяемый HTTP-слой
+- `GET /api/v1/health`
-- Observability без протекания OpenTelemetry во внутренние слои
+- `POST /api/v1/create` с `chat_id: UUID`
+- одну активную sandbox на чат
+- reuse активной sandbox до истечения TTL
+- cleanup просроченных sandbox в фоне
+- startup reconciliation по Docker labels после рестарта сервиса
+- chat mount `rw`, dependencies mount `ro`, lambda-tools mount `ro`
+- логи, метрики и трейсы через порты `Logger`, `Metrics`, `Tracer`
+
+Пока вне scope:
+- auth и access control
+- p2p/WebSocket lease
+- workspace/chat CRUD API
+- central DB, artifacts, S3, quota и retention policy
+
+## Как устроен проект
+
+- FastAPI живет только во внешнем adapter слое
+- Docker живет только во внешнем adapter слое
+- конфиг собирается из `config/app.yaml`, `.env` и env vars в один dataclass tree
+- repository и usecase создаются один раз на старте в `adapter/di/container.py`
+- observability не протекает во внутренние слои через OpenTelemetry SDK
+
+## Структура
+
+- `domain/` — core model и domain errors
+- `usecase/` — use cases и interfaces
+- `repository/` — in-memory и другие repository implementations
+- `adapter/config/` — typed config models и loader
+- `adapter/docker/` — Docker sandbox runtime
+- `adapter/observability/` — logger/metrics/tracer runtime factory
+- `adapter/otel/` — OpenTelemetry adapters
+- `adapter/di/` — composition root
+- `adapter/http/fastapi/` — app, middleware, schemas, routers
+- `adapter/sandbox/` — sandbox reconciliation logic
+- `config/` — YAML config files
+- `docs/` — ADR и проектные гайды
 
 ## Быстрый старт
 
+### Требования
+
+- Python 3.13
+- `uv`
+- локальный Docker daemon
+- секреты `APP_API_TOKEN` и `APP_SIGNING_KEY`
+
+### Установка
+
 ```bash
 make install
+```
+
+### Локальный запуск
+
+```bash
 APP_API_TOKEN=local-api-token APP_SIGNING_KEY=local-signing-key make run
 ```
 
-Приложение стартует на `http://0.0.0.0:8123` и публикует versioned API под `/api/v1`.
+Это поднимет сам API, но для успешного `POST /api/v1/create` локально нужен еще рабочий sandbox runtime:
+
+- Docker daemon должен быть доступен по `docker.base_url`
+- образ `sandbox.image` должен существовать локально
+- директории `sandbox.dependencies_host_path` и `sandbox.lambda_tools_host_path` должны существовать
+
+В дефолтном `config/app.yaml` это значит:
+
+```bash
+mkdir -p var/sandbox/dependencies var/sandbox/lambda-tools
+docker image inspect ai-agent:latest >/dev/null
+```
+
+Если у тебя нет готового `ai-agent:latest`, проще начать с Docker Compose smoke path ниже.
+
+После старта сервис доступен на:
+- `http://127.0.0.1:8123/api/v1/health`
+
+Проверка health:
+
+```bash
+curl http://127.0.0.1:8123/api/v1/health
+```
+
+Создание или reuse sandbox:
+
+```bash
+curl -X POST http://127.0.0.1:8123/api/v1/create \
+  -H 'Content-Type: application/json' \
+  -d '{"chat_id":"11111111-1111-1111-1111-111111111111"}'
+```
+
+Пример ответа:
+
+```json
+{
+  "session_id": "3701cfe3-e05e-48af-8385-442dcd954ca2",
+  "chat_id": "11111111-1111-1111-1111-111111111111",
+  "container_id": "64d839c6007de9396ee08ad4af4a22a59a6410ec5f4892a9277a87eb49c3ff5d",
+  "status": "running",
+  "expires_at": "2026-04-02T21:11:38.292893Z"
+}
+```
+
+## Запуск через Docker Compose
+
+Для локального smoke-run есть `docker-compose.yml`.
+Он поднимает:
+- `app`
+- `docker-engine` в режиме Docker-in-Docker
+- `otel-collector`
+
+При этом `app` получает compose-specific config из:
+- `config/docker-compose.yml`
+
+Запуск:
+
+```bash
+make compose-up
+```
+
+Проверка:
+
+```bash
+make compose-ps
+make compose-logs
+```
+
+Остановка:
+
+```bash
+make compose-down
+```
+
+Важно:
+- в `config/docker-compose.yml` сейчас для smoke-проверки стоит `sandbox.image: nginx:1.27-alpine`
+- для реального agent runtime замени `sandbox.image` на образ своего sandbox/agent контейнера
+- в compose auth env vars нужны для startup config, но текущий MVP API еще не проверяет request token
+
+## Как конфигурировать
+
+### Источники конфига
+
+Конфиг собирается в таком порядке:
+1. базовый YAML из `config/app.yaml`
+2. значения из `.env`
+3. process env vars поверх `.env`
+
+То есть env vars имеют наивысший приоритет.
+
+### Обязательные секреты
+
+Нужны всегда:
+- `APP_API_TOKEN`
+- `APP_SIGNING_KEY`
+
+Сейчас это startup config, а не активная request auth для `/api/v1/create` и `/api/v1/health`.
+То есть в текущем MVP токен не нужно передавать в HTTP headers для вызова этих endpoint.
+
+### Основные секции YAML
+
+В `config/app.yaml` и `config/docker-compose.yml` есть секции:
+- `app`
+- `http`
+- `logging`
+- `metrics`
+- `tracing`
+- `otel`
+- `docker`
+- `sandbox`
+- `security`
+
+### Полезные env overrides
+
+Чаще всего полезны:
+
+#### Общие
+- `APP_NAME`
+- `APP_ENV`
+- `APP_HTTP_HOST`
+- `APP_HTTP_PORT`
+
+#### Логирование и observability
+- `APP_LOGGING_LEVEL`
+- `APP_LOGGING_OUTPUT`
+- `APP_LOGGING_FORMAT`
+- `APP_LOGGING_FILE_PATH`
+- `APP_METRICS_ENABLED`
+- `APP_TRACING_ENABLED`
+- `APP_OTEL_SERVICE_NAME`
+- `APP_OTEL_LOGS_ENDPOINT`
+- `APP_OTEL_METRICS_ENDPOINT`
+- `APP_OTEL_TRACES_ENDPOINT`
+
+#### Docker runtime
+- `APP_DOCKER_BASE_URL`
+
+#### Sandbox
+- `APP_SANDBOX_IMAGE`
+- `APP_SANDBOX_TTL_SECONDS`
+- `APP_SANDBOX_CLEANUP_INTERVAL_SECONDS`
+- `APP_SANDBOX_CHATS_ROOT`
+- `APP_SANDBOX_DEPENDENCIES_HOST_PATH`
+- `APP_SANDBOX_LAMBDA_TOOLS_HOST_PATH`
+- `APP_SANDBOX_CHAT_MOUNT_PATH`
+- `APP_SANDBOX_DEPENDENCIES_MOUNT_PATH`
+- `APP_SANDBOX_LAMBDA_TOOLS_MOUNT_PATH`
+
+#### Security
+- `APP_API_TOKEN_HEADER`
+- `APP_API_TOKEN`
+- `APP_SIGNING_KEY`
+
+### Что важно в sandbox config
+
+- `docker.base_url` — адрес Docker daemon
+- `sandbox.image` — образ sandbox контейнера
+- `sandbox.ttl_seconds` — TTL sandbox
+- `sandbox.cleanup_interval_seconds` — частота cleanup loop
+- `sandbox.chats_root` — корень chat directories
+- `sandbox.dependencies_host_path` — host path для dependency cache
+- `sandbox.lambda_tools_host_path` — host path для read-only lambda-tools
+- `sandbox.chat_mount_path` — путь внутри sandbox для chat volume
+- `sandbox.dependencies_mount_path` — путь внутри sandbox для dependency cache
+- `sandbox.lambda_tools_mount_path` — путь внутри sandbox для lambda-tools
+
+## Основные команды
+
+- `make install` — установить зависимости через `uv`
+- `make run` — локальный запуск
+- `make run-otel` — запуск с OTel endpoints из env
+- `make test` — `pytest`
+- `make lint` — `ruff`
+- `make typecheck` — `mypy`
+- `make pre-commit` — lint + typecheck + test
+- `make compose-build` — собрать compose images
+- `make compose-up` — поднять локальный stack
+- `make compose-down` — остановить stack
+- `make compose-logs` — смотреть логи
+- `make compose-ps` — смотреть статус сервисов
 
 ## Документация
 
 ### Гайды
 
 - [Правила проекта и ограничения для агента](AGENTS.md)
-- [Кодстайл проекта для AI-агента](docs/CODESTYLE.md)
+- [Кодстайл проекта](docs/CODESTYLE.md)
 - [Чистая архитектура, SOLID, DIP, Protocol и repository](docs/CLEAN_ARCHITECTURE_RU.md)
 - [Логи, метрики и трейсы в этом проекте](docs/OBSERVABILITY_RU.md)
 - [Как чистая архитектура реализована здесь](docs/PROJECT_GUIDE_RU.md)
@@ -43,43 +275,24 @@ APP_API_TOKEN=local-api-token APP_SIGNING_KEY=local-signing-key make run
 - [003 Observability Via Interfaces](docs/003-observability-via-interfaces.md)
 - [004 Versioned HTTP API](docs/004-versioned-http-api.md)
 - [005 Early FastAPI OTel Instrumentation](docs/005-fastapi-otel-early-instrumentation.md)
+- [006 MVP Docker Sandbox Orchestration](docs/006-mvp-docker-sandbox-orchestration.md)
+- [007 Startup Sandbox Reconciliation](docs/007-startup-sandbox-reconciliation.md)
+- [008 Sandbox Lifecycle Observability](docs/008-sandbox-lifecycle-observability.md)
 
-## Структура проекта
+## Для AI-агента
 
-- `domain/` - core-сущности и доменные ошибки
+Если ты меняешь проект как AI-агент, сначала прочитай:
-- `usecase/` - прикладные сценарии и порты
-- `repository/` - реализации repository
-- `adapter/config/` - загрузка и модели типизированного конфига
-- `adapter/observability/` - выбор runtime для logger, metrics и tracer
-- `adapter/otel/` - OpenTelemetry adapters
-- `adapter/di/` - composition root и singleton wiring
-- `adapter/http/fastapi/` - HTTP-схемы, dependencies, middleware и routers
-- `config/` - YAML-конфиг приложения и локального OTel collector
 
-## Для ИИ
+1. [AGENTS.md](AGENTS.md)
+2. [docs/CODESTYLE.md](docs/CODESTYLE.md)
+3. [docs/PROJECT_GUIDE_RU.md](docs/PROJECT_GUIDE_RU.md)
+4. [docs/CLEAN_ARCHITECTURE_RU.md](docs/CLEAN_ARCHITECTURE_RU.md)
+5. [docs/OBSERVABILITY_RU.md](docs/OBSERVABILITY_RU.md)
+6. релевантные ADR в `docs/`
+7. [tasks.md](tasks.md)
 
-Если ты AI-агент и собираешься что-то менять в проекте, сначала прочитай документы в таком порядке:
+Главные правила:
-
+- сначала определи слой изменения
-1. [Правила проекта и ограничения агента](AGENTS.md) - обязательные правила работы в этом репозитории
+- зависимости только внутрь
-2. [Кодстайл проекта для AI-агента](docs/CODESTYLE.md) - границы слоев, стиль кода и правила зависимостей
+- не тащи FastAPI и OpenTelemetry во внутренние слои
-3. [Как чистая архитектура реализована здесь](docs/PROJECT_GUIDE_RU.md) - практическая карта проекта и типовые сценарии изменений
+- архитектурные решения сверяй с ADR
-4. [Чистая архитектура, SOLID, DIP, Protocol и repository](docs/CLEAN_ARCHITECTURE_RU.md) - базовые архитектурные принципы и примеры
-5. [Логи, метрики и трейсы в этом проекте](docs/OBSERVABILITY_RU.md) - читать перед любыми изменениями в observability, middleware и runtime wiring
-6. [ADR в `docs/`](docs/001-composition-root-and-lifetimes.md) - читать релевантные решения перед изменением архитектуры или startup wiring
-7. [План задач и история работ](tasks.md) - понять, что уже сделано, что отложено и какие ограничения были зафиксированы
-
-Перед началом работы:
-
-- Определи, в каком слое будет изменение: `domain/`, `usecase/`, `repository/` или `adapter/`
-- Убедись, что зависимости идут только внутрь
-- Не тащи FastAPI и OpenTelemetry во внутренние слои
-- Сначала изучи существующий код в нужной директории, потом вноси изменения
-- Если задача затрагивает архитектурное решение, сначала сверяйся с ADR и проектными правилами
-
-## Запуск и команды
-
-- Для локального запуска нужны `APP_API_TOKEN` и `APP_SIGNING_KEY`
-- `make run` запускает приложение локально
-- `make run-otel` запускает приложение с локальными OTel endpoints из env vars
-- `make pre-commit` запускает `ruff`, `mypy` и `pytest`
-- `make compose-up` поднимает приложение и локальный LGTM stack через Docker Compose

adapter/config/loader.py

@@ -8,10 +8,12 @@ from dotenv import dotenv_values
 from .model import (
     AppConfig,
     AppSectionConfig,
+    DockerConfig,
     HttpConfig,
     LoggingConfig,
     MetricsConfig,
     OtelConfig,
+    SandboxConfig,
     SecurityConfig,
     TracingConfig,
 )
@@ -38,6 +40,8 @@ def load_config(
     logging_section = _section(yaml_data, 'logging')
     metrics_section = _section(yaml_data, 'metrics')
     tracing_section = _section(yaml_data, 'tracing')
+    docker_section = _section(yaml_data, 'docker')
+    sandbox_section = _section(yaml_data, 'sandbox')
     security_section = _section(yaml_data, 'security')
 
     logging_output = _yaml_or_env_choice(
@@ -128,6 +132,16 @@ def load_config(
             enable_metrics=metrics_enabled,
             enable_tracing=tracing_enabled,
         ),
+        docker=DockerConfig(
+            base_url=_yaml_or_env_str(
+                docker_section,
+                'base_url',
+                'docker.base_url',
+                env_values,
+                'APP_DOCKER_BASE_URL',
+            )
+        ),
+        sandbox=_load_sandbox_config(sandbox_section, env_values),
         security=SecurityConfig(
             token_header=_yaml_or_env_str(
                 security_section,
@@ -221,6 +235,77 @@ def _optional_section(data: Mapping[str, object], name: str) -> dict[str, object
     return section
 
 
+def _load_sandbox_config(
+    section: Mapping[str, object],
+    env: Mapping[str, str],
+) -> SandboxConfig:
+    return SandboxConfig(
+        image=_yaml_or_env_str(
+            section,
+            'image',
+            'sandbox.image',
+            env,
+            'APP_SANDBOX_IMAGE',
+        ),
+        ttl_seconds=_yaml_or_env_int(
+            section,
+            'ttl_seconds',
+            'sandbox.ttl_seconds',
+            env,
+            'APP_SANDBOX_TTL_SECONDS',
+        ),
+        cleanup_interval_seconds=_yaml_or_env_int(
+            section,
+            'cleanup_interval_seconds',
+            'sandbox.cleanup_interval_seconds',
+            env,
+            'APP_SANDBOX_CLEANUP_INTERVAL_SECONDS',
+        ),
+        chats_root=_yaml_or_env_str(
+            section,
+            'chats_root',
+            'sandbox.chats_root',
+            env,
+            'APP_SANDBOX_CHATS_ROOT',
+        ),
+        dependencies_host_path=_yaml_or_env_str(
+            section,
+            'dependencies_host_path',
+            'sandbox.dependencies_host_path',
+            env,
+            'APP_SANDBOX_DEPENDENCIES_HOST_PATH',
+        ),
+        lambda_tools_host_path=_yaml_or_env_str(
+            section,
+            'lambda_tools_host_path',
+            'sandbox.lambda_tools_host_path',
+            env,
+            'APP_SANDBOX_LAMBDA_TOOLS_HOST_PATH',
+        ),
+        chat_mount_path=_yaml_or_env_str(
+            section,
+            'chat_mount_path',
+            'sandbox.chat_mount_path',
+            env,
+            'APP_SANDBOX_CHAT_MOUNT_PATH',
+        ),
+        dependencies_mount_path=_yaml_or_env_str(
+            section,
+            'dependencies_mount_path',
+            'sandbox.dependencies_mount_path',
+            env,
+            'APP_SANDBOX_DEPENDENCIES_MOUNT_PATH',
+        ),
+        lambda_tools_mount_path=_yaml_or_env_str(
+            section,
+            'lambda_tools_mount_path',
+            'sandbox.lambda_tools_mount_path',
+            env,
+            'APP_SANDBOX_LAMBDA_TOOLS_MOUNT_PATH',
+        ),
+    )
+
+
 def _load_otel_config(
     data: Mapping[str, object],
     env: Mapping[str, str],

adapter/config/model.py

@@ -40,6 +40,24 @@ class OtelConfig:
     metric_export_interval: int
 
 
+@dataclass(frozen=True, slots=True)
+class DockerConfig:
+    base_url: str
+
+
+@dataclass(frozen=True, slots=True)
+class SandboxConfig:
+    image: str
+    ttl_seconds: int
+    cleanup_interval_seconds: int
+    chats_root: str
+    dependencies_host_path: str
+    lambda_tools_host_path: str
+    chat_mount_path: str
+    dependencies_mount_path: str
+    lambda_tools_mount_path: str
+
+
 @dataclass(frozen=True, slots=True)
 class SecurityConfig:
     token_header: str
@@ -55,4 +73,6 @@ class AppConfig:
     metrics: MetricsConfig
     tracing: TracingConfig
     otel: OtelConfig
+    docker: DockerConfig
+    sandbox: SandboxConfig
     security: SecurityConfig

adapter/di/container.py

@@ -1,24 +1,32 @@
 from collections.abc import Mapping
 from dataclasses import dataclass, field
+from datetime import UTC, datetime, timedelta
 from pathlib import Path
 
+import docker
+from docker import DockerClient
+
 from adapter.config.loader import load_config
 from adapter.config.model import AppConfig
+from adapter.docker.runtime import DockerSandboxRuntime
 from adapter.observability.factory import build_observability
 from adapter.observability.runtime import ObservabilityRuntime
-from domain.user import User
+from adapter.sandbox.reconciliation import SandboxSessionReconciler
-from repository.user import InMemoryUserRepository
+from repository.sandbox_lock import ProcessLocalSandboxLifecycleLocker
-from usecase.user import GetUser
+from repository.sandbox_session import InMemorySandboxSessionRepository
+from usecase.interface import Clock
+from usecase.sandbox import CleanupExpiredSandboxes, CreateSandbox
 
 
 @dataclass(frozen=True, slots=True)
 class AppRepositories:
-    user: InMemoryUserRepository
+    sandbox_session: InMemorySandboxSessionRepository
 
 
 @dataclass(frozen=True, slots=True)
 class AppUsecases:
-    get_user: GetUser
+    create_sandbox: CreateSandbox
+    cleanup_expired_sandboxes: CleanupExpiredSandboxes
 
 
 @dataclass(slots=True)
@@ -27,16 +35,30 @@ class AppContainer:
     observability: ObservabilityRuntime
     repositories: AppRepositories
     usecases: AppUsecases
+    sandbox_reconciler: SandboxSessionReconciler = field(repr=False)
+    _docker_client: DockerClient = field(repr=False)
     _is_shutdown: bool = field(default=False, init=False, repr=False)
 
     def shutdown(self) -> None:
         if self._is_shutdown:
             return
 
-        try:
+        self._is_shutdown = True
-            self.observability.shutdown()
+        errors: list[Exception] = []
-        finally:
+
-            self._is_shutdown = True
+        for action in (self._docker_client.close, self.observability.shutdown):
+            try:
+                action()
+            except Exception as exc:
+                errors.append(exc)
+
+        if errors:
+            raise ExceptionGroup('shutdown failed', errors)
+
+
+class SystemClock(Clock):
+    def now(self) -> datetime:
+        return datetime.now(tz=UTC)
 
 
 def build_container(
@@ -54,17 +76,46 @@ def build_container(
         )
 
     observability = build_observability(app_config)
+    clock = SystemClock()
+    docker_client = docker.DockerClient(base_url=app_config.docker.base_url)
 
-    user_repository = InMemoryUserRepository(
+    sandbox_repository = InMemorySandboxSessionRepository()
-        observability.tracer, [User(id='123', email='aza@gglamer.ru', name='gglamer')]
+    sandbox_locker = ProcessLocalSandboxLifecycleLocker()
+    sandbox_runtime = DockerSandboxRuntime(
+        app_config.sandbox,
+        docker_client,
+        observability.metrics,
+        observability.tracer,
     )
-    repositories = AppRepositories(user=user_repository)
+    sandbox_reconciler = SandboxSessionReconciler(
+        state_source=sandbox_runtime,
+        registry=sandbox_repository,
+        logger=observability.logger,
+        metrics=observability.metrics,
+        tracer=observability.tracer,
+    )
+
+    repositories = AppRepositories(sandbox_session=sandbox_repository)
     usecases = AppUsecases(
-        get_user=GetUser(
+        create_sandbox=CreateSandbox(
-            repository=user_repository,
+            repository=sandbox_repository,
+            locker=sandbox_locker,
+            runtime=sandbox_runtime,
+            clock=clock,
             logger=observability.logger,
+            metrics=observability.metrics,
             tracer=observability.tracer,
-        )
+            ttl=timedelta(seconds=app_config.sandbox.ttl_seconds),
+        ),
+        cleanup_expired_sandboxes=CleanupExpiredSandboxes(
+            repository=sandbox_repository,
+            locker=sandbox_locker,
+            runtime=sandbox_runtime,
+            clock=clock,
+            logger=observability.logger,
+            metrics=observability.metrics,
+            tracer=observability.tracer,
+        ),
     )
 
     return AppContainer(
@@ -72,4 +123,6 @@ def build_container(
         observability=observability,
         repositories=repositories,
         usecases=usecases,
+        sandbox_reconciler=sandbox_reconciler,
+        _docker_client=docker_client,
     )

adapter/docker/runtime.py

@@ -0,0 +1,315 @@
+import time
+from datetime import datetime
+from pathlib import Path
+from uuid import UUID
+
+from docker import DockerClient
+from docker.errors import DockerException, NotFound
+from docker.types import Mount
+
+from adapter.config.model import SandboxConfig
+from domain.error import SandboxError, SandboxStartError
+from domain.sandbox import SandboxSession, SandboxStatus
+from usecase.interface import Metrics, SandboxRuntime, Span, Tracer
+
+SANDBOX_LABELS = ('session_id', 'chat_id', 'expires_at')
+
+
+class DockerSandboxRuntime(SandboxRuntime):
+    def __init__(
+        self,
+        config: SandboxConfig,
+        client: DockerClient,
+        metrics: Metrics,
+        tracer: Tracer,
+    ) -> None:
+        self._config = config
+        self._client = client
+        self._metrics = metrics
+        self._tracer = tracer
+
+    def create(
+        self,
+        *,
+        session_id: UUID,
+        chat_id: UUID,
+        created_at: datetime,
+        expires_at: datetime,
+    ) -> SandboxSession:
+        started_at = time.perf_counter()
+        result = 'error'
+
+        with self._tracer.start_span(
+            'adapter.docker.create_sandbox',
+            attrs={
+                'chat.id': str(chat_id),
+                'session.id': str(session_id),
+            },
+        ) as span:
+            try:
+                try:
+                    chat_path = self._chat_path(chat_id)
+                    dependencies_path = self._readonly_host_path(
+                        self._config.dependencies_host_path
+                    )
+                    lambda_tools_path = self._readonly_host_path(
+                        self._config.lambda_tools_host_path
+                    )
+                    chat_path.mkdir(parents=True, exist_ok=True)
+                    container = self._client.containers.run(
+                        self._config.image,
+                        detach=True,
+                        labels=self._labels(session_id, chat_id, expires_at),
+                        mounts=self._mounts(
+                            chat_path,
+                            dependencies_path,
+                            lambda_tools_path,
+                        ),
+                    )
+                except (DockerException, OSError, ValueError) as exc:
+                    raise SandboxStartError(str(chat_id)) from exc
+
+                container_id = str(getattr(container, 'id', '')).strip()
+                if not container_id:
+                    raise SandboxStartError(str(chat_id))
+
+                result = 'created'
+                span.set_attribute('container.id', container_id)
+                span.set_attribute('sandbox.result', result)
+                return SandboxSession(
+                    session_id=session_id,
+                    chat_id=chat_id,
+                    container_id=container_id,
+                    status=SandboxStatus.RUNNING,
+                    created_at=created_at,
+                    expires_at=expires_at,
+                )
+            except Exception as exc:
+                span.set_attribute('sandbox.result', result)
+                span.record_error(exc)
+                self._metrics.increment(
+                    'sandbox.runtime.error.total',
+                    attrs=_runtime_error_metric_attrs('create', _error_type(exc)),
+                )
+                raise
+            finally:
+                self._metrics.record(
+                    'sandbox.runtime.create.duration_ms',
+                    _duration_ms(started_at),
+                    attrs=_runtime_metric_attrs('create', result),
+                )
+
+    def stop(self, container_id: str) -> None:
+        started_at = time.perf_counter()
+        result = 'error'
+
+        with self._tracer.start_span(
+            'adapter.docker.stop_sandbox',
+            attrs={'container.id': container_id},
+        ) as span:
+            try:
+                container = self._client.containers.get(container_id)
+                _set_span_container_attrs(span, container)
+                container.stop()
+                result = 'stopped'
+                span.set_attribute('sandbox.result', result)
+            except NotFound:
+                result = 'not_found'
+                span.set_attribute('sandbox.result', result)
+                return
+            except DockerException as exc:
+                span.set_attribute('sandbox.result', result)
+                span.record_error(exc)
+                self._metrics.increment(
+                    'sandbox.runtime.error.total',
+                    attrs=_runtime_error_metric_attrs('stop', type(exc).__name__),
+                )
+                raise SandboxError('sandbox_stop_failed') from exc
+            finally:
+                self._metrics.record(
+                    'sandbox.runtime.stop.duration_ms',
+                    _duration_ms(started_at),
+                    attrs=_runtime_metric_attrs('stop', result),
+                )
+
+    def list_active_sessions(self) -> list[SandboxSession]:
+        started_at = time.perf_counter()
+        result = 'error'
+
+        with self._tracer.start_span(
+            'adapter.docker.list_active_sandboxes',
+        ) as span:
+            try:
+                try:
+                    containers = self._client.containers.list(
+                        filters={'label': list(SANDBOX_LABELS)}
+                    )
+                except DockerException as exc:
+                    raise SandboxError('sandbox_list_failed') from exc
+
+                sessions: list[SandboxSession] = []
+                for container in containers:
+                    session = self._session_from_container(container)
+                    if session is None:
+                        continue
+                    sessions.append(session)
+
+                result = 'listed'
+                span.set_attribute('sandbox.container_count', len(containers))
+                span.set_attribute('sandbox.active_count', len(sessions))
+                span.set_attribute('sandbox.result', result)
+                return sessions
+            except Exception as exc:
+                span.set_attribute('sandbox.result', result)
+                span.record_error(exc)
+                self._metrics.increment(
+                    'sandbox.runtime.error.total',
+                    attrs=_runtime_error_metric_attrs('list_active', _error_type(exc)),
+                )
+                raise
+            finally:
+                self._metrics.record(
+                    'sandbox.runtime.list_active.duration_ms',
+                    _duration_ms(started_at),
+                    attrs=_runtime_metric_attrs('list_active', result),
+                )
+
+    def _labels(
+        self,
+        session_id: UUID,
+        chat_id: UUID,
+        expires_at: datetime,
+    ) -> dict[str, str]:
+        return {
+            'session_id': str(session_id),
+            'chat_id': str(chat_id),
+            'expires_at': expires_at.isoformat(),
+        }
+
+    def _mounts(
+        self,
+        chat_path: Path,
+        dependencies_path: Path,
+        lambda_tools_path: Path,
+    ) -> list[Mount]:
+        return [
+            Mount(
+                target=self._config.chat_mount_path,
+                source=str(chat_path),
+                type='bind',
+            ),
+            Mount(
+                target=self._config.dependencies_mount_path,
+                source=str(dependencies_path),
+                type='bind',
+                read_only=True,
+            ),
+            Mount(
+                target=self._config.lambda_tools_mount_path,
+                source=str(lambda_tools_path),
+                type='bind',
+                read_only=True,
+            ),
+        ]
+
+    def _chat_path(self, chat_id: UUID) -> Path:
+        chats_root = self._host_path(self._config.chats_root)
+        chat_path = (chats_root / str(chat_id)).resolve(strict=False)
+        if not chat_path.is_relative_to(chats_root):
+            raise ValueError('invalid chat path')
+        return chat_path
+
+    def _readonly_host_path(self, path_value: str) -> Path:
+        host_path = self._host_path(path_value)
+        if not host_path.exists():
+            raise ValueError('invalid host path')
+        return host_path
+
+    def _session_from_container(self, container: object) -> SandboxSession | None:
+        container_id = str(getattr(container, 'id', '')).strip()
+        labels = getattr(container, 'labels', None)
+        if not container_id or not isinstance(labels, dict):
+            return None
+
+        try:
+            session_id = UUID(labels['session_id'])
+            chat_id = UUID(labels['chat_id'])
+            created_at = self._container_created_at(container)
+            expires_at = _parse_datetime(labels['expires_at'])
+        except (KeyError, TypeError, ValueError):
+            return None
+
+        return SandboxSession(
+            session_id=session_id,
+            chat_id=chat_id,
+            container_id=container_id,
+            status=SandboxStatus.RUNNING,
+            created_at=created_at,
+            expires_at=expires_at,
+        )
+
+    def _container_created_at(self, container: object) -> datetime:
+        attrs = getattr(container, 'attrs', None)
+        if not isinstance(attrs, dict):
+            reload_container = getattr(container, 'reload', None)
+            if callable(reload_container):
+                reload_container()
+                attrs = getattr(container, 'attrs', None)
+
+        if not isinstance(attrs, dict):
+            raise ValueError('invalid container attrs')
+
+        raw_created_at = attrs.get('Created')
+        if not isinstance(raw_created_at, str):
+            raise ValueError('invalid created_at')
+
+        return _parse_datetime(raw_created_at)
+
+    def _host_path(self, path_value: str) -> Path:
+        return Path(path_value).expanduser().resolve(strict=False)
+
+
+def _parse_datetime(value: str) -> datetime:
+    normalized = f'{value[:-1]}+00:00' if value.endswith('Z') else value
+    return datetime.fromisoformat(normalized)
+
+
+def _duration_ms(started_at: float) -> float:
+    return (time.perf_counter() - started_at) * 1000
+
+
+def _runtime_metric_attrs(operation: str, result: str) -> dict[str, str]:
+    return {
+        'operation': operation,
+        'result': result,
+    }
+
+
+def _runtime_error_metric_attrs(
+    operation: str,
+    error_type: str,
+) -> dict[str, str]:
+    return {
+        'operation': operation,
+        'error.type': error_type,
+    }
+
+
+def _error_type(error: Exception) -> str:
+    if isinstance(error.__cause__, Exception):
+        return type(error.__cause__).__name__
+    return type(error).__name__
+
+
+def _set_span_container_attrs(span: Span, container: object) -> None:
+    labels = getattr(container, 'labels', None)
+    if not isinstance(labels, dict):
+        return
+
+    session_id = labels.get('session_id')
+    if isinstance(session_id, str) and session_id:
+        span.set_attribute('session.id', session_id)
+
+    chat_id = labels.get('chat_id')
+    if isinstance(chat_id, str) and chat_id:
+        span.set_attribute('chat.id', chat_id)

adapter/http/fastapi/app.py

@@ -1,5 +1,7 @@
-from collections.abc import Callable
+import asyncio
+from collections.abc import Awaitable, Callable
 
+from fastapi import FastAPI
 from opentelemetry.instrumentation.fastapi import FastAPIInstrumentor
 
 from adapter.config.loader import load_config
@@ -8,9 +10,10 @@ from adapter.di.container import AppContainer, build_container
 from adapter.http.fastapi.dependencies import APP_CONFIG_STATE, APP_CONTAINER_STATE
 from adapter.http.fastapi.middleware import register_middleware
 from adapter.http.fastapi.routers.v1.router import router as v1_router
-from fastapi import FastAPI
 
 API_V1_PREFIX = '/api/v1'
+APP_CLEANUP_TASK_STATE = 'cleanup_task'
+APP_CLEANUP_STOP_STATE = 'cleanup_stop'
 
 
 def create_app(config: AppConfig | None = None) -> FastAPI:
@@ -22,6 +25,7 @@ def create_app(config: AppConfig | None = None) -> FastAPI:
         app = FastAPI(title=app_config.app.name)
         setattr(app.state, APP_CONFIG_STATE, app_config)
         setattr(app.state, APP_CONTAINER_STATE, container)
+        app.add_event_handler('startup', _build_startup_handler(app, container))
         app.add_event_handler('shutdown', _build_shutdown_handler(app, container))
         register_middleware(app, app_config)
         app.include_router(v1_router, prefix=API_V1_PREFIX)
@@ -43,19 +47,112 @@ def create_app(config: AppConfig | None = None) -> FastAPI:
         raise
 
 
+def _build_startup_handler(
+    app: FastAPI,
+    container: AppContainer,
+) -> Callable[[], Awaitable[None]]:
+    async def startup() -> None:
+        task = _get_cleanup_task(app)
+        if task is not None and not task.done():
+            return
+
+        await asyncio.to_thread(container.sandbox_reconciler.execute)
+
+        stop_event = asyncio.Event()
+        setattr(app.state, APP_CLEANUP_STOP_STATE, stop_event)
+        setattr(
+            app.state,
+            APP_CLEANUP_TASK_STATE,
+            asyncio.create_task(
+                _run_cleanup_loop(container, stop_event),
+                name='sandbox_cleanup',
+            ),
+        )
+
+    return startup
+
+
 def _build_shutdown_handler(
     app: FastAPI,
     container: AppContainer,
-) -> Callable[[], None]:
+) -> Callable[[], Awaitable[None]]:
-    def shutdown() -> None:
+    async def shutdown() -> None:
+        errors: list[Exception] = []
+
+        try:
+            await _stop_cleanup_loop(app)
+        except Exception as exc:
+            errors.append(exc)
+
         try:
             _uninstrument_app(app)
-        finally:
+        except Exception as exc:
+            errors.append(exc)
+
+        try:
             container.shutdown()
+        except Exception as exc:
+            errors.append(exc)
+
+        if errors:
+            raise ExceptionGroup('app shutdown failed', errors)
 
     return shutdown
 
 
+async def _run_cleanup_loop(
+    container: AppContainer,
+    stop_event: asyncio.Event,
+) -> None:
+    interval = container.config.sandbox.cleanup_interval_seconds
+
+    while not stop_event.is_set():
+        try:
+            await asyncio.to_thread(
+                container.usecases.cleanup_expired_sandboxes.execute
+            )
+        except Exception as exc:
+            container.observability.logger.error(
+                'sandbox_cleanup_failed',
+                attrs={
+                    'error': type(exc).__name__,
+                },
+            )
+
+        try:
+            await asyncio.wait_for(stop_event.wait(), timeout=interval)
+        except asyncio.TimeoutError:
+            continue
+
+
+async def _stop_cleanup_loop(app: FastAPI) -> None:
+    stop_event = _get_cleanup_stop_event(app)
+    if stop_event is not None:
+        stop_event.set()
+
+    task = _get_cleanup_task(app)
+    try:
+        if task is not None:
+            await task
+    finally:
+        setattr(app.state, APP_CLEANUP_TASK_STATE, None)
+        setattr(app.state, APP_CLEANUP_STOP_STATE, None)
+
+
+def _get_cleanup_task(app: FastAPI) -> asyncio.Task[None] | None:
+    task = getattr(app.state, APP_CLEANUP_TASK_STATE, None)
+    if isinstance(task, asyncio.Task):
+        return task
+    return None
+
+
+def _get_cleanup_stop_event(app: FastAPI) -> asyncio.Event | None:
+    stop_event = getattr(app.state, APP_CLEANUP_STOP_STATE, None)
+    if isinstance(stop_event, asyncio.Event):
+        return stop_event
+    return None
+
+
 def _uninstrument_app(app: FastAPI) -> None:
     if _is_instrumented(app):
         FastAPIInstrumentor.uninstrument_app(app)

adapter/http/fastapi/dependencies.py

@@ -1,8 +1,7 @@
-from typing import cast
+from fastapi import Depends, Request
 
 from adapter.di.container import AppContainer
-from fastapi import Depends, Request
+from usecase.sandbox import CreateSandbox
-from usecase.user import GetUser
 
 APP_CONTAINER_STATE = 'container'
 APP_CONFIG_STATE = 'config'
@@ -10,10 +9,12 @@ APP_CONFIG_STATE = 'config'
 
 def get_container(request: Request) -> AppContainer:
     container = getattr(request.app.state, APP_CONTAINER_STATE, None)
-    if container is None:
+    if not isinstance(container, AppContainer):
         raise RuntimeError('container unavailable')
-    return cast(AppContainer, container)
+    return container
 
 
-def get_get_user(container: AppContainer = Depends(get_container)) -> GetUser:
+def get_create_sandbox(
-    return container.usecases.get_user
+    container: AppContainer = Depends(get_container),
+) -> CreateSandbox:
+    return container.usecases.create_sandbox

adapter/http/fastapi/middleware.py

@@ -1,8 +1,9 @@
 from time import perf_counter
 
+from fastapi import FastAPI, Request, Response
+
 from adapter.config.model import AppConfig
 from adapter.http.fastapi.dependencies import get_container
-from fastapi import FastAPI, Request, Response
 
 
 def register_middleware(app: FastAPI, config: AppConfig) -> None:

adapter/http/fastapi/routers/v1/router.py

@@ -1,10 +1,19 @@
 from fastapi import APIRouter, Depends, HTTPException, status
 
 from adapter.di.container import AppContainer
-from adapter.http.fastapi.dependencies import get_container, get_get_user
+from adapter.http.fastapi.dependencies import (
-from adapter.http.fastapi.schemas import ErrorResponse, HealthResponse, UserResponse
+    get_container,
-from domain.error import UserNotFoundError
+    get_create_sandbox,
-from usecase.user import GetUser, GetUserQuery
+)
+from adapter.http.fastapi.schemas import (
+    CreateSandboxRequest,
+    ErrorResponse,
+    HealthResponse,
+    SandboxSessionResponse,
+)
+from domain.error import SandboxError, SandboxStartError
+from domain.sandbox import SandboxSession
+from usecase.sandbox import CreateSandbox, CreateSandboxCommand
 
 router = APIRouter()
 
@@ -22,19 +31,40 @@ def health(container: AppContainer = Depends(get_container)) -> HealthResponse:
     )
 
 
-@router.get(
+@router.post(
-    '/users/{user_id}',
+    '/create',
-    response_model=UserResponse,
+    response_model=SandboxSessionResponse,
-    responses={status.HTTP_404_NOT_FOUND: {'model': ErrorResponse}},
+    responses={
+        status.HTTP_503_SERVICE_UNAVAILABLE: {'model': ErrorResponse},
+        status.HTTP_500_INTERNAL_SERVER_ERROR: {'model': ErrorResponse},
+    },
     status_code=status.HTTP_200_OK,
 )
-def get_user(user_id: str, usecase: GetUser = Depends(get_get_user)) -> UserResponse:
+def create_sandbox(
+    request: CreateSandboxRequest,
+    usecase: CreateSandbox = Depends(get_create_sandbox),
+) -> SandboxSessionResponse:
     try:
-        user = usecase.execute(GetUserQuery(user_id=user_id))
+        session = usecase.execute(CreateSandboxCommand(chat_id=request.chat_id))
-    except UserNotFoundError as exc:
+    except SandboxStartError as exc:
         raise HTTPException(
-            status_code=status.HTTP_404_NOT_FOUND,
+            status_code=status.HTTP_503_SERVICE_UNAVAILABLE,
+            detail=str(exc),
+        ) from exc
+    except SandboxError as exc:
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
             detail=str(exc),
         ) from exc
 
-    return UserResponse(id=user.id, email=user.email, name=user.name)
+    return _to_sandbox_session_response(session)
+
+
+def _to_sandbox_session_response(session: SandboxSession) -> SandboxSessionResponse:
+    return SandboxSessionResponse(
+        session_id=session.session_id,
+        chat_id=session.chat_id,
+        container_id=session.container_id,
+        status=session.status.value,
+        expires_at=session.expires_at,
+    )

adapter/http/fastapi/schemas.py

@@ -1,4 +1,7 @@
-from pydantic import BaseModel
+from datetime import datetime
+from uuid import UUID
+
+from pydantic import BaseModel, ConfigDict
 
 
 class HealthResponse(BaseModel):
@@ -7,10 +10,18 @@ class HealthResponse(BaseModel):
     env: str
 
 
-class UserResponse(BaseModel):
+class CreateSandboxRequest(BaseModel):
-    id: str
+    model_config = ConfigDict(extra='forbid')
-    email: str
+
-    name: str
+    chat_id: UUID
+
+
+class SandboxSessionResponse(BaseModel):
+    session_id: UUID
+    chat_id: UUID
+    container_id: str
+    status: str
+    expires_at: datetime
 
 
 class ErrorResponse(BaseModel):

adapter/observability/noop.py

@@ -20,6 +20,14 @@ class NoopMetrics:
     ) -> None:
         return None
 
+    def set(
+        self,
+        name: str,
+        value: int | float,
+        attrs: Attrs | None = None,
+    ) -> None:
+        return None
+
 
 class NoopSpan:
     def set_attribute(self, name: str, value: AttrValue) -> None:

adapter/otel/metrics.py

@@ -5,12 +5,21 @@ from opentelemetry.metrics import Counter, Histogram, Meter
 from usecase.interface import Attrs
 
 
+class _GaugeAdapter:
+    def __init__(self, gauge: object) -> None:
+        self._gauge = gauge
+
+    def set(self, value: int | float, attributes: object = None) -> None:
+        getattr(self._gauge, 'set')(value, attributes=attributes)
+
+
 class OtelMetrics:
     def __init__(self, meter: Meter) -> None:
         self._meter = meter
         self._lock = Lock()
         self._counters: dict[str, Counter] = {}
         self._histograms: dict[str, Histogram] = {}
+        self._gauges: dict[str, _GaugeAdapter] = {}
 
     def increment(
         self,
@@ -34,6 +43,17 @@ class OtelMetrics:
             attributes=None if attrs is None else dict(attrs),
         )
 
+    def set(
+        self,
+        name: str,
+        value: int | float,
+        attrs: Attrs | None = None,
+    ) -> None:
+        self._gauge(name).set(
+            value,
+            attributes=None if attrs is None else dict(attrs),
+        )
+
     def _counter(self, name: str) -> Counter:
         counter = self._counters.get(name)
         if counter is not None:
@@ -57,3 +77,15 @@ class OtelMetrics:
                 histogram = self._meter.create_histogram(name)
                 self._histograms[name] = histogram
         return histogram
+
+    def _gauge(self, name: str) -> _GaugeAdapter:
+        gauge = self._gauges.get(name)
+        if gauge is not None:
+            return gauge
+
+        with self._lock:
+            gauge = self._gauges.get(name)
+            if gauge is None:
+                gauge = _GaugeAdapter(self._meter.create_gauge(name))
+                self._gauges[name] = gauge
+        return gauge

adapter/sandbox/reconciliation.py

@@ -0,0 +1,57 @@
+from dataclasses import dataclass
+from typing import Protocol
+from uuid import UUID
+
+from domain.sandbox import SandboxSession
+from usecase.interface import Logger, Metrics, Tracer
+
+
+class SandboxSessionStateSource(Protocol):
+    def list_active_sessions(self) -> list[SandboxSession]: ...
+
+
+class SandboxSessionRegistry(Protocol):
+    def replace_all(self, sessions: list[SandboxSession]) -> None: ...
+
+    def count_active(self) -> int: ...
+
+
+@dataclass(frozen=True, slots=True)
+class SandboxSessionReconciler:
+    state_source: SandboxSessionStateSource
+    registry: SandboxSessionRegistry
+    logger: Logger
+    metrics: Metrics
+    tracer: Tracer
+
+    def execute(self) -> list[SandboxSession]:
+        with self.tracer.start_span(
+            'adapter.sandbox.reconcile_sessions',
+        ) as span:
+            try:
+                sessions_by_chat_id: dict[UUID, SandboxSession] = {}
+                discovered_sessions = self.state_source.list_active_sessions()
+                span.set_attribute('sandbox.discovered_count', len(discovered_sessions))
+                for session in sorted(
+                    discovered_sessions,
+                    key=lambda item: item.created_at,
+                ):
+                    sessions_by_chat_id[session.chat_id] = session
+
+                sessions = list(sessions_by_chat_id.values())
+                self.registry.replace_all(sessions)
+                active_count = self.registry.count_active()
+                self.metrics.set('sandbox.active.count', active_count)
+                span.set_attribute('sandbox.active_count', active_count)
+                span.set_attribute('sandbox.result', 'reconciled')
+                self.logger.info(
+                    'sandbox_reconciled',
+                    attrs={
+                        'session_count': active_count,
+                    },
+                )
+                return sessions
+            except Exception as exc:
+                span.set_attribute('sandbox.result', 'error')
+                span.record_error(exc)
+                raise

config/app.yaml

@@ -24,5 +24,19 @@ otel:
   traces_endpoint: http://localhost:4318/v1/traces
   metric_export_interval: 1000
 
+docker:
+  base_url: unix:///var/run/docker.sock
+
+sandbox:
+  image: ai-agent:latest
+  ttl_seconds: 300
+  cleanup_interval_seconds: 60
+  chats_root: var/sandbox/chats
+  dependencies_host_path: var/sandbox/dependencies
+  lambda_tools_host_path: var/sandbox/lambda-tools
+  chat_mount_path: /workspace/chat
+  dependencies_mount_path: /opt/dependencies
+  lambda_tools_mount_path: /opt/lambda-tools
+
 security:
   token_header: X-API-Token

config/docker-compose.yml

@@ -0,0 +1,42 @@
+app:
+  name: master-service
+  env: docker-compose
+
+http:
+  host: 0.0.0.0
+  port: 8123
+
+logging:
+  level: INFO
+  output: otel
+  format: json
+
+metrics:
+  enabled: true
+
+tracing:
+  enabled: true
+
+otel:
+  service_name: master-service
+  logs_endpoint: http://otel-collector:4318/v1/logs
+  metrics_endpoint: http://otel-collector:4318/v1/metrics
+  traces_endpoint: http://otel-collector:4318/v1/traces
+  metric_export_interval: 1000
+
+docker:
+  base_url: tcp://docker-engine:2375
+
+sandbox:
+  image: nginx:1.27-alpine
+  ttl_seconds: 300
+  cleanup_interval_seconds: 60
+  chats_root: /var/lib/master-sandbox/chats
+  dependencies_host_path: /var/lib/master-dependencies
+  lambda_tools_host_path: /var/lib/master-lambda-tools
+  chat_mount_path: /workspace/chat
+  dependencies_mount_path: /opt/dependencies
+  lambda_tools_mount_path: /opt/lambda-tools
+
+security:
+  token_header: X-API-Token

docker-compose.yml

@@ -4,22 +4,43 @@ services:
       context: .
       dockerfile: Dockerfile
       target: run
+    user: root
     depends_on:
-      - otel-collector
+      docker-engine:
+        condition: service_healthy
+      otel-collector:
+        condition: service_started
     environment:
-      APP_API_TOKEN: ${APP_API_TOKEN:?APP_API_TOKEN is required}
+      APP_API_TOKEN: local-api-token
-      APP_SIGNING_KEY: ${APP_SIGNING_KEY:?APP_SIGNING_KEY is required}
+      APP_SIGNING_KEY: local-signing-key
-      APP_ENV: docker
-      APP_HTTP_HOST: 0.0.0.0
-      APP_HTTP_PORT: '8123'
-      APP_LOGGING_OUTPUT: otel
-      APP_METRICS_ENABLED: 'true'
-      APP_TRACING_ENABLED: 'true'
-      APP_OTEL_LOGS_ENDPOINT: http://otel-collector:4318/v1/logs
-      APP_OTEL_METRICS_ENDPOINT: http://otel-collector:4318/v1/metrics
-      APP_OTEL_TRACES_ENDPOINT: http://otel-collector:4318/v1/traces
     ports:
       - '127.0.0.1:8123:8123'
+    volumes:
+      - ./config/docker-compose.yml:/app/config/app.yaml:ro
+      - sandbox-data:/var/lib/master-sandbox
+      - sandbox-dependencies:/var/lib/master-dependencies:ro
+      - sandbox-tools:/var/lib/master-lambda-tools:ro
+
+  docker-engine:
+    image: docker:28-dind
+    privileged: true
+    environment:
+      DOCKER_TLS_CERTDIR: ''
+    command:
+      - --host=tcp://0.0.0.0:2375
+    healthcheck:
+      test:
+        - CMD
+        - docker
+        - info
+      interval: 5s
+      timeout: 5s
+      retries: 12
+    volumes:
+      - docker-data:/var/lib/docker
+      - sandbox-data:/var/lib/master-sandbox
+      - sandbox-dependencies:/var/lib/master-dependencies
+      - sandbox-tools:/var/lib/master-lambda-tools
 
   otel-collector:
     image: grafana/otel-lgtm:latest
@@ -29,4 +50,8 @@ services:
       - lgtm-data:/data
 
 volumes:
+  docker-data:
   lgtm-data:
+  sandbox-data:
+  sandbox-dependencies:
+  sandbox-tools:

docs/006-mvp-docker-sandbox-orchestration.md

@@ -0,0 +1,19 @@
+# 006 MVP Docker Sandbox Orchestration
+
+Context
+- The service needs a first MVP for sandbox orchestration behind `/api/v1/create`.
+- The first version must stay small, avoid auth, and preserve clean architecture boundaries.
+
+Decision
+- Use Docker as the outer runtime adapter for sandbox start and stop operations.
+- Keep sandbox entities and errors in `domain/` and orchestration ports in `usecase/`.
+- Use an in-memory session repository for the MVP instead of a central database.
+- Keep one active sandbox per `chat_id` and reuse it until TTL expiry.
+- Set default sandbox TTL to 300 seconds.
+- Mount chat storage as `rw`, dependencies as `ro`, and lambda-tools as `ro`.
+- Run expired sandbox cleanup as an in-process background loop in the HTTP app lifecycle.
+
+Consequences
+- Inner layers stay free from Docker and FastAPI details.
+- The MVP is single-instance oriented and not yet suitable for multi-node coordination.
+- Repository and runtime can be replaced later without changing usecase contracts.

docs/007-startup-sandbox-reconciliation.md

@@ -0,0 +1,17 @@
+# 007 Startup Sandbox Reconciliation
+
+Context
+- Active sandboxes outlive the process because Docker keeps containers running across master-service restarts.
+- The in-memory session repository is rebuilt on each start and otherwise loses running sandbox state.
+
+Decision
+- Reconcile sandbox state during app startup before the cleanup loop starts serving requests.
+- Read running Docker containers through sandbox labels `session_id`, `chat_id`, and `expires_at`.
+- Rebuild the in-memory registry from the reconciled sessions and prefer the newest session per `chat_id`.
+- Let the normal cleanup flow handle reconciled sessions that are already expired.
+- Do not stop healthy sandbox containers during service shutdown; shutdown only stops background control-plane work and closes local resources.
+
+Consequences
+- A restarted master-service reuses existing sandboxes instead of starting duplicates for the same chat.
+- Startup now depends on Docker state access and should fail fast if runtime state cannot be listed.
+- The reconciliation rule stays local to outer layers and does not leak Docker into usecases.

docs/008-sandbox-lifecycle-observability.md

@@ -0,0 +1,18 @@
+# 008 Sandbox lifecycle observability
+
+## Context
+- FR-034 требует метрики по active sandbox, startup latency и cleanup
+- Issue #11 требует трассировку sandbox usecase и Docker adapter steps
+- Inner layers должны знать только observability ports
+
+## Decision
+- Usecase sandbox lifecycle использует только `Logger`, `Metrics`, `Tracer`
+- `Metrics` получает `set(...)` для current-state signals
+- `sandbox.active.count` считается из session registry через `count_active()`
+- M19 добавляет только contracts и adapter support для будущих lifecycle signals
+- M20 и M21 отдельно добавят spans и runtime metrics в usecase и Docker adapter
+
+## Consequences
+- OTel gauge остается в outer adapter, не протекает во внутренние слои
+- Active sandbox count синхронизируется после create, cleanup и reconciliation
+- Tests могут проверять observability через fake ports без реального OTel backend

docs/009-storage-foundation.md

@@ -0,0 +1,17 @@
+# 009 Storage foundation
+
+## Context
+- v1 storage slice needs workspace, chat and file flows before durable DB
+- trusted caller passes `user_id`, and one workspace belongs to one user
+- chat content must live outside sandbox lifecycle and survive sandbox restart
+
+## Decision
+- metadata repositories are in-memory for the first storage slice
+- `Workspace`, `Chat` and `ChatFile` are first-class domain entities
+- filesystem access stays behind storage ports in outer layers
+- sandbox later integrates through chat metadata and storage ports, not raw path math in usecases
+
+## Consequences
+- metadata is lost on restart in this phase
+- storage usecases and HTTP API can be built before durable persistence
+- later durable metadata can replace in-memory adapters behind the same ports

docs/010-chat-history-policy.md

@@ -0,0 +1,17 @@
+# 010 Chat history policy
+
+## Context
+- v1 keeps chat history in filesystem, not in central DB
+- chat metadata must not depend on parsing history content
+- each chat already maps to an isolated working directory
+
+## Decision
+- each chat owns one `history.md` inside its chat directory
+- `history.md` is created with chat layout initialization
+- chat metadata stores identity and lifecycle fields separately from history content
+- history read and write stay behind storage ports in outer layers
+
+## Consequences
+- history survives sandbox restart with chat storage
+- metadata and content evolve independently
+- later migration to another history backend can keep the same chat identity model

docs/master.md

@@ -0,0 +1,125 @@
+# Требования к master-service
+
+Источник: `meetings/meeting_1_document.md` + уточнение от тимлида
+
+## Назначение
+
+Master-service — это control plane платформы AI-агентов. Он управляет sandbox-контейнерами с AI-Agent, рабочими пространствами пользователей, чатами, пользовательскими файлами, проверкой доступа и выдачей временного p2p-доступа к контейнеру.
+
+После старта sandbox и успешной авторизации Master должен оставаться плоскостью управления, а не постоянным data-plane прокси: рабочий трафик к агенту должен идти напрямую по выданному p2p-каналу.
+
+## Границы ответственности
+
+### В scope
+
+- оркестрация sandbox
+- управление workspace и chat volume
+- хранение чатов? и файлов пользователя
+- хранение и выдача артефактов
+- проверка доступа
+- выдача и отзыв p2p-доступа
+- учет сессий, TTL и cleanup-процессов
+- хранение служебных метаданных
+
+### Вне scope
+
+- Telegram/Discord/Slack интеграции и их transport-логика
+- бизнес-логика AI-Agent внутри sandbox
+- реализация LLM-level защиты от prompt injection
+- реализация конкретных lambda-tools и browser tools
+
+## Основные сущности
+
+- **User** — владелец workspace
+- **Workspace** — пользовательское хранилище с мягкой квотой 10 GB
+- **Chat** — изолированная рабочая область внутри workspace
+- **SandboxSession** — активный runtime AI-Agent для конкретного чата
+- **AccessLease** — временное право на p2p-доступ к sandbox
+- **ChatFile** — файл, прикрепленный к чату
+- **Artifact** — результат, созданный агентом и подготовленный к выдаче наружу
+- **DependencyBundle** — предзагруженный набор зависимостей tool/runtime
+- **LambdaToolBundle** — платформенный read-only набор тулов
+
+## Функциональные требования
+
+### 1. Управление workspace и chat storage
+
+- **FR-001** Master должен создавать workspace пользователя при первом обращении
+- **FR-002** Master должен вести мягкую квоту workspace в 10 GB на пользователя без физического резервирования места
+- **FR-003** Master должен создавать, получать, перечислять и удалять чаты пользователя
+- **FR-004** Каждый chat должен иметь собственную изолированную директорию с файлами и историей
+- **FR-005** История сообщений чата в первой версии должна храниться рядом с чатом в файле `history.md`
+- **FR-006** Master должен сохранять пользовательские вложения в директорию соответствующего чата до запуска или во время работы sandbox
+- **FR-007** Master должен предоставлять операции списка, чтения метаданных, удаления и очистки файлов чата
+
+### 2. Оркестрация sandbox
+
+- **FR-008** Master должен поднимать sandbox-контейнер AI-Agent по первому сообщению или при отсутствии активной sandbox-сессии
+- **FR-009** Master должен переиспользовать активную sandbox-сессию, пока не истек TTL бездействия
+- **FR-010** Master должен завершать sandbox после периода неактивности; стартовое значение TTL — около 1 часа, но параметр должен быть конфигурируемым
+- **FR-011** После завершения sandbox данные чата и workspace должны сохраняться и использоваться при следующем старте
+- **FR-012** Master должен уметь пересоздать sandbox на другом узле при сохранности данных и доступности нужных volume
+
+### 3. Подключение volume и runtime-ресурсов
+
+- **FR-013** При старте sandbox Master должен подключать volume текущего чата
+- **FR-014** При старте sandbox Master должен подключать volume с dependency cache
+- **FR-015** При старте sandbox Master должен подключать volume с lambda-tools в режиме read-only
+- **FR-016** Если используются user-tools, Master должен подключать их так, чтобы агент мог их только исполнять, но не менять
+- **FR-017** Master должен проверять наличие нужных dependency bundle на текущем узле и при отсутствии инициировать загрузку из внутреннего хранилища
+
+### 4. Доступ и p2p-соединение
+
+- **FR-018** Master должен аутентифицировать внутренние запросы от интеграционного слоя и других доверенных сервисов
+- **FR-019** Master должен авторизовывать доступ по связке user/workspace/chat/sandbox
+- **FR-020** После успешной проверки доступа Master должен выдавать временные параметры p2p-подключения к sandbox
+- **FR-021** Выданный p2p-доступ должен быть ограничен по времени и привязан к конкретной sandbox-сессии
+- **FR-022** Master должен уметь отзывать p2p-доступ при остановке sandbox, смене прав или завершении сессии
+- **FR-023** Master не должен проксировать постоянный трафик к агенту после выдачи p2p-доступа, кроме управляющих операций
+
+### 5. Хранение чатов, файлов и артефактов
+
+- **FR-024** Master должен хранить центральные метаданные о пользователях, чатах, sandbox-сессиях, lease, файлах и артефактах
+- **FR-025** Master должен отделять метаданные чата в центральной БД от фактической истории сообщений в `history.db`
+- **FR-026** Master должен поддерживать загрузку артефактов в S3-совместимое объектное хранилище
+- **FR-027** Master должен хранить связь между артефактом, пользователем, чатом и способом внешней доставки
+- **FR-028** После подтвержденной отправки артефакт должен быть помечен на удаление или удален по policy
+
+### 6. Lifecycle и cleanup
+
+- **FR-029** Master должен отслеживать `last_read` и/или `last_update` для файлов чата
+- **FR-030** Файлы чата должны автоматически удаляться после 3 дней неактивности
+- **FR-031** Master должен поддерживать удаление всего workspace после длительной неактивности аккаунта; срок должен быть конфигурируемым
+- **FR-032** Cleanup должен быть безопасным: нельзя удалять данные активного sandbox или чата с действующим lease
+
+### 7. Наблюдаемость и аудит
+
+- **FR-033** Master должен логировать создание и остановку sandbox, подключение volume, выдачу доступа и cleanup-операции
+- **FR-034** Master должен собирать метрики по активным sandbox, времени запуска, ошибкам старта, cleanup и storage usage
+- **FR-035** Master должен предоставлять аудитный след по операциям доступа к chat/file/artifact ресурсам
+
+## Нефункциональные требования
+
+- **NFR-001** Master должен быть control-plane сервисом и не содержать бизнес-логику AI-Agent
+- **NFR-002** Все данные разных пользователей должны быть изолированы на уровне путей, mount policy и access policy
+- **NFR-003** Повторный запрос на создание chat или запуск sandbox не должен приводить к дублированию ресурсов при одинаковом идемпотентном ключе
+- **NFR-004** Политики TTL, cleanup, квоты и lease должны задаваться конфигом
+- **NFR-005** Решение должно позволять горизонтально масштабировать Master и пересоздавать sandbox независимо от конкретного инстанса сервиса
+- **NFR-006** Отказ sandbox не должен приводить к потере chat history и пользовательских файлов
+- **NFR-007** Все security-чувствительные операции должны быть трассируемы и наблюдаемы
+
+## Явные ограничения первой версии
+
+- один master-service отвечает за orchestration и storage metadata
+- история сообщений хранится в `history.md` внутри chat directory
+- workspace quota считается по фактическому размеру файлов, без hard reservation
+- heavy tools и browser automation не выносятся в отдельный сервис, пока это не потребуется
+- интеграции мессенджеров считаются внешними клиентами master-service
+
+## Открытые вопросы
+
+- какой именно transport используется для p2p-доступа: raw TCP, WebSocket, gRPC или другой вариант = WebSocket
+- какой срок inactivity cleanup для всего аккаунта считается продуктовым дефолтом
+- нужен ли отдельный API для скачивания chat files наружу или достаточно metadata + object/file gateway
+- подтверждается ли доставка артефакта внешней интеграцией явно или cleanup идет только по TTL
+- одна sandbox-сессия должна соответствовать одному chat или допустим multiplex нескольких чатов на один runtime

domain/chat.py

@@ -0,0 +1,35 @@
+from dataclasses import dataclass
+from datetime import datetime
+from uuid import UUID
+
+HISTORY_FILE_NAME = 'history.md'
+
+
+@dataclass(frozen=True, slots=True)
+class ChatAttachmentName:
+    value: str
+
+    def __post_init__(self) -> None:
+        if not self.value or self.value in {'.', '..'}:
+            raise ValueError('invalid attachment name')
+        if '/' in self.value or '\\' in self.value:
+            raise ValueError('invalid attachment name')
+        if self.value == HISTORY_FILE_NAME:
+            raise ValueError('reserved attachment name')
+
+
+@dataclass(frozen=True, slots=True)
+class Chat:
+    chat_id: UUID
+    workspace_id: UUID
+    created_at: datetime
+
+
+@dataclass(frozen=True, slots=True)
+class ChatFile:
+    file_id: UUID
+    chat_id: UUID
+    name: ChatAttachmentName
+    content_type: str | None
+    size_bytes: int
+    created_at: datetime

domain/error.py

@@ -1,3 +1,6 @@
+from uuid import UUID
+
+
 class DomainError(Exception):
     pass
 
@@ -16,3 +19,61 @@ class UserConflictError(UserError):
     def __init__(self, email: str) -> None:
         super().__init__('user_conflict')
         self.email = email
+
+
+class WorkspaceError(DomainError):
+    pass
+
+
+class WorkspaceNotFoundError(WorkspaceError):
+    def __init__(self, workspace_id: UUID) -> None:
+        super().__init__('workspace_not_found')
+        self.workspace_id = workspace_id
+
+
+class WorkspaceQuotaExceededError(WorkspaceError):
+    def __init__(self, workspace_id: UUID) -> None:
+        super().__init__('workspace_quota_exceeded')
+        self.workspace_id = workspace_id
+
+
+class ChatError(DomainError):
+    pass
+
+
+class ChatNotFoundError(ChatError):
+    def __init__(self, chat_id: UUID) -> None:
+        super().__init__('chat_not_found')
+        self.chat_id = chat_id
+
+
+class ChatHasActiveSandboxError(ChatError):
+    def __init__(self, chat_id: UUID) -> None:
+        super().__init__('chat_has_active_sandbox')
+        self.chat_id = chat_id
+
+
+class ChatFileError(DomainError):
+    pass
+
+
+class ChatFileNotFoundError(ChatFileError):
+    def __init__(self, file_id: UUID) -> None:
+        super().__init__('chat_file_not_found')
+        self.file_id = file_id
+
+
+class SandboxError(DomainError):
+    pass
+
+
+class SandboxStartError(SandboxError):
+    def __init__(self, chat_id: str) -> None:
+        super().__init__('sandbox_start_failed')
+        self.chat_id = chat_id
+
+
+class SandboxAlreadyRunningError(SandboxError):
+    def __init__(self, chat_id: str) -> None:
+        super().__init__('sandbox_already_running')
+        self.chat_id = chat_id

domain/sandbox.py

@@ -0,0 +1,22 @@
+from dataclasses import dataclass
+from datetime import datetime
+from enum import Enum
+from uuid import UUID
+
+
+class SandboxStatus(str, Enum):
+    STARTING = 'starting'
+    RUNNING = 'running'
+    STOPPING = 'stopping'
+    STOPPED = 'stopped'
+    FAILED = 'failed'
+
+
+@dataclass(frozen=True, slots=True)
+class SandboxSession:
+    session_id: UUID
+    chat_id: UUID
+    container_id: str
+    status: SandboxStatus
+    created_at: datetime
+    expires_at: datetime

domain/workspace.py

@@ -0,0 +1,17 @@
+from dataclasses import dataclass
+from datetime import datetime
+from uuid import UUID
+
+
+@dataclass(frozen=True, slots=True)
+class Workspace:
+    workspace_id: UUID
+    user_id: UUID
+    created_at: datetime
+
+
+@dataclass(frozen=True, slots=True)
+class WorkspaceUsage:
+    workspace_id: UUID
+    used_bytes: int
+    quota_bytes: int

meetings/meeting_1_document.md

@@ -0,0 +1,405 @@
+# Meeting 1 — Platform Architecture Design
+**Date:** 2026-03-21 (estimated from transcript context)
+**Duration:** ~1:30:18 (00:00 - 1:30:18)
+**Participants:** Azamat N (lead architect, screen-sharing), Egor (Егор), Yaroslav (Ярослав), David Shvarts (Давид Шварц), Andrey Bakhtiozin (Бахтиозин Андрей)
+**Platform:** DION video conferencing
+**Collaborative tool:** Excalidraw (shared whiteboard)
+
+## Summary
+
+The Lambda Lab 3.0 team held their first architecture planning meeting to design the AI-agent platform. Azamat presented a draft architecture he had developed with Egor over the preceding two days, walking through the system on a shared Excalidraw whiteboard. The platform consists of two main applications: a **Master** service (orchestrator managing sessions, containers, and context) and an **AI-Agent** service (running inside isolated Docker containers with a per-user workspace). The team discussed the workspace structure (10GB per user with chat-scoped files and history), three categories of tools (built-in, lambda/platform, and user-generated), dependency management via mount points, artifact storage via S3, and message history storage options. Key side topics included security concerns around prompt injection and file leakage, the agent-browser tool for web interaction, infrastructure decisions (Forgejo for Git hosting, Matrix for team communication), and upcoming deadlines — specifically presenting this architecture to Kalinin the next day.
+
+The architecture is explicitly described as a first draft created in two days, meant to establish a direction rather than be final. The team assigned responsibilities: Azamat handles Forgejo repository setup, wiki documentation, and architecture presentation preparation. The meeting also touched on developer tooling — using Claude/OpenCode with agent profiles, CLAUDE.md-style configuration files, and CI/CD pipelines for documentation.
+
+## Discussion
+
+### Setup and Technical Issues
+**[00:00 — 02:14]**
+
+The meeting opened with participants joining the DION video call and resolving recording/connection issues. Azamat briefly disconnected and reconnected. Five participants were present: Azamat N, Egor, Yaroslav, David Shvarts, and Andrey Bakhtiozin.
+
+**Visual context:** DION video conferencing interface showing participant tiles with initials and names.
+
+---
+
+### Platform Architecture Overview
+**[02:17 — 04:50]**
+
+Azamat shared his Excalidraw screen and began presenting the architecture he had drafted with Egor. He explained the overall platform goal: creating AI agents accessible through various integrations (Telegram, Discord, Slack), where the agent lives in an isolated environment.
+
+The architecture has two main applications:
+
+1. **Master** — orchestrator that manages agent sessions. When a user writes in Telegram, the message hits the Master, which creates a session and asks Docker to spin up a container with the AI-Agent.
+
+2. **AI-Agent** — the second application, living inside an isolated Docker container. Once the container is running, communication happens directly between the messaging integration and the agent container (peer-to-peer), bypassing the Master.
+
+> **Azamat** @ 02:37: "Вот наша задача сделать разными схемами платформу, которые подключаются к различной интеграции по типу Telegram, Discord и всего такого."
+
+> **Azamat** @ 02:51: "Наша задача именно сделать агента, который живет в изолированной среде."
+
+**Visual context:** Excalidraw diagram showing Stream (Telegram, Discord, Slack) -> Docker (multiple AI-Agent instances) -> Master (S3, DB). Arrows indicate data flow between components.
+
+---
+
+### Workspace Structure and Projects
+**[05:15 — 07:28]**
+
+Azamat described the workspace allocated per user — 10GB of storage. Within this workspace, users can create projects (later renamed to "chats"), each containing files the agent interacts with (PDFs, images, documents, videos). This is analogous to ChatGPT's project feature.
+
+> **Azamat** @ 05:15: "Мы ему выделяем 10 гигабайт памяти. Просто. И он в эти 10 гигабайт может делать, что хочет."
+
+> **Azamat** @ 07:14: "У юзера, вот как в chat.gpt можно создавать проекты, здесь будет то же самое."
+
+**Visual context:** Excalidraw diagram showing "Workspace 10GB" containing P1 (file1.pdf, file2.png), P2 (file3.docx), P3 (video.mp4).
+
+---
+
+### Tool Categories
+**[07:30 — 12:35]**
+
+The team established three categories of tools:
+
+**1. Built-in tools** (embedded in agent code):
+- `web-search` — web search capability
+- `fetch-url` — fetch a web page (e.g., `http://simple-file.ru/index.html`)
+- `bash` — command-line interaction (cd, ls, git, python, etc.)
+
+**2. Lambda tools (platform-wide, read-only):**
+- Written by other teams within the organization
+- Example: `markitdown` — converts various formats (PDF, DOCX, HTML) to Markdown
+- Must support `--help`, `--json`/`--agent` flags for agent-friendly interaction
+- Must support `man-agent <tool>` for machine-readable documentation
+- Read-only, mounted into every agent container
+
+**3. User tools (per-user, execute-only):**
+- Agent can write scripts to solve specific user tasks
+- Example: a simple task-tracker with `add` and `close` commands, backed by sqlite.db
+- Execute-only permissions (`+x only`) for security
+
+> **Azamat** @ 10:38: "Когда мы вызываем условно какой-нибудь такой тул, у нас должно быть типа help и желательно еще минус-минус JSON, чтобы вот это агент понимал."
+
+**Visual context:** Excalidraw showing AI-Agent sandbox with built-in tools listed, "Сторонние тулы RO" (third-party tools read-only) section, and "User-tool +x only" section. Workspace with P1/P2/P3 projects below.
+
+---
+
+### Security Discussion: User Tools
+**[15:58 — 16:30]**
+
+Egor raised a security concern: if user tools are writable by the agent, a malicious prompt injection could cause the agent to modify user tools, inject code like `print(token)`, and exfiltrate credentials stored in environment variables.
+
+> **Egor** @ 15:58: "Потому что это поможет подцепить где-нибудь вирусный промпт, полезть в папку UserTools, ввести туда какой-нибудь принт токен и слить токены оттуда."
+
+The team agreed that user tools should have execute-only permissions and that this area needs further discussion.
+
+---
+
+### Dependency Management and Mount Points
+**[17:37 — 23:47]**
+
+The team discussed how dependencies for tools (Python libraries, etc.) would be managed. Since downloading gigabytes of dependencies on every container launch is impractical, they decided to:
+
+- **Pre-mount dependencies** on the host machine
+- Check if dependencies exist on the current node at container launch
+- If not present, download from their own S3/cloud storage (in the same data center) rather than external sources
+- Store as mounted volumes, not inside containers
+
+Mount points defined:
+- `Mount cur-chat` — current chat/project files
+- `Mount dependencies` — pre-cached tool dependencies
+- `Mount lambda-tools` — shared platform tools (RO)
+
+> **Azamat** @ 18:15: "Нам надо где-то сохранять, потому что на каждый запуск еще тянуть гигабайт со скоростью 100 килобит в секунду не очень хочется."
+
+**Visual context:** Updated Excalidraw diagram showing mount points connected to the AI-Agent workspace, with arrows from host machine storage to container mounts.
+
+---
+
+### S3 Storage for Artifacts
+**[16:46 — 17:36]**
+
+S3 is used for artifact storage — files the agent generates (reports, presentations, images) that need to be sent back to users via Telegram or other messengers. The agent uploads artifacts to S3, and they are delivered to the user through the messaging integration.
+
+> **Azamat** @ 21:32: "Просто файл-помойка. Типа, если ты хочешь знать, как с ним работать, ты файлик просто по http туда загружаешь."
+
+> **David** @ 22:08: "Считай просто это как Google диск, только наш."
+
+---
+
+### Projects vs. Chats Naming
+**[36:07 — 37:50]**
+
+Confusion arose around the "project" naming because in ChatGPT, a project contains multiple chats. The team decided to rename "projects" to **"chats"** (C1, C2, C3) to avoid confusion. Each chat has its own files and history.db. The question of supporting multiple chats per project was deferred.
+
+> **Azamat** @ 36:29: "Знаешь, короче, как чаты просто в GPT-шке, только мы их называли проектами."
+
+> **Egor** @ 37:30: "Confusing naming тут."
+
+---
+
+### Message History Storage
+**[30:55 — 35:32]**
+
+The team debated where to store message history. Options discussed:
+
+1. **Central database** — rejected because querying by project_id across scattered chunks is inefficient
+2. **Per-chat file** (history.db) inside the workspace — accepted as the initial approach
+
+Egor argued that per-chat SQLite files are more efficient since you always need the full history for a chat, not individual messages. The database fields discussed: `project_id`, `timestamp`, `msg`.
+
+> **Egor** @ 34:45: "Каждый запрос в BD, ему нужно пройтись по индексу, найти все записи вот с этим project id, а они раскиданы все по разным чанкам."
+
+> **Azamat** @ 36:04: "Я вспомнил про файлы типа memory.md, soul.md. Кто-то же так делает."
+
+**Visual context:** Database schema shown with fields: session_id/project_id, user_id, timestamp, msg.
+
+---
+
+### File Lifecycle and Cleanup
+**[37:50 — 42:12]**
+
+The team discussed file retention policy within chats:
+
+- **Decision:** Files inside chats live for **3 days** by default
+- Track `last_read`/`last_update` timestamps
+- After 3 days without access, files are auto-deleted
+- If a user needs a deleted file, the bot asks them to re-send it
+- For inactive accounts: delete entire workspace after N months (like Telegram's 6-month inactivity deletion)
+- Artifacts generated by the agent are deleted after being sent to the user via messenger
+
+> **Azamat** @ 40:20: "Условно сделаем хранение три дня файлов, то есть если три дня никто их не трогает."
+
+> **Azamat** @ 48:33: "Давай просто как у телеграма. Если профиль не активен 3 месяца, он удаляет аккаунт."
+
+---
+
+### Responsibilities and Team Boundaries
+**[43:00 — 43:43]**
+
+The team explicitly marked areas of responsibility on the Excalidraw diagram. The green-highlighted sections (Telegram/Discord/Slack integrations and Stream) are **not their responsibility** — those belong to "Путиловский-team" (Putilovsky's team). Lambda Lab 3.0 owns the Master, AI-Agent, workspace, and tools infrastructure.
+
+**Visual context:** Diagram annotated with "Путиловский-team" labels on the Stream/integration components.
+
+---
+
+### Machine-Level Architecture
+**[44:29 — 47:01]**
+
+Egor suggested visualizing the machine-level architecture: a Linux base, shared environment layer with lambda-tools, then multiple Agent-AI containers above, each with their own environment and code layer. This shows how lambda tools are shared across all agents on a machine, while each agent has its own isolated workspace.
+
+**Visual context:** New Excalidraw section labeled "Машина" (Machine) showing: Linux -> Env -> lambda-tools -> multiple Agent-AI instances.
+
+---
+
+### Storage Overcommit Discussion
+**[49:16 — 51:32]**
+
+Azamat suggested that the 10GB per user wouldn't be physically reserved — similar to VPS overcommit, they'd simply check that each user's directory doesn't exceed 10GB. If a machine runs out of space, Kubernetes can migrate workloads to another node. The team agreed this is acceptable for the initial version and deferred detailed capacity planning.
+
+> **Azamat** @ 49:26: "Мы просто их будем создавать на диске и смотреть, что папка пользователей не больше 10 гигавайт."
+
+---
+
+### Container Lifecycle and TTL
+**[55:44 — 56:10]**
+
+When a user sends their first message, a container spins up. The container has a **TTL** — if there are no messages for about an hour, the container is killed. When the user returns, a new container is launched with the same mounted workspace data, so everything persists.
+
+> **Azamat** @ 55:55: "У него есть какой-то TTL. Ну, типа, если там сообщений нет час, ну, контейнер сдохнет."
+
+---
+
+### Testing Strategy
+**[52:40 — 53:25]**
+
+The question of testing without integrations was raised. Since the Telegram/Discord/Slack integration is another team's responsibility, testing the AI-Agent would require generating mock events. Azamat suggested using AI tools to quickly generate test client instruments.
+
+---
+
+### Repository Structure
+**[53:28 — 54:17]**
+
+The team decided on **two repositories**: one for Master, one for AI-Agent. Both will be hosted on Forgejo (their self-hosted Git platform). The architecture document and repository structure are due by Friday (per Kalinin's report).
+
+> **Azamat** @ 53:54: "Написано в отчете: команда к пятнице — предоставить структуру репозитория и архитектурный документ."
+
+---
+
+### Infrastructure: Forgejo, Matrix, Communication
+**[57:06 — 1:03:34]**
+
+- **Forgejo** — self-hosted Git service at lambda.coredump.ru, used instead of GitHub for code privacy and Docker registry access
+- **Matrix** — team communication platform, registration at element.lambda.coredump.ru (E2E encryption disabled on server). To be used as backup if Telegram gets blocked.
+- The team discussed registering on both platforms
+
+> **Azamat** @ 1:01:01: "Наш личный гитаб."
+
+**Visual context:** Forgejo landing page showing "MAI Lambda Lab 3.0" with self-hosted Git repository features. Telegram chat showing Matrix registration instructions.
+
+---
+
+### Agent-Browser Tool
+**[1:12:11 — 1:15:32]**
+
+The team discussed the agent-browser tool — a CLI tool for browser automation. They reviewed the GitHub repository for an open-source agent-browser (23.6k stars, Rust-based, ~1MB binary). Key concern: Chromium dependency (~800MB + RAM usage).
+
+**Decision:** Start with running the browser tool locally inside the container. If it proves too heavy, move to a separate API-based service.
+
+> **Azamat** @ 1:14:51: "Давай по умолчанию, ну что, вот сейчас примем решение, что оно по API ходит."
+
+**Visual context:** GitHub repository page for agent-browser showing quick-start commands, installation instructions, and browser automation commands.
+
+---
+
+### Scripting and Tool Implementation
+**[1:16:00 — 1:19:07]**
+
+Discussion about how agents should implement tools at runtime. The team considered Lua scripting as a lightweight alternative to Python for agent-generated scripts, avoiding dependency issues. The idea is that the agent writes small scripts to process data without needing to call the LLM again for each step.
+
+> **Azamat** @ 1:18:43: "Проще питон, потому что у вас питон уже будет поднятый."
+
+---
+
+### Security: Prompt Injection and Data Leakage
+**[1:19:27 — 1:23:58]**
+
+Extended discussion about security risks:
+
+- **Prompt injection** — malicious prompts embedded in web pages or files could instruct the agent to exfiltrate user data
+- **File leakage** — sensitive files (e.g., NDA documents, movie scripts) uploaded by users could be leaked by a compromised agent
+- **Defense options discussed:**
+  - Agent creates an execution plan before acting, which can be reviewed
+  - User tools should have execute-only permissions
+  - LLM-level defense is "useless" per Egor
+- **Decision:** Security is not Lambda Lab 3.0's sole responsibility — there's another team doing "LLM research" that should add **security metrics** and test defense mechanisms
+
+> **Egor** @ 1:23:28: "И защищаться на уровне ЛЛМ это бесполезно."
+
+> **Azamat** @ 1:23:50: "Придумать метрику безопасности."
+
+---
+
+### Tomorrow's Presentation and Next Steps
+**[1:24:33 — 1:30:18]**
+
+The team confirmed that the Excalidraw architecture diagram is the main presentation material for the next day's meeting with Kalinin. Azamat will:
+- Set up the Forgejo repository
+- Create wiki documentation
+- Set up the architecture document
+- Share Forgejo registration links
+
+The meeting ended with Azamat demonstrating his OpenCode setup — showing agent profiles (code-refactorer, bug-hunter-fixer, worker, documentation-generator, feature-developer, code-explorer, test-engineer, code-reviewer) and system-design.md configuration files that define how AI coding agents interact with the project.
+
+> **Azamat** @ 1:25:09: "Касательно завтрашней презентации, мы же будем как-то вот показывать эту схему, не на пальцах же объяснять?"
+
+**Visual context:** Terminal showing OpenCode CLI with Claude Opus 4.6 model, agent profiles, system-design.md configuration file showing subagent delegation rules.
+
+---
+
+## Decisions Made
+
+| # | Decision | Context | Timestamp |
+|---|----------|---------|-----------|
+| 1 | Platform consists of two applications: Master (orchestrator) and AI-Agent (sandboxed) | Core architecture decision | 03:01 |
+| 2 | 10GB workspace per user with overcommit (not physically reserved) | Storage allocation strategy | 05:15, 49:26 |
+| 3 | Three tool categories: built-in (in agent code), lambda (platform-wide RO), user (per-user +x only) | Tool taxonomy | 07:30 |
+| 4 | Tools must support `--help`, `--json`/`--agent` flags and `man-agent` command | Tool interface contract | 10:38 |
+| 5 | User tools have execute-only permissions for security | Security measure | 16:19 |
+| 6 | Dependencies pre-mounted from host, cached in S3 | Performance optimization | 18:15 |
+| 7 | Rename "projects" to "chats" (C1, C2, C3) | Naming clarity | 37:40 |
+| 8 | Message history stored as per-chat history.db files inside workspace | Storage simplicity | 33:14 |
+| 9 | Files inside chats auto-delete after 3 days of inactivity | File lifecycle policy | 40:20 |
+| 10 | Inactive accounts deleted after N months (Telegram-style) | Account lifecycle | 48:33 |
+| 11 | Container TTL ~1 hour without messages | Resource management | 55:55 |
+| 12 | Two repositories: one for Master, one for AI-Agent | Code organization | 54:00 |
+| 13 | Use Forgejo for Git hosting | Infrastructure | 57:30 |
+| 14 | Use Matrix as backup communication (Element client) | Team communication | 1:00:37 |
+| 15 | Agent-browser starts as local tool, move to API if too heavy | Browser tool strategy | 1:14:51 |
+| 16 | Security is a shared responsibility — LLM metrics team should add security benchmarks | Security ownership | 1:23:48 |
+
+## Action Items
+
+| # | Owner | Task | Deadline | Timestamp |
+|---|-------|------|----------|-----------|
+| 1 | Azamat | Create Forgejo repository structure for Master and AI-Agent | Today (meeting day) | 1:00:12 |
+| 2 | Azamat | Set up wiki and upload architecture documentation | Today | 1:00:19 |
+| 3 | Azamat | Share Forgejo registration link with the team | Today | 1:01:24 |
+| 4 | Azamat | Create Matrix chat for the team | Today | 1:03:09 |
+| 5 | Azamat | Prepare architecture presentation for Kalinin | Tomorrow | 1:24:33 |
+| 6 | Team | Register on Forgejo (lambda.coredump.ru) | Today | 1:01:06 |
+| 7 | Team | Register on Matrix (element.lambda.coredump.ru) | Today | 1:00:37 |
+| 8 | Team | Define tool interface contract (--help, --json, man-agent) | Not specified | 11:15 |
+| 9 | Team | Set up dev-environment configuration (CLAUDE.md-style) | Not specified | 1:26:22 |
+| 10 | Other team | Define security metrics for LLM safety | Not specified | 1:23:50 |
+
+## Key Quotes
+
+> "Наша задача именно сделать агента, который живет в изолированной среде, при этом он может пользоваться какими-то тулами." — Azamat, 02:51
+
+> "Мы ему выделяем 10 гигабайт памяти. Просто. И он в эти 10 гигабайт может делать, что хочет." — Azamat, 05:15
+
+> "Потому что это поможет подцепить где-нибудь вирусный промпт, полезть в папку UserTools, ввести туда какой-нибудь принт токен и слить токены оттуда." — Egor, 15:58
+
+> "Каждый запрос в BD, ему нужно пройтись по индексу, найти все записи вот с этим project id, а они раскиданы все по разным чанкам." — Egor, 34:45
+
+> "Считай просто это как Google диск, только наш." — David, 22:08
+
+> "И защищаться на уровне ЛЛМ это бесполезно." — Egor, 1:23:28
+
+> "Мы же, понимаешь, у нас же, это типа за два дня накинутая просто первая архитектура. Чтобы мы начали думать в этом направлении хотя бы." — Azamat, 50:14
+
+## Visual Artifacts
+
+### Architecture Diagram (Excalidraw — primary artifact)
+
+The main diagram evolved throughout the meeting on Excalidraw and shows the following system:
+
+**Top-level architecture:**
+- **Stream layer** (Путиловский-team responsibility): Telegram, Discord, Slack integrations
+- **Docker layer**: Multiple AI-Agent containers
+- **Master**: Orchestrator connected to DB and S3
+- **S3**: Artifact/file storage
+- **DB**: Session management, metadata
+
+**AI-Agent Sandbox (per-user):**
+- **Workspace 10GB** containing:
+  - Chats: C1 (file1.pdf, file2.png, history.db), C2 (file3.docx, history.db), C3 (video.mp4, history.db)
+  - User-tool (+x only)
+- **Mount points:**
+  - `Mount cur-chat` — current chat data
+  - `Mount dependencies` — cached tool dependencies
+  - `Mount lambda-tools` — shared platform tools
+- **Built-in tools:** web-search, fetch-url, bash
+- **Lambda tools (RO):** markitdown (with --help, --json/--agent, man-agent interface)
+
+**Machine-level view:**
+- Linux base
+- Shared Env layer
+- Lambda-tools (shared)
+- Multiple Agent-AI instances, each with own workspace
+
+### ChatGPT Projects Reference
+**[36:30]** Azamat showed ChatGPT's project interface to illustrate the concept of projects containing chats with attached files.
+
+### Forgejo Instance
+**[57:30]** The team's Forgejo instance at "MAI Lambda Lab 3.0" — self-hosted Git repository service.
+
+### Agent-Browser GitHub Repository
+**[1:13:30]** agent-browser repository (23.6k stars, 58 releases, v0.21.2, Rust-based CLI tool for browser automation).
+
+### OpenCode Agent Setup
+**[1:28:00 — 1:29:30]** Azamat's terminal showing OpenCode CLI with Claude Opus 4.6, multiple agent profiles (@code-refactorer, @bug-hunter-fixer, @worker, @documentation-generator, @feature-developer, @code-explorer, @test-engineer, @code-reviewer), and system-design.md configuration for agent behavior.
+
+## Open Questions
+
+1. **How exactly should the tool interface contract work?** The `--help`, `--json`/`--agent` flags and `man-agent` command were proposed but not finalized.
+2. **Should user tools be writable by the agent or only loaded from external sources?** Security implications were raised but not fully resolved.
+3. **What happens when a heavy tool (like agent-browser) requires more resources than the container allows?** The "start local, move to API" approach was agreed on but details remain.
+4. **How to handle the scenario where an agent needs a library that isn't pre-installed?** The scripting discussion suggested Python is available, but library management wasn't resolved.
+5. **Security architecture:** Who is responsible for defending against prompt injection attacks? The LLM metrics team was suggested, but no formal agreement was reached.
+6. **Memory and context management:** SOUL.md/MEMORY.md-style files were mentioned but not designed.
+7. **What should the Kubernetes setup look like?** Mentioned as needed "in the second half of spring" but not planned.
+8. **Detailed flow for file lifecycle:** When exactly are S3 artifacts cleaned up? The "3-day rule" applies to workspace files, but S3 cleanup timing remains unclear.
+9. **Browser Use vs. agent-browser:** The relationship with Durevich's team regarding browser integration needs clarification.
+
+## Source
+Original transcript: `meeting_1.txt`

pyproject.toml

@@ -5,6 +5,7 @@ description = ""
 readme = "README.md"
 requires-python = ">=3.13"
 dependencies = [
+    "docker>=7.1.0",
     "fastapi>=0.116.1",
     "opentelemetry-api>=1.31.1",
     "opentelemetry-exporter-otlp-proto-http>=1.31.1",
@@ -21,6 +22,7 @@ dev = [
     "mypy>=1.18.2",
     "pytest>=8.4.2",
     "ruff>=0.13.1",
+    "types-docker>=7.1.0.20260402",
     "types-pyyaml>=6.0.12.20250915",
 ]
 

repository/chat.py

@@ -0,0 +1,24 @@
+from uuid import UUID
+
+from domain.chat import Chat
+from usecase.interface import ChatRepository
+
+
+class InMemoryChatRepository(ChatRepository):
+    def __init__(self) -> None:
+        self._by_id: dict[UUID, Chat] = {}
+
+    def get(self, chat_id: UUID) -> Chat | None:
+        return self._by_id.get(chat_id)
+
+    def list_by_workspace_id(self, workspace_id: UUID) -> list[Chat]:
+        return sorted(
+            (c for c in self._by_id.values() if c.workspace_id == workspace_id),
+            key=lambda c: (c.created_at, c.chat_id),
+        )
+
+    def save(self, chat: Chat) -> None:
+        self._by_id[chat.chat_id] = chat
+
+    def delete(self, chat_id: UUID) -> None:
+        self._by_id.pop(chat_id, None)

repository/chat_file.py

@@ -0,0 +1,57 @@
+from uuid import UUID
+
+from domain.chat import ChatAttachmentName, ChatFile
+from usecase.interface import ChatFileRepository
+
+
+class InMemoryChatFileRepository(ChatFileRepository):
+    def __init__(self) -> None:
+        self._by_id: dict[UUID, ChatFile] = {}
+        self._by_chat_and_name: dict[tuple[UUID, str], UUID] = {}
+
+    def get(self, file_id: UUID) -> ChatFile | None:
+        return self._by_id.get(file_id)
+
+    def get_by_chat_id_and_name(
+        self,
+        chat_id: UUID,
+        name: ChatAttachmentName,
+    ) -> ChatFile | None:
+        fid = self._by_chat_and_name.get((chat_id, name.value))
+        if fid is None:
+            return None
+        return self._by_id.get(fid)
+
+    def list_by_chat_id(self, chat_id: UUID) -> list[ChatFile]:
+        return sorted(
+            (f for f in self._by_id.values() if f.chat_id == chat_id),
+            key=lambda f: (f.created_at, f.file_id),
+        )
+
+    def save(self, chat_file: ChatFile) -> None:
+        key = (chat_file.chat_id, chat_file.name.value)
+        existing_at_key = self._by_chat_and_name.get(key)
+        if existing_at_key is not None and existing_at_key != chat_file.file_id:
+            self._by_id.pop(existing_at_key, None)
+
+        previous = self._by_id.get(chat_file.file_id)
+        if previous is not None:
+            prev_key = (previous.chat_id, previous.name.value)
+            if self._by_chat_and_name.get(prev_key) == previous.file_id:
+                del self._by_chat_and_name[prev_key]
+
+        self._by_id[chat_file.file_id] = chat_file
+        self._by_chat_and_name[key] = chat_file.file_id
+
+    def delete(self, file_id: UUID) -> None:
+        chat_file = self._by_id.pop(file_id, None)
+        if chat_file is None:
+            return
+        key = (chat_file.chat_id, chat_file.name.value)
+        if self._by_chat_and_name.get(key) == file_id:
+            del self._by_chat_and_name[key]
+
+    def delete_by_chat_id(self, chat_id: UUID) -> None:
+        file_ids = [f.file_id for f in self._by_id.values() if f.chat_id == chat_id]
+        for fid in file_ids:
+            self.delete(fid)

repository/sandbox_lock.py

@@ -0,0 +1,70 @@
+import threading
+from dataclasses import dataclass
+from types import TracebackType
+from typing import Protocol
+from uuid import UUID
+
+from usecase.interface import LockContext, SandboxLifecycleLocker
+
+
+class _SyncLock(Protocol):
+    def acquire(self, blocking: bool = True, timeout: float = -1) -> bool: ...
+
+    def release(self) -> None: ...
+
+
+@dataclass(slots=True)
+class _LockEntry:
+    lock: _SyncLock
+    users: int = 0
+
+
+class _ChatLock(LockContext):
+    def __init__(
+        self,
+        locker: 'ProcessLocalSandboxLifecycleLocker',
+        chat_id: UUID,
+        entry: _LockEntry,
+    ) -> None:
+        self._locker = locker
+        self._chat_id = chat_id
+        self._entry = entry
+
+    def __enter__(self) -> None:
+        self._entry.lock.acquire()
+
+    def __exit__(
+        self,
+        exc_type: type[BaseException] | None,
+        exc: BaseException | None,
+        traceback: TracebackType | None,
+    ) -> bool | None:
+        self._entry.lock.release()
+        self._locker._release(self._chat_id, self._entry)
+        return None
+
+
+class ProcessLocalSandboxLifecycleLocker(SandboxLifecycleLocker):
+    def __init__(self) -> None:
+        self._registry_lock = threading.Lock()
+        self._locks_by_chat_id: dict[UUID, _LockEntry] = {}
+
+    def lock(self, chat_id: UUID) -> LockContext:
+        with self._registry_lock:
+            entry = self._locks_by_chat_id.get(chat_id)
+            if entry is None:
+                entry = _LockEntry(lock=threading.Lock())
+                self._locks_by_chat_id[chat_id] = entry
+            entry.users += 1
+
+        return _ChatLock(self, chat_id, entry)
+
+    def _release(self, chat_id: UUID, entry: _LockEntry) -> None:
+        with self._registry_lock:
+            entry.users -= 1
+            if entry.users != 0:
+                return
+
+            current_entry = self._locks_by_chat_id.get(chat_id)
+            if current_entry is entry:
+                del self._locks_by_chat_id[chat_id]

repository/sandbox_session.py

@@ -0,0 +1,45 @@
+import threading
+from datetime import datetime
+from uuid import UUID
+
+from domain.sandbox import SandboxSession
+from usecase.interface import SandboxSessionRepository
+
+
+class InMemorySandboxSessionRepository(SandboxSessionRepository):
+    def __init__(self) -> None:
+        self._sessions_by_chat_id: dict[UUID, SandboxSession] = {}
+        self._lock = threading.Lock()
+
+    def replace_all(self, sessions: list[SandboxSession]) -> None:
+        with self._lock:
+            self._sessions_by_chat_id = {
+                session.chat_id: session for session in sessions
+            }
+
+    def get_active_by_chat_id(self, chat_id: UUID) -> SandboxSession | None:
+        with self._lock:
+            return self._sessions_by_chat_id.get(chat_id)
+
+    def list_expired(self, now: datetime) -> list[SandboxSession]:
+        with self._lock:
+            return [
+                session
+                for session in self._sessions_by_chat_id.values()
+                if session.expires_at <= now
+            ]
+
+    def count_active(self) -> int:
+        with self._lock:
+            return len(self._sessions_by_chat_id)
+
+    def save(self, session: SandboxSession) -> None:
+        with self._lock:
+            self._sessions_by_chat_id[session.chat_id] = session
+
+    def delete(self, session_id: UUID) -> None:
+        with self._lock:
+            for chat_id, session in tuple(self._sessions_by_chat_id.items()):
+                if session.session_id == session_id:
+                    del self._sessions_by_chat_id[chat_id]
+                    return

repository/workspace.py

@@ -0,0 +1,26 @@
+from uuid import UUID
+
+from domain.workspace import Workspace
+from usecase.interface import WorkspaceRepository
+
+
+class InMemoryWorkspaceRepository(WorkspaceRepository):
+    def __init__(self) -> None:
+        self._by_id: dict[UUID, Workspace] = {}
+        self._user_id_to_workspace_id: dict[UUID, UUID] = {}
+
+    def get(self, workspace_id: UUID) -> Workspace | None:
+        return self._by_id.get(workspace_id)
+
+    def get_by_user_id(self, user_id: UUID) -> Workspace | None:
+        wid = self._user_id_to_workspace_id.get(user_id)
+        if wid is None:
+            return None
+        return self._by_id.get(wid)
+
+    def save(self, workspace: Workspace) -> None:
+        existing_wid = self._user_id_to_workspace_id.get(workspace.user_id)
+        if existing_wid is not None and existing_wid != workspace.workspace_id:
+            self._by_id.pop(existing_wid, None)
+        self._by_id[workspace.workspace_id] = workspace
+        self._user_id_to_workspace_id[workspace.user_id] = workspace.workspace_id

tasks.md

@@ -1,142 +1,361 @@
-# План работ: web-python-skelet
+# План работ: master-service MVP sandbox
 
 ## Контекст
 
-- Источник требований: `AGENTS.md` и ADR `docs/001`-`docs/004`
+- Источники требований: `docs/master.md`, `meetings/meeting_1_document.md`, `README.md`, `docs/*`
-- Текущее состояние: в `adapter/`, `domain/`, `usecase/`, `repository/`, `test/` пока только `__init__.py`
+- Базовый template уже готов: typed config, DI container, observability, FastAPI adapter и versioned API под `/api/v1`
-- Отсутствуют рабочие каталоги и файлы из целевой структуры: `adapter/config/`, `adapter/otel/`, `adapter/di/`, `adapter/http/fastapi/`, `config/`, `main.py`
+- Текущая цель: минимальное управление Docker sandbox без auth
-- Ограничения: `docs/` и `tasks.md` не добавлять в git; коммиты не делать; работать по одной задаче
+- MVP API: `POST /api/v1/create`
-- ADR пока покрывают архитектуру, новые ADR нужны только если по ходу работ изменится решение
+- Sandbox policy: TTL `300` секунд, одна активная sandbox на `chat_id`, повторный `create` переиспользует активную сессию
+- Volume policy: chat volume `rw`, dependencies volume `ro`, lambda-tools volume `ro`
+- Host paths вычисляются из typed config, а HTTP request передает только `chat_id`
+- Cleanup выполняется периодическим in-process loop внутри master-service
+
+## Вне текущего scope
+
+- auth и access control
+- p2p lease и WebSocket transport
+- workspace/chat CRUD API
+- chat files, artifacts, S3, quota и retention policy
+- центральная БД и multi-node orchestration
 
 ## Правила выполнения
 
-- Каждую задачу выполнять отдельным заходом, без параллельной реализации
+- Выполняем по одной задаче
-- Каждый субагент отдает diff, список измененных файлов и проверку, но не делает commit
+- Коммиты не делаем
-- Если в задаче всплывает архитектурное изменение, остановиться и вынести вопрос на согласование
+- Если по ходу нужна смена архитектуры, останавливаемся и согласуем решение
+- `domain/` и `usecase/` не импортируют Docker, FastAPI, OpenTelemetry, env или YAML
+- Inner layers работают только через минимальные domain сущности и usecase порты
 
 ## Очередь задач
 
-### T01. Базовый каркас домена и usecase
+### M01. ADR и минимальный sandbox scaffolding
 
-- Исполнитель: `primary-agent` (scaffolding)
+- Исполнитель: `primary-agent`
 - Статус: completed
 - Зависимости: нет
 - Commit required: no
-- Scope: создать базовые файлы и контракты в `domain/`, `usecase/`, `repository/`
+- Scope: зафиксировать MVP-решение в ADR и создать минимальные сущности, ошибки и usecase-контракты для sandbox orchestration
-- Файлы: `domain/error.py`, `domain/user.py`, `usecase/interface.py`, `usecase/user.py`, `repository/user.py`
+- Файлы: `docs/006-mvp-docker-sandbox-orchestration.md`, `domain/sandbox.py`, `domain/error.py`, `usecase/interface.py`, `usecase/sandbox.py`
-- Критерии приемки: зависимости направлены внутрь; в `domain/` и `usecase/` нет FastAPI/OTel; есть пример сущности, ошибок, портов и простого usecase
+- Критерии приемки: в `domain/` есть минимальная `SandboxSession` и sandbox-ошибки; в `usecase/` есть порты `SandboxSessionRepository`, `SandboxRuntime` и `Clock`; созданы скелеты `CreateSandbox` и `CleanupExpiredSandboxes`; ADR занимает 10-20 строк
 
-### T02. Конфиг из YAML и env
+### M02. Typed config для sandbox runtime
 
 - Субагент: `feature-developer`
 - Статус: completed
-- Зависимости: `T01`
+- Зависимости: `M01`
 - Commit required: no
-- Scope: собрать typed-config слой в `adapter/config/` и подготовить базовые yaml-файлы
+- Scope: расширить typed-config слоем `sandbox` с настройками image, TTL, cleanup interval, host paths и container mount paths
-- Файлы: `adapter/config/*`, `config/app.yaml`
+- Файлы: `adapter/config/model.py`, `adapter/config/loader.py`, `config/app.yaml`
-- Критерии приемки: конфиг собирается в одну dataclass-структуру; секреты читаются из env; парсинг и валидация не протекают в inner layers
+- Решение: chat host path строится как путь под общим `sandbox.chats_root/<chat_id>`; request не передает host path напрямую
+- Критерии приемки: конфиг собирается в typed dataclass-дерево; дефолтный TTL равен `300`; есть отдельные настройки для `chats_root`, `dependencies_host_path`, `lambda_tools_host_path`, `chat_mount_path`, `dependencies_mount_path`, `lambda_tools_mount_path`, `cleanup_interval_seconds`; inner layers не читают env
 
-### T03. Observability порты и OTel adapter
+### M03. Docker runtime adapter для sandbox lifecycle
 
 - Субагент: `feature-developer`
 - Статус: completed
-- Зависимости: `T01`, `T02`
+- Зависимости: `M01`, `M02`
 - Commit required: no
-- Scope: реализовать логгер, метрики, трейсинг и bootstrap OTel в `adapter/otel/` через интерфейсы из `usecase/interface.py`
+- Scope: реализовать outer adapter над Docker для создания и остановки sandbox контейнера с нужными labels и mount policy
-- Файлы: `adapter/otel/*`, `config/otel-collector.yaml`
+- Файлы: `adapter/docker/runtime.py`, `adapter/docker/__init__.py`
-- Критерии приемки: inner layers знают только интерфейсы; OTLP exporter настраивается из конфига; нет кастомного trace middleware
+- Ограничения: все Docker-детали остаются в `adapter/`; runtime не должен протекать во внутренние слои
+- Критерии приемки: runtime умеет создать sandbox container по входным параметрам usecase; chat volume монтируется как `rw`; dependency и lambda-tools volumes монтируются как `ro`; контейнер получает labels с `session_id`, `chat_id` и `expires_at`; runtime переводит ошибки Docker в понятные исключения адаптера
 
-### T04. Composition root и lifetime singleton-объектов
+### M04. In-memory session repository и usecase `CreateSandbox`
 
 - Субагент: `feature-developer`
 - Статус: completed
-- Зависимости: `T01`, `T02`, `T03`
+- Зависимости: `M01`, `M02`, `M03`
 - Commit required: no
-- Scope: собрать контейнер и startup wiring в `adapter/di/`
+- Scope: реализовать in-memory registry активных sandbox-сессий и usecase создания sandbox с логикой reuse по `chat_id`
-- Файлы: `adapter/di/container.py`, `adapter/di/__init__.py`
+- Файлы: `repository/sandbox_session.py`, `usecase/sandbox.py`, `adapter/di/container.py`
-- Критерии приемки: repository/usecase создаются один раз на старте; контейнер хранит инстансы явно; нет пересоздания на HTTP-запрос
+- Решение: если по `chat_id` есть неистекшая сессия, usecase возвращает ее без нового container start; если сессия истекла, usecase инициирует stop старой sandbox и создает новую
+- Критерии приемки: одна активная sandbox на `chat_id`; TTL-логика использует порт `Clock`; usecase не импортирует Docker; container wiring остается singleton-based
 
-### T05. FastAPI adapter как заменяемый web layer
+### M05. Cleanup expired sandboxes и lifecycle wiring
 
 - Субагент: `feature-developer`
 - Статус: completed
-- Зависимости: `T04`
+- Зависимости: `M04`
 - Commit required: no
-- Scope: поднять HTTP adapter в `adapter/http/fastapi/` с app factory, lifespan, dependencies, middleware и router ` /api/v1`
+- Scope: реализовать usecase cleanup просроченных sandbox и подключить периодический cleanup loop в FastAPI lifecycle
-- Файлы: `adapter/http/fastapi/*`, `main.py`
+- Файлы: `usecase/sandbox.py`, `adapter/di/container.py`, `adapter/http/fastapi/app.py`, при необходимости `adapter/http/fastapi/dependencies.py`
-- Критерии приемки: FastAPI изолирован в adapter-слое; handlers тонкие; request logging и metrics middleware подключены; usecase/repository берутся из контейнера
+- Ограничения: не ломать ADR про раннее OTel instrumentation; lifecycle loop должен стартовать и останавливаться один раз
+- Критерии приемки: cleanup находит истекшие сессии, останавливает sandbox через runtime и удаляет их из registry; интервал cleanup берется из конфига; shutdown корректно завершает фоновую задачу
 
-### T06. Локальный runtime и compose-окружение
+### M06. HTTP endpoint `POST /api/v1/create`
 
 - Субагент: `feature-developer`
 - Статус: completed
-- Зависимости: `T02`, `T03`, `T05`
+- Зависимости: `M04`
 - Commit required: no
-- Scope: добавить контейнерный runtime для сервиса и compose-окружение с OTel и UI для просмотра логов, метрик и трейсов
+- Scope: добавить минимальную HTTP ручку для создания или переиспользования sandbox без auth
-- Файлы: `Dockerfile`, `docker-compose.yml`
+- Файлы: `adapter/http/fastapi/schemas.py`, `adapter/http/fastapi/dependencies.py`, `adapter/http/fastapi/routers/v1/router.py`, при необходимости `adapter/di/container.py`
-- Ограничения: не трогать репозиторный `config/app.yaml`; docker должен прокидывать свой runtime-config внутрь контейнера; Dockerfile только с двумя стадиями `build` и `run`
+- Request: `{ "chat_id": "..." }`
-- Критерии приемки: `make compose-build` и `make compose-up` опираются на существующие файлы; сервис поднимается в контейнере; OTel telemetry уходит в dockerized stack; есть UI для просмотра логов, метрик и трейсов; для локального docker-окружения достаточно только `Dockerfile` и `docker-compose.yml`
+- Response: `session_id`, `chat_id`, `container_id`, `status`, `expires_at`
+- Критерии приемки: router остается тонким; handler только переводит HTTP input в команду usecase и маппит ошибки в HTTP; endpoint живет под `/api/v1/create`; auth не добавляется
 
-### T07. Тесты на lifetimes, config и HTTP smoke
+### M07. Тесты для create, reuse, TTL и mount policy
 
 - Субагент: `test-engineer`
-- Статус: pending
+- Статус: completed
-- Зависимости: `T01`, `T02`, `T03`, `T04`, `T05`, `T06`
+- Зависимости: `M03`, `M04`, `M05`, `M06`
 - Commit required: no
-- Scope: покрыть тестами ключевые архитектурные гарантии
+- Scope: покрыть тестами ключевое поведение MVP без запуска реального production Docker stack
 - Файлы: `test/*`
-- Критерии приемки: есть тест на singleton lifetime для repository/usecase; есть тест merge YAML+env; есть smoke-тест для ` /api/v1`; тесты не тянут FastAPI/OTel в inner layers
+- Критерии приемки: есть unit-тесты для `CreateSandbox` и `CleanupExpiredSandboxes` с fake clock; есть HTTP smoke-тест для `POST /api/v1/create`; есть adapter-level тест с mock Docker client на mount policy `chat=rw`, `deps=ro`, `tools=ro`; тесты не тащат FastAPI или Docker в inner-layer тесты
 
-### T08. Архитектурный и boundary review
+### M08. Архитектурный и boundary review по MVP sandbox
 
 - Субагент: `code-reviewer`
-- Статус: pending
+- Статус: completed
-- Зависимости: `T07`
+- Зависимости: `M07`
 - Commit required: no
-- Scope: проверить импорты, соблюдение слоев, startup lifetimes и заменяемость web adapter
+- Scope: проверить соблюдение clean architecture, dependency direction и соответствие MVP-ограничениям
 - Файлы: весь измененный код
-- Критерии приемки: dependency direction не нарушен; FastAPI и OTel не протекают в `domain/` и `usecase/`; замечания сформулированы как точечные правки или подтверждение готовности к review
+- Критерии приемки: Docker остается только во внешнем adapter; FastAPI не протекает в `domain/` и `usecase/`; TTL и mount policy читаются как явные, тестируемые правила; замечания сформулированы как точечные правки или подтверждение готовности
 
-### T09. Конфигурируемый runtime observability
+## Follow-up после M08 review
+
+### M09. Сериализация lifecycle sandbox по `chat_id`
 
 - Субагент: `feature-developer`
 - Статус: completed
-- Зависимости: `T03`, `T04`, `T05`
+- Зависимости: `M08`
 - Commit required: no
-- Scope: сделать конфигурируемый runtime observability с настраиваемым sink и форматом логов, плюс отдельными флагами для метрик и трейсов
+- Scope: убрать гонки между параллельными `create` и cleanup для одного `chat_id`
-- Файлы: `adapter/config/*`, `adapter/observability/*`, `adapter/otel/*`, `adapter/di/*`, `adapter/http/fastapi/*`, `config/app.yaml`, при необходимости `main.py`
+- Файлы: `usecase/interface.py`, `usecase/sandbox.py`, `repository/sandbox_lock.py` или другой outer-layer lock implementation, `adapter/di/container.py`
-- Конфиг: `logging.output=stdout|file|otel`, `logging.format=text|json`, `logging.file_path=...`, `metrics.enabled=true|false`, `tracing.enabled=true|false`
+- Решение: ввести явный usecase-port для process-local lock по `chat_id`; outer-layer реализация держит per-chat lock registry; `CreateSandbox` и `CleanupExpiredSandboxes` выполняют мутации сессии под этим lock
-- Решение: вынести выбор runtime в отдельный adapter-layer factory; `domain/` и `usecase/` не менять; для `stdout` и `file` поддержать оба формата `text` и `json`; `logging.file_path` использовать только при `logging.output=file`; при отключенных метриках и трейсах использовать `Noop`-реализации; OTel runtime поднимать только если нужен хотя бы для одного сигнала
+- Критерии приемки: для одного `chat_id` не поднимаются два sandbox при concurrent create; create-vs-cleanup не оставляет orphan container; locking не протекает в HTTP и Docker adapter как бизнес-логика
-- Критерии приемки: при `logging.output=stdout` логи идут в stdout в формате `text` или `json` по конфигу; при `logging.output=file` логи пишутся в файл по пути `logging.file_path` в формате `text` или `json`; при `logging.output=otel` логи уходят в collector; `metrics.enabled=false` отключает метрики и metrics middleware; `tracing.enabled=false` отключает FastAPI instrumentation и tracer runtime; DI продолжает отдавать единый runtime через контейнер; внутренние слои по-прежнему знают только порты
 
-### T10. ADR: раннее подключение OTel к FastAPI
+### M10. Устойчивый cleanup и вынос blocking cleanup из event loop
-
-- Исполнитель: `primary-agent` (docs)
-- Статус: completed
-- Зависимости: нет
-- Commit required: no
-- Scope: зафиксировать правило, что FastAPI OTel instrumentation выполняется до первой сборки `middleware_stack`
-- Файлы: `docs/005-fastapi-otel-early-instrumentation.md`
-- Критерии приемки: ADR занимает 10-20 строк; описаны context, decision, consequences; решение не переписывает историю прошлых ADR
-
-### T11. Перенос FastAPI OTel bootstrap в app factory
 
 - Субагент: `feature-developer`
 - Статус: completed
-- Зависимости: `T10`
+- Зависимости: `M09`
 - Commit required: no
-- Scope: перенести создание container, установку `FastAPIInstrumentor.instrument_app(...)` из `lifespan` в `create_app`, оставив в `lifespan` только shutdown
+- Scope: сделать cleanup устойчивым к частичным ошибкам и не блокировать FastAPI event loop синхронным Docker stop
-- Файлы: `adapter/http/fastapi/app.py`, `adapter/http/fastapi/lifespan.py`, при необходимости `adapter/http/fastapi/dependencies.py`
+- Файлы: `usecase/sandbox.py`, `adapter/http/fastapi/app.py`
-- Ограничения: не использовать ручной rebuild `app.middleware_stack`; не менять `domain/` и `usecase/`; не добавлять бизнес-логику; сохранить singleton-lifetime container
+- Решение: `CleanupExpiredSandboxes` обрабатывает stop/delete по каждой сессии отдельно и продолжает batch; HTTP cleanup loop выносит blocking cleanup work в thread через adapter-layer orchestration
-- Критерии приемки: instrumentation происходит до первой сборки middleware stack; `OpenTelemetryMiddleware` попадает в runtime stack без workaround; shutdown закрывает instrumentation и runtime один раз; compose-конфиг продолжает работать
+- Критерии приемки: ошибка на одной expired session не мешает чистить остальные; background cleanup loop не умирает после ошибки; blocking cleanup больше не выполняется прямо в event loop
 
-### T12. Регрессионная проверка HTTP telemetry wiring
+### M11. Удаление не-MVP user surface из приложения
+
+- Субагент: `feature-developer`
+- Статус: completed
+- Зависимости: `M08`
+- Commit required: no
+- Scope: убрать из runtime app неотносящиеся к MVP user endpoint и seed user wiring
+- Файлы: `adapter/http/fastapi/routers/v1/router.py`, `adapter/http/fastapi/dependencies.py`, `adapter/http/fastapi/schemas.py`, `adapter/di/container.py`
+- Решение: оставить в MVP только `health` и sandbox API; примерный user code может остаться в репозитории как template, но не должен быть подключен в runtime app
+- Критерии приемки: `GET /api/v1/users/{user_id}` больше не опубликован; container не создает seeded user repository/usecase для runtime app; незапрошенная user-surface area исчезает
+
+### M12. Регрессионные тесты на race conditions и cleanup resilience
 
 - Субагент: `test-engineer`
-- Статус: pending
+- Статус: completed
-- Зависимости: `T11`
+- Зависимости: `M09`, `M10`, `M11`
 - Commit required: no
-- Scope: добавить проверку, что раннее instrumentation wiring сохраняется и не требует ручного rebuild middleware stack
+- Scope: добавить тесты на новые гарантии после review fixes
 - Файлы: `test/*`
-- Ограничения: без реального collector; проверять через ASGI/lifespan или локальные assertions по app runtime; не тянуть FastAPI и OTel в inner-layer тесты
+- Критерии приемки: есть тест на duplicate create для одного `chat_id`; есть тест на create-vs-cleanup race или эквивалентную сериализацию; есть тест, что cleanup продолжает batch после stop failure; HTTP smoke/regression тесты обновлены под удаление user endpoint
-- Критерии приемки: тест подтверждает, что при включенных metrics/tracing `OpenTelemetryMiddleware` присутствует в runtime stack; тест не зависит от внешнего OTel collector; существующие архитектурные границы не нарушены
+
+### M13. Повторный boundary review после fix-pass
+
+- Субагент: `code-reviewer`
+- Статус: completed
+- Зависимости: `M12`
+- Commit required: no
+- Scope: проверить, что must-fix и should-fix замечания из `M08` закрыты без нарушения clean architecture
+- Файлы: весь измененный код после `M09`-`M12`
+- Критерии приемки: нет гонки на one-sandbox-per-chat; cleanup не блокирует event loop и не валится на первом stop failure; runtime app не публикует лишний user API; замечания сведены к minor или отсутствуют
+
+### M14. Починка mypy-типизации тестов после sandbox MVP
+
+- Субагент: `test-engineer`
+- Статус: completed
+- Зависимости: `M07`
+- Commit required: no
+- Scope: устранить текущие ошибки `make pre-commit` в test-suite без изменения production behavior
+- Файлы: `test/test_docker_runtime.py`, `test/test_create_http.py`, при необходимости общие test helpers в `test/*`
+- Ошибки: несовместимый fake Docker client для `DockerSandboxRuntime`, неточная типизация `run_calls` и ASGI message payload, использование `object` вместо типизированных test doubles для `AppRepositories`, `AppUsecases`, `AppContainer`
+- Решение: сделать test doubles типизированными через совместимые fake classes или локальные protocols; убрать `object` и неиндексируемые `dict[str, object]` там, где mypy не может вывести типы
+- Критерии приемки: `uv run mypy .` проходит; `make pre-commit` доходит как минимум до pytest stage; production code не меняется или меняется только при явной необходимости для testability
+
+## Follow-up после M13 review
+
+### M15. Канонизация и валидация `chat_id`
+
+- Субагент: `feature-developer`
+- Статус: completed
+- Зависимости: `M13`
+- Commit required: no
+- Scope: сделать `chat_id` строго UUID и убрать path alias/whole-root mount риск через неканоничные значения
+- Файлы: `adapter/http/fastapi/schemas.py`, `adapter/docker/runtime.py`, при необходимости `usecase/sandbox.py` и тесты в `test/*`
+- Решение: принять `chat_id` как UUID на HTTP boundary, использовать его каноничную строковую форму дальше в usecase/repository/path construction и не принимать произвольные path-like строки
+- Критерии приемки: не-UUID значения отклоняются на HTTP boundary с `400/422`; UUID используется как единое каноничное значение для lock key, repository key и filesystem path; появляются регрессионные тесты на invalid `chat_id`
+
+### M16. Lifecycle reconciliation на startup/shutdown
+
+- Субагент: `feature-developer`
+- Статус: completed
+- Зависимости: `M13`
+- Commit required: no
+- Scope: устранить restart-gap между in-memory registry и уже запущенными Docker containers
+- Файлы: `adapter/docker/runtime.py`, `adapter/di/container.py`, `adapter/http/fastapi/app.py`, при необходимости новые outer-layer helper files и тесты в `test/*`
+- Решение: основная стратегия — reconciliation по Docker labels на startup, чтобы после restart master-service продолжал видеть уже запущенные sandbox и не поднимал дубликаты; graceful shutdown cleanup остается опциональным дополнением
+- Критерии приемки: после restart master-service может восстановить/синхронизировать state по Docker labels без потери работающих agent containers; one-sandbox-per-chat не нарушается из-за пустого in-memory registry; lifecycle policy явно зафиксирована и покрыта тестами
+
+### M17. Управление жизненным циклом per-chat locks
+
+- Субагент: `feature-developer`
+- Статус: completed
+- Зависимости: `M13`
+- Commit required: no
+- Scope: ограничить неограниченный рост registry locks по числу когда-либо увиденных `chat_id`
+- Файлы: `repository/sandbox_lock.py`, при необходимости тесты в `test/*`
+- Решение: добавить eviction/ref-count/weakref policy во внешнем lock registry без нарушения сериализации lifecycle для активного `chat_id`
+- Критерии приемки: registry locks не растет бесконечно без причины; сериализация для активных чатов сохраняется; поведение покрыто тестами
+
+### M18. Перевести sandbox ids на UUID types
+
+- Субагент: `feature-developer`
+- Статус: completed
+- Зависимости: `M15`
+- Commit required: no
+- Scope: сделать `chat_id` и `session_id` типом `UUID` внутри sandbox scope, оставив `container_id` строкой как внешний Docker identifier
+- Файлы: `domain/sandbox.py`, `usecase/interface.py`, `usecase/sandbox.py`, `repository/sandbox_session.py`, `adapter/http/fastapi/*`, `adapter/docker/runtime.py`, `adapter/di/container.py`, `test/*`
+- Решение: HTTP boundary принимает/возвращает UUID, usecase и repository работают с UUID objects, Docker labels продолжают сериализоваться в строки через `str(uuid)`
+- Критерии приемки: внутри sandbox flow `chat_id` и `session_id` больше не строки; `container_id` остается `str`; pydantic корректно сериализует UUID в response; `make pre-commit` проходит
+
+## Follow-up после issue #11 observability
+
+### M19. ADR и observability contracts для sandbox lifecycle
+
+- Исполнитель: `primary-agent`
+- Статус: completed
+- Зависимости: `M18`
+- Commit required: yes
+- Commit message: `add sandbox observability contracts`
+- Scope: зафиксировать sandbox lifecycle observability policy в ADR-lite и подготовить минимальные контракты для traces и current-state metrics без нарушения clean architecture
+- Файлы: `docs/008-sandbox-lifecycle-observability.md`, `usecase/interface.py`, `repository/sandbox_session.py`, `adapter/otel/metrics.py`, `adapter/observability/noop.py`
+- Решение: добавить в `Metrics` порт операцию `set(...)` для gauge-like current-state сигналов; добавить в `SandboxSessionRepository` `count_active()` как источник truth для `sandbox.active.count`
+- Критерии приемки: ADR занимает 10-20 строк; inner layers по-прежнему знают только порты `Logger`/`Metrics`/`Tracer`; current-state метрика активных sandbox выражается без OTel imports во внутреннем слое
+
+### M20. Трейсы и метрики в sandbox usecases
+
+- Субагент: `feature-developer`
+- Статус: completed
+- Зависимости: `M19`
+- Commit required: yes
+- Commit message: `instrument sandbox usecases`
+- Scope: добавить spans и ключевые lifecycle metrics в `CreateSandbox` и `CleanupExpiredSandboxes`
+- Файлы: `usecase/sandbox.py`, `adapter/di/container.py`, при необходимости тесты в `test/*`
+- Решение: usecase получает `Metrics` и `Tracer` через конструктор; `CreateSandbox` и `CleanupExpiredSandboxes` публикуют `sandbox.create.total`, `sandbox.cleanup.total`, `sandbox.cleanup.error.total` и обновляют `sandbox.active.count` после мутаций registry
+- Критерии приемки: есть spans `usecase.create_sandbox` и `usecase.cleanup_expired_sandboxes`; span attrs и metric attrs включают ключевые lifecycle identifiers/result fields; reuse/replace/cleanup paths наблюдаемы без OTel imports в usecase
+
+### M21. Трейсы и runtime metrics в Docker adapter и reconciliation
+
+- Субагент: `feature-developer`
+- Статус: completed
+- Зависимости: `M19`
+- Commit required: yes
+- Commit message: `instrument sandbox docker runtime`
+- Scope: добавить observability в `DockerSandboxRuntime` и reconciliation path для Docker operations и current-state sync
+- Файлы: `adapter/docker/runtime.py`, `adapter/sandbox/reconciliation.py`, `adapter/di/container.py`, при необходимости тесты в `test/*`
+- Решение: `DockerSandboxRuntime` получает `Metrics` и `Tracer`; create/stop/list публикуют duration histograms `sandbox.runtime.create.duration_ms`, `sandbox.runtime.stop.duration_ms`, `sandbox.runtime.list_active.duration_ms`, error counter `sandbox.runtime.error.total` и span attrs по chat/session/container; reconciliation обновляет `sandbox.active.count` по registry snapshot
+- Критерии приемки: Docker adapter остается во внешнем слое; ошибки Docker операций отражаются в spans и metrics; после startup reconciliation current-state метрика активных sandbox синхронизирована с registry
+
+### M22. Тесты на sandbox observability
+
+- Субагент: `test-engineer`
+- Статус: completed
+- Зависимости: `M20`, `M21`
+- Commit required: yes
+- Commit message: `add sandbox observability tests`
+- Scope: покрыть regression tests новую observability policy без реального OTel backend
+- Файлы: `test/test_sandbox_usecase.py`, `test/test_docker_runtime.py`, при необходимости новые focused tests в `test/*`
+- Решение: использовать типизированные fake metrics/tracer implementations и проверить names/attrs ключевых spans и metrics на create/reuse/replace/cleanup/runtime paths
+- Критерии приемки: тесты подтверждают spans и metrics на usecase и adapter paths; constructor wiring обновлен без mypy regressions; `make typecheck` и релевантный `pytest` проходят
+
+### M23. Boundary review для sandbox observability
+
+- Субагент: `code-reviewer`
+- Статус: in_progress
+- Зависимости: `M22`
+- Commit required: no
+- Scope: проверить, что observability изменения закрывают issue #11 и FR-034 без нарушения clean architecture
+- Файлы: весь измененный код после `M19`-`M22`
+- Критерии приемки: inner layers не импортируют OTel; Docker-specific tracing остается в `adapter/docker/`; current-state и duration metrics достаточно покрывают sandbox lifecycle; замечания сведены к minor или отсутствуют
+
+## Follow-up после M23 boundary review
+
+### M24. Исправить replace trace identity в CreateSandbox
+
+- Субагент: `feature-developer`
+- Статус: completed
+- Зависимости: `M23`
+- Commit required: yes
+- Commit message: `fix sandbox replace trace identity`
+- Scope: устранить смешение old/new sandbox identifiers в replace path usecase tracing
+- Файлы: `usecase/sandbox.py`, при необходимости точечные тесты в `test/*`
+- Решение: сохранять старые и новые sandbox identifiers в отдельных span attrs или child spans так, чтобы replace success и replace failure оставались однозначно трассируемыми
+- Критерии приемки: replace path не перетирает previous/new identifiers; при replace failure span остается консистентным и отражает обе стороны lifecycle
+
+### M25. Добрать failure-path observability regression tests
+
+- Субагент: `test-engineer`
+- Статус: completed
+- Зависимости: `M24`
+- Commit required: yes
+- Commit message: `add sandbox observability failure tests`
+- Scope: покрыть tests для replace-failure trace, cleanup error metrics/spans и Docker stop observability
+- Файлы: `test/test_sandbox_usecase.py`, `test/test_docker_runtime.py`, при необходимости другие focused tests в `test/*`
+- Решение: использовать presence-based assertions и проверять ключевые span/metric contracts без brittle exact-order checks
+- Критерии приемки: есть тест на replace failure tracing; есть тест на `sandbox.cleanup.error.total`; есть тесты на Docker stop observability для success/error/not_found или эквивалентного набора outcome paths
+
+### M26. Повторный boundary review для sandbox observability
+
+- Субагент: `code-reviewer`
+- Статус: in_progress
+- Зависимости: `M25`
+- Commit required: no
+- Scope: подтвердить, что follow-up fixes закрыли M23 замечания без новых boundary нарушений
+- Файлы: весь измененный код после `M24`-`M25`
+- Критерии приемки: нет замечаний по replace tracing identity и missing failure-path observability coverage; clean architecture по-прежнему соблюдена
+
+## Follow-up после M26 boundary review
+
+### M27. Компенсация save failure после runtime.create
+
+- Субагент: `feature-developer`
+- Статус: completed
+- Зависимости: `M26`
+- Commit required: yes
+- Commit message: `fix sandbox create rollback gap`
+- Scope: не оставлять untracked running container и неконсистентный `sandbox.active.count` при падении `repository.save()` после успешного `runtime.create()`
+- Файлы: `usecase/sandbox.py`, при необходимости точечные тесты в `test/*`
+- Решение: сделать create/replace path registry-safe через rollback или другой явный compensation path без нарушения clean architecture
+- Критерии приемки: save failure не оставляет новый container в runtime без registry state; `sandbox.active.count` отражает финальное committed state; replace и fresh-create failure paths консистентны
+
+### M28. Регрессии на rollback и startup failure observability
+
+- Субагент: `test-engineer`
+- Статус: completed
+- Зависимости: `M27`
+- Commit required: yes
+- Commit message: `add sandbox rollback regression tests`
+- Scope: покрыть tests для save-failure rollback и startup observability failure paths
+- Файлы: `test/test_sandbox_usecase.py`, `test/test_docker_runtime.py`, `test/test_create_http.py`, при необходимости другие focused tests в `test/*`
+- Решение: добавить tests на fresh-create/replace save failure compensation, `list_active` failure observability и reconciliation failure span/metric expectations где применимо
+- Критерии приемки: rollback path покрыт; list/reconciliation failure observability не регрессирует; tests остаются presence-based и стабильными
+
+### M29. Финальный boundary review для sandbox observability
+
+- Субагент: `code-reviewer`
+- Статус: completed
+- Зависимости: `M28`
+- Commit required: no
+- Scope: подтвердить, что M27-M28 закрыли remaining M26 замечания
+- Файлы: весь измененный код после `M27`-`M28`
+- Критерии приемки: нет замечаний по rollback gap и startup failure observability coverage; sandbox observability slice приемлем as-is

tasks/roadmap.md

@@ -0,0 +1,279 @@
+# Roadmap: workspace/chat/files + artifacts
+
+## Цель
+
+Следующий продуктовый приоритет для `master-service`:
+1. управление `workspace` / `chat` / `chat files`
+2. хранение и выдача `artifacts`
+
+Идея этапа: превратить текущий sandbox MVP в storage-centric control plane, где sandbox работает поверх явных пользовательских данных, а не только поверх `chat_id`.
+
+## Что считаем в scope этого roadmap
+
+- workspace metadata
+- chat metadata и chat directories
+- `history.md` рядом с чатом
+- upload/list/download/delete/clear для chat files
+- quota / usage accounting
+- artifact metadata
+- object storage adapter для artifacts
+- delivery / mark-for-delete flow для artifacts
+- интеграция storage model с текущим sandbox lifecycle
+
+## Что пока вне scope
+
+- полноценный auth/access control
+- access lease и WebSocket handoff
+- multi-node orchestration как отдельный эпик
+- full retention engine для workspace/account lifecycle
+- внешние messenger integrations
+
+## Зафиксированные продуктовые решения
+
+- trusted caller передает `user_id: UUID`
+- в v1 один `Workspace` на одного `User`
+- metadata adapters пока делаем `in-memory`
+- chat history в v1 хранится в `history.md`
+- chat files отдаем через master: metadata API + download endpoint
+- soft quota 10 GB блокирует quota-relevant write-path операции, но не делает hard reservation
+- chat create сам по себе не блокируется quota check, пока не появляется значимый file/artifact payload
+- delete в v1 — hard delete; chat delete запрещен при active sandbox
+- artifact delivery подтверждается явным ack
+- artifact blob отдается через presigned URL, metadata — через master
+- object storage layout: один bucket на environment + prefixes
+- artifact states: `created`, `stored`, `delivered`, `delivery_failed`, `marked_for_delete`, `deleted`
+- artifact retention: после ack -> `marked_for_delete` -> async cleanup; без ack действует отдельный TTL
+
+## Принципы выполнения
+
+- Clean Architecture и dependency direction сохраняются
+- filesystem и object storage живут только во внешних adapter/repository слоях
+- `domain/` и `usecase/` не знают про FastAPI, Docker, OpenTelemetry, env
+- один delivery slice = одна атомарная задача = один commit
+- архитектурные развилки сначала фиксируются кратким ADR-lite в `docs/`
+
+## Легенда исполнителей
+
+- `primary-agent` — архитектура, ADR, domain/contracts, финальная сборка решений
+- `junior` — простые in-memory adapters и базовые unit-level задачи
+- `junior+opus` — usecase/adapter/API slices со средним уровнем связности
+- `test-engineer` — regression и integration test packs
+- `code-reviewer` — review-only этапы
+
+---
+
+## Phase 0. Design checkpoints
+
+### R00. ADR: storage source of truth
+- **Рекомендуемый исполнитель:** `primary-agent`
+- **Зачем:** определить, что является source of truth для `Workspace`, `Chat`, `ChatFile`, `Artifact`
+- **Нужно решить:** как устроен in-memory-first этап, как metadata связаны с filesystem paths, как sandbox получает path текущего chat
+- **Выход:** ADR-lite + список сущностей и ownership boundaries
+
+### R01. ADR: artifact lifecycle
+- **Рекомендуемый исполнитель:** `primary-agent`
+- **Зачем:** зафиксировать states и delivery contract для artifact flow
+- **Нужно решить:** `created -> stored -> delivered -> delivery_failed -> marked_for_delete -> deleted`
+- **Выход:** ADR-lite + минимальный state machine
+
+### R02. ADR: chat history policy
+- **Рекомендуемый исполнитель:** `primary-agent`
+- **Зачем:** зафиксировать, как `history.md` соотносится с metadata чата
+- **Нужно решить:** кто создает файл, кто пишет initial header, как читать/обновлять metadata без парсинга файла
+- **Выход:** ADR-lite + правила sync между metadata и history file
+
+---
+
+## Phase 1. Foundation for workspace/chat/file domain
+
+### R10. Domain model for user storage
+- **Рекомендуемый исполнитель:** `primary-agent`
+- **Scope:** `Workspace`, `Chat`, `ChatFile`, domain errors
+- **Layer:** `domain/`
+- **Выход:** минимальные сущности, value objects, ошибки конфликтов/не-найдено/квота
+- **Depends on:** `R00`, `R02`
+
+### R11. Usecase ports for storage and metadata
+- **Рекомендуемый исполнитель:** `primary-agent`
+- **Scope:** repository/storage interfaces для workspace/chat/chat-file/history/quota
+- **Layer:** `usecase/`
+- **Выход:** порты для metadata repo, file storage, usage reader, id generator, clock и trusted caller identity input
+- **Depends on:** `R10`
+
+### R12. In-memory metadata adapter foundation
+- **Рекомендуемый исполнитель:** `junior`
+- **Scope:** первая metadata implementation для workspace/chat/chat-file/artifact metadata
+- **Layer:** `repository/`
+- **Примечание:** это промежуточный этап; durable DB идет отдельным follow-up после storage/product slice
+- **Выход:** in-memory CRUD для workspace/chat/chat-file/artifact metadata
+- **Depends on:** `R11`
+
+### R13. Filesystem storage adapter foundation
+- **Рекомендуемый исполнитель:** `junior+opus`
+- **Scope:** storage adapter для chat directories, `history.md`, uploads и file metadata extraction
+- **Layer:** `adapter/` или `repository/` в зависимости от финальной границы
+- **Выход:** операции create/list/delete/read metadata для chat files, history path management и download path resolution
+- **Depends on:** `R11`
+
+### R14. Sandbox integration with chat metadata
+- **Рекомендуемый исполнитель:** `junior+opus`
+- **Scope:** перестать считать `chat_id` единственным источником layout, использовать metadata-backed chat path
+- **Layer:** `usecase/` + outer adapters
+- **Выход:** sandbox mounts current chat directory, созданную через storage model
+- **Depends on:** `R12`, `R13`
+
+---
+
+## Phase 2. Workspace and chat API
+
+### R20. Create workspace on first touch
+- **Рекомендуемый исполнитель:** `junior+opus`
+- **Scope:** auto-create workspace при первом запросе на user storage
+- **API:** внутренний usecase + при необходимости явная HTTP ручка
+- **Выход:** гарантированный `Workspace` для пользователя по `user_id`
+- **Depends on:** `R12`
+
+### R21. Chat CRUD
+- **Рекомендуемый исполнитель:** `junior+opus`
+- **Scope:** create/get/list/delete chat
+- **API:** versioned HTTP under `/api/v1`
+- **Выход:** first-class chat metadata вместо implicit-only `chat_id`
+- **Depends on:** `R20`, `R12`, `R13`
+
+### R22. History file lifecycle
+- **Рекомендуемый исполнитель:** `junior+opus`
+- **Scope:** создавать `history.md` вместе с chat directory, читать и обновлять metadata
+- **Выход:** первый стабильный contract для истории чата в filesystem
+- **Depends on:** `R21`, `R13`
+
+### R23. Chat file upload and file metadata API
+- **Рекомендуемый исполнитель:** `junior+opus`
+- **Scope:** upload/list/get metadata/delete single file/clear chat files
+- **API:** HTTP adapter + schemas
+- **Выход:** базовый file management API на chat scope + master download endpoint
+- **Depends on:** `R21`, `R13`
+
+### R24. Usage and quota accounting
+- **Рекомендуемый исполнитель:** `junior+opus`
+- **Scope:** считать storage usage по workspace/chat, мягкая quota 10 GB
+- **Выход:** usecases и metrics для current usage; file/artifact write-path reject при превышении soft quota
+- **Depends on:** `R23`
+
+### R25. Tests and review for storage slice
+- **Рекомендуемый исполнитель:** `test-engineer`
+- **Scope:** unit + adapter + HTTP tests, boundary review
+- **Выход:** regression coverage для workspace/chat/files
+- **Depends on:** `R20`-`R24`
+
+---
+
+## Phase 3. Artifact pipeline
+
+### R30. Artifact domain and metadata model
+- **Рекомендуемый исполнитель:** `primary-agent`
+- **Scope:** `Artifact` entity, states, delivery metadata, linkage to `user/chat`
+- **Layer:** `domain/`, `usecase/`
+- **Выход:** базовая artifact model + errors + repository ports для in-memory metadata stage
+- **Depends on:** `R01`, `R12`
+
+### R31. Object storage adapter
+- **Рекомендуемый исполнитель:** `junior+opus`
+- **Scope:** S3-compatible adapter for artifact blobs
+- **Layer:** outer `adapter/` or `repository/`
+- **Выход:** upload/get/delete primitives behind interface + presigned URL support
+- **Depends on:** `R30`
+
+### R32. Artifact registration and upload flow
+- **Рекомендуемый исполнитель:** `junior+opus`
+- **Scope:** usecase, который связывает metadata и object storage
+- **Выход:** artifact можно зарегистрировать, сохранить blob и получить external reference
+- **Depends on:** `R30`, `R31`
+
+### R33. Artifact list/get metadata API
+- **Рекомендуемый исполнитель:** `junior+opus`
+- **Scope:** list artifacts by chat/user, get artifact metadata/status
+- **Выход:** внешний API для управления artifact metadata и выдачи presigned download references
+- **Depends on:** `R32`
+
+### R34. Delivery acknowledgement flow
+- **Рекомендуемый исполнитель:** `junior+opus`
+- **Scope:** mark artifact as delivered / delivery_failed
+- **Выход:** подтвержденный delivery state через explicit ack и база для retention
+- **Depends on:** `R32`
+
+### R35. Artifact delete / mark-for-delete policy
+- **Рекомендуемый исполнитель:** `junior+opus`
+- **Scope:** policy-driven cleanup hooks после delivery confirmation или TTL
+- **Выход:** artifact lifecycle завершен, metadata и blob cleanup согласованы, `marked_for_delete` используется как переходное состояние
+- **Depends on:** `R34`
+
+### R36. Tests and review for artifact slice
+- **Рекомендуемый исполнитель:** `test-engineer`
+- **Scope:** unit + adapter + HTTP tests, boundary review
+- **Выход:** regression coverage для artifact lifecycle и object storage integration
+- **Depends on:** `R30`-`R35`
+
+---
+
+## Phase 4. Cross-slice hardening
+
+### R40. Observability for storage usage and artifacts
+- **Рекомендуемый исполнитель:** `junior+opus`
+- **Scope:** metrics по usage/quota, artifact upload latency, artifact errors
+- **Выход:** dashboards and alerts for storage-centric flows
+- **Depends on:** `R24`, `R32`, `R35`
+
+### R41. Cleanup safety rules
+- **Рекомендуемый исполнитель:** `junior+opus`
+- **Scope:** не удалять chat files/artifacts, если есть активный sandbox или незавершенный delivery flow
+- **Выход:** safe cleanup invariants
+- **Depends on:** `R24`, `R35`
+
+### R42. Docs refresh
+- **Рекомендуемый исполнитель:** `primary-agent`
+- **Scope:** README, ADRs, API docs, operator notes
+- **Выход:** актуальная документация для storage and artifact flows
+- **Depends on:** все предыдущие slices
+
+---
+
+## Рекомендуемый порядок выполнения
+
+### Priority A — workspace/chat/files
+1. `R00` `R02`
+2. `R10` `R11`
+3. `R12` `R13`
+4. `R20` `R21` `R22`
+5. `R23` `R24`
+6. `R14` `R25`
+
+### Priority B — artifacts
+1. `R01`
+2. `R30` `R31`
+3. `R32` `R33`
+4. `R34` `R35`
+5. `R36`
+
+### Future follow-up after Priority A+B
+1. durable metadata repository instead of in-memory adapters
+2. auth/access control and access lease
+3. multi-node storage/session coordination
+
+### Priority C — hardening
+1. `R40`
+2. `R41`
+3. `R42`
+
+---
+
+## Что даст этот roadmap
+
+После выполнения Priority A + B сервис сможет:
+- создавать и хранить workspace/chat metadata
+- управлять файлами пользователя в chat scope
+- хранить историю чата в предсказуемом layout
+- поднимать sandbox поверх first-class chat storage
+- хранить artifact metadata и blob отдельно
+- отдавать artifact metadata наружу и подтверждать доставку
+
+То есть `master-service` станет не только sandbox orchestrator, но и полноценным control-plane для user storage и artifact lifecycle.

tasks/sprint-01-storage-foundation.md

@@ -0,0 +1,130 @@
+# Sprint 01 — storage foundation
+
+## Цель спринта
+
+Подготовить первый исполнимый storage slice для `workspace` / `chat` / `chat files` без durable DB и без auth, но с корректными clean-architecture границами.
+
+## Зафиксированные допущения
+
+- trusted caller передает `user_id: UUID`
+- один `Workspace` на одного `User`
+- metadata adapters пока `in-memory`
+- история чата живет в `history.md`
+- chat files отдаются через master download endpoint
+- soft quota блокирует quota-relevant write-path операции
+- chat create сам по себе quota не блокирует
+- delete в v1 — hard delete; chat delete запрещен при active sandbox
+
+## Scope in
+
+- domain model для `Workspace`, `Chat`, `ChatFile`
+- usecase ports для metadata, filesystem storage и usage
+- in-memory metadata adapters
+- filesystem adapter для chat directories, `history.md` и file operations
+- usecases для auto-create workspace, chat CRUD и file CRUD
+- HTTP API для chat/file operations
+- базовый usage/quota check
+
+## Scope out
+
+- durable DB
+- auth/access control
+- artifacts
+- retention engine
+- p2p/access lease
+- multi-node behavior
+
+## Легенда исполнителей
+
+- `primary-agent` — архитектура, ADR, domain/contracts
+- `junior` — простые repositories и базовые unit tasks
+- `junior+opus` — adapter/usecase/API задачи средней сложности
+- `test-engineer` — test packs
+- `code-reviewer` — review-only этап
+
+## Порядок задач
+
+### S01. ADR-lite и storage contracts
+- **Рекомендуемый исполнитель:** `primary-agent`
+- **Commit:** `add storage foundation contracts`
+- **Scope:** зафиксировать storage source-of-truth и history policy; добавить domain entities и usecase ports
+- **Files:** `docs/009-storage-foundation.md`, `docs/010-chat-history-policy.md`, `domain/*`, `usecase/interface.py` или новые storage usecase files
+- **Acceptance:** есть минимальные сущности/ошибки/порты; ADR краткий и консистентный
+
+### S02. In-memory metadata adapters
+- **Рекомендуемый исполнитель:** `junior`
+- **Commit:** `add in-memory storage repositories`
+- **Scope:** `WorkspaceRepository`, `ChatRepository`, `ChatFileRepository` in-memory implementations
+- **Files:** `repository/*`
+- **Acceptance:** CRUD и базовые query paths работают без HTTP
+
+### S03. Filesystem chat storage adapter
+- **Рекомендуемый исполнитель:** `junior+opus`
+- **Commit:** `add chat filesystem storage`
+- **Scope:** create chat directory, create `history.md`, save/list/delete files, collect file metadata
+- **Files:** outer adapter/repository storage files
+- **Acceptance:** chat directory layout стабилен; adapter не протекает внутрь
+
+### S04. Workspace and chat usecases
+- **Рекомендуемый исполнитель:** `junior+opus`
+- **Commit:** `add workspace and chat usecases`
+- **Scope:** auto-create workspace on first touch, create/get/list/delete chat
+- **Files:** `usecase/*`, `adapter/di/container.py`
+- **Acceptance:** delete chat конфликтует при active sandbox; history path создается через storage adapter
+
+### S05. Chat file usecases + quota
+- **Рекомендуемый исполнитель:** `junior+opus`
+- **Commit:** `add chat file usecases`
+- **Scope:** upload/list/metadata/delete/clear files; current usage and soft quota reject on write
+- **Files:** `usecase/*`, `adapter/di/container.py`
+- **Acceptance:** file write-path reject при quota overflow; usage включает `history.md` и chat files; chat metadata create не блокируется quota check
+
+### S06. HTTP API for chat and files
+- **Рекомендуемый исполнитель:** `junior+opus`
+- **Commit:** `add chat and file http api`
+- **Scope:** versioned routes, schemas, error mapping, file download endpoint
+- **Files:** `adapter/http/fastapi/*`
+- **Acceptance:** API тонкий; FastAPI не протекает во внутренние слои
+
+### S07. Tests for storage foundation
+- **Рекомендуемый исполнитель:** `test-engineer`
+- **Commit:** `add storage foundation tests`
+- **Scope:** unit + adapter + HTTP tests для workspace/chat/files/quota
+- **Files:** `test/*`
+- **Acceptance:** есть regression coverage для CRUD, history creation, quota reject, delete conflict, file download
+
+### S08. Boundary review
+- **Рекомендуемый исполнитель:** `code-reviewer`
+- **Commit:** no
+- **Scope:** final review storage foundation slice
+- **Acceptance:** clean architecture соблюдена, нет must-fix замечаний
+
+## Definition of done
+
+Спринт считается завершенным, когда:
+- можно создать chat для `user_id`
+- для chat создается directory и `history.md`
+- можно загрузить и удалить chat files
+- можно получить metadata и скачать файл через master
+- quota check блокирует quota-relevant file write operations
+- все тесты проходят
+- boundary review не содержит must-fix замечаний
+
+## Риски
+
+- in-memory metadata не переживает restart
+- download endpoint может потребовать уточнения streaming contract
+- quota semantics нужно держать простыми, чтобы не породить скрытый retention scope
+
+## Что идет следующим спринтом
+
+Сразу после этого спринта нужен bridge-спринт для sandbox/storage integration:
+- metadata-backed chat path для sandbox flow
+- wiring текущего `CreateSandbox` поверх first-class chat metadata
+- review, что sandbox lifecycle не расходится с новым storage slice
+
+И только потом логично брать artifact slice:
+- artifact domain + metadata
+- object storage adapter
+- artifact upload/list/status
+- delivery ack flow

test/test_create_http.py

@@ -0,0 +1,909 @@
+import asyncio
+import json
+from datetime import UTC, datetime, timedelta
+from uuid import UUID
+
+import pytest
+from docker import DockerClient
+from fastapi import FastAPI
+from starlette.types import Message, Scope
+
+import adapter.di.container as container_module
+from adapter.config.model import (
+    AppConfig,
+    AppSectionConfig,
+    DockerConfig,
+    HttpConfig,
+    LoggingConfig,
+    MetricsConfig,
+    OtelConfig,
+    SandboxConfig,
+    SecurityConfig,
+    TracingConfig,
+)
+from adapter.di.container import AppContainer, AppRepositories, AppUsecases
+from adapter.docker.runtime import DockerSandboxRuntime
+from adapter.http.fastapi import app as app_module
+from adapter.observability.noop import NoopMetrics, NoopTracer
+from adapter.observability.runtime import ObservabilityRuntime
+from adapter.sandbox.reconciliation import SandboxSessionReconciler
+from domain.error import SandboxError, SandboxStartError
+from domain.sandbox import SandboxSession, SandboxStatus
+from repository.sandbox_lock import ProcessLocalSandboxLifecycleLocker
+from repository.sandbox_session import InMemorySandboxSessionRepository
+from usecase.interface import Attrs
+from usecase.sandbox import CleanupExpiredSandboxes, CreateSandbox, CreateSandboxCommand
+
+CHAT_ID = UUID('123e4567-e89b-12d3-a456-426614174000')
+NON_CANONICAL_CHAT_ID = '123E4567E89B12D3A456426614174000'
+SESSION_ID = UUID('00000000-0000-0000-0000-000000000011')
+
+
+class FakeLogger:
+    def __init__(self) -> None:
+        self.messages: list[tuple[str, str, Attrs | None]] = []
+
+    def debug(self, message: str, attrs: Attrs | None = None) -> None:
+        self.messages.append(('debug', message, attrs))
+
+    def info(self, message: str, attrs: Attrs | None = None) -> None:
+        self.messages.append(('info', message, attrs))
+
+    def warning(self, message: str, attrs: Attrs | None = None) -> None:
+        self.messages.append(('warning', message, attrs))
+
+    def error(self, message: str, attrs: Attrs | None = None) -> None:
+        self.messages.append(('error', message, attrs))
+
+
+class FakeCreateSandboxUsecase(CreateSandbox):
+    def __init__(
+        self, session: SandboxSession | None = None, error: Exception | None = None
+    ) -> None:
+        self._session = session
+        self._error = error
+        self.commands: list[CreateSandboxCommand] = []
+
+    def execute(self, command: CreateSandboxCommand) -> SandboxSession:
+        self.commands.append(command)
+        if self._error is not None:
+            raise self._error
+        if self._session is None:
+            raise AssertionError('missing session')
+        return self._session
+
+
+class FakeCleanupExpiredSandboxes(CleanupExpiredSandboxes):
+    def __init__(self) -> None:
+        self.calls = 0
+
+    def execute(self) -> list[SandboxSession]:
+        self.calls += 1
+        return []
+
+
+class FakeDockerClient(DockerClient):
+    def __init__(self, base_url: str | None = None) -> None:
+        self.base_url = base_url
+        self.close_calls = 0
+
+    def close(self) -> None:
+        self.close_calls += 1
+
+
+class EmptySandboxState:
+    def __init__(self) -> None:
+        self.calls = 0
+
+    def list_active_sessions(self) -> list[SandboxSession]:
+        self.calls += 1
+        return []
+
+
+class FakeClock:
+    def __init__(self, now: datetime) -> None:
+        self._now = now
+
+    def now(self) -> datetime:
+        return self._now
+
+
+class RecordingMetrics:
+    def __init__(self) -> None:
+        self.increment_calls: list[tuple[str, int, Attrs | None]] = []
+        self.record_calls: list[tuple[str, float, Attrs | None]] = []
+        self.set_calls: list[tuple[str, int | float, Attrs | None]] = []
+
+    def increment(
+        self,
+        name: str,
+        value: int = 1,
+        attrs: Attrs | None = None,
+    ) -> None:
+        self.increment_calls.append((name, value, attrs))
+
+    def record(
+        self,
+        name: str,
+        value: float,
+        attrs: Attrs | None = None,
+    ) -> None:
+        self.record_calls.append((name, value, attrs))
+
+    def set(
+        self,
+        name: str,
+        value: int | float,
+        attrs: Attrs | None = None,
+    ) -> None:
+        self.set_calls.append((name, value, attrs))
+
+
+class RecordingSpan:
+    def __init__(self) -> None:
+        self.attrs: dict[str, str | int | float | bool] = {}
+        self.errors: list[Exception] = []
+
+    def set_attribute(self, name: str, value: str | int | float | bool) -> None:
+        self.attrs[name] = value
+
+    def record_error(self, error: Exception) -> None:
+        self.errors.append(error)
+
+
+class RecordingSpanContext:
+    def __init__(self, span: RecordingSpan) -> None:
+        self._span = span
+
+    def __enter__(self) -> RecordingSpan:
+        return self._span
+
+    def __exit__(
+        self,
+        exc_type: type[BaseException] | None,
+        exc: BaseException | None,
+        traceback: object,
+    ) -> bool | None:
+        return None
+
+
+class RecordingTracer:
+    def __init__(self) -> None:
+        self.spans: list[tuple[str, Attrs | None, RecordingSpan]] = []
+
+    def start_span(
+        self,
+        name: str,
+        attrs: Attrs | None = None,
+    ) -> RecordingSpanContext:
+        span = RecordingSpan()
+        self.spans.append((name, attrs, span))
+        return RecordingSpanContext(span)
+
+
+class FakeLifecycleRuntime:
+    def __init__(self, sessions: list[SandboxSession]) -> None:
+        self._sessions = list(sessions)
+        self.list_calls = 0
+        self.create_calls: list[CreateSandboxCommand] = []
+        self.stop_calls: list[str] = []
+
+    def list_active_sessions(self) -> list[SandboxSession]:
+        self.list_calls += 1
+        return list(self._sessions)
+
+    def create(
+        self,
+        *,
+        session_id: UUID,
+        chat_id: UUID,
+        created_at: datetime,
+        expires_at: datetime,
+    ) -> SandboxSession:
+        self.create_calls.append(CreateSandboxCommand(chat_id=chat_id))
+        session = SandboxSession(
+            session_id=session_id,
+            chat_id=chat_id,
+            container_id=f'container-{session_id}',
+            status=SandboxStatus.RUNNING,
+            created_at=created_at,
+            expires_at=expires_at,
+        )
+        self._sessions = [
+            existing for existing in self._sessions if existing.chat_id != chat_id
+        ]
+        self._sessions.append(session)
+        return session
+
+    def stop(self, container_id: str) -> None:
+        self.stop_calls.append(container_id)
+
+
+class FixedSandboxState:
+    def __init__(self, sessions: list[SandboxSession]) -> None:
+        self._sessions = list(sessions)
+
+    def list_active_sessions(self) -> list[SandboxSession]:
+        return list(self._sessions)
+
+
+class FailingSandboxState:
+    def __init__(self, error: Exception) -> None:
+        self._error = error
+        self.calls = 0
+
+    def list_active_sessions(self) -> list[SandboxSession]:
+        self.calls += 1
+        raise self._error
+
+
+class CountingRegistry:
+    def __init__(self, count_active_result: int) -> None:
+        self._count_active_result = count_active_result
+        self.replaced_sessions: list[SandboxSession] = []
+
+    def replace_all(self, sessions: list[SandboxSession]) -> None:
+        self.replaced_sessions = list(sessions)
+
+    def count_active(self) -> int:
+        return self._count_active_result
+
+
+class FailingRegistry:
+    def __init__(self, error: Exception, *, fail_on: str = 'replace_all') -> None:
+        self._error = error
+        self._fail_on = fail_on
+        self.replaced_sessions: list[SandboxSession] = []
+        self.count_calls = 0
+
+    def replace_all(self, sessions: list[SandboxSession]) -> None:
+        self.replaced_sessions = list(sessions)
+        if self._fail_on == 'replace_all':
+            raise self._error
+
+    def count_active(self) -> int:
+        self.count_calls += 1
+        if self._fail_on == 'count_active':
+            raise self._error
+        return 0
+
+
+def build_config() -> AppConfig:
+    return AppConfig(
+        app=AppSectionConfig(name='master', env='test'),
+        http=HttpConfig(host='127.0.0.1', port=8000),
+        logging=LoggingConfig(
+            level='INFO', output='stdout', format='json', file_path=None
+        ),
+        metrics=MetricsConfig(enabled=False),
+        tracing=TracingConfig(enabled=False),
+        otel=OtelConfig(
+            service_name='master',
+            logs_endpoint='http://localhost:4318/v1/logs',
+            metrics_endpoint='http://localhost:4318/v1/metrics',
+            traces_endpoint='http://localhost:4318/v1/traces',
+            metric_export_interval=1000,
+        ),
+        docker=DockerConfig(base_url='unix:///var/run/docker.sock'),
+        sandbox=SandboxConfig(
+            image='sandbox:latest',
+            ttl_seconds=300,
+            cleanup_interval_seconds=60,
+            chats_root='/tmp/chats',
+            dependencies_host_path='/tmp/dependencies',
+            lambda_tools_host_path='/tmp/lambda-tools',
+            chat_mount_path='/workspace/chat',
+            dependencies_mount_path='/workspace/dependencies',
+            lambda_tools_mount_path='/workspace/lambda-tools',
+        ),
+        security=SecurityConfig(
+            token_header='Authorization',
+            api_token='token',
+            signing_key='signing-key',
+        ),
+    )
+
+
+def build_container(
+    config: AppConfig,
+    create_sandbox_usecase: CreateSandbox,
+    cleanup_usecase: CleanupExpiredSandboxes,
+    logger: FakeLogger,
+    docker_client: FakeDockerClient,
+    sandbox_reconciler: SandboxSessionReconciler | None = None,
+) -> AppContainer:
+    observability = ObservabilityRuntime(
+        logger=logger,
+        metrics=NoopMetrics(),
+        tracer=NoopTracer(),
+    )
+    repositories = AppRepositories(sandbox_session=InMemorySandboxSessionRepository())
+    reconciler = sandbox_reconciler
+    if reconciler is None:
+        reconciler = SandboxSessionReconciler(
+            state_source=EmptySandboxState(),
+            registry=repositories.sandbox_session,
+            logger=logger,
+            metrics=observability.metrics,
+            tracer=observability.tracer,
+        )
+    usecases = AppUsecases(
+        create_sandbox=create_sandbox_usecase,
+        cleanup_expired_sandboxes=cleanup_usecase,
+    )
+    return AppContainer(
+        config=config,
+        observability=observability,
+        repositories=repositories,
+        usecases=usecases,
+        sandbox_reconciler=reconciler,
+        _docker_client=docker_client,
+    )
+
+
+async def request_json(
+    app: FastAPI,
+    method: str,
+    path: str,
+    payload: dict[str, str] | None = None,
+) -> tuple[int, dict[str, object]]:
+    body = b'' if payload is None else json.dumps(payload).encode()
+    messages: list[Message] = []
+    request_sent = False
+
+    async def receive() -> Message:
+        nonlocal request_sent
+        if request_sent:
+            await asyncio.sleep(0)
+            return {'type': 'http.disconnect'}
+
+        request_sent = True
+        return {
+            'type': 'http.request',
+            'body': body,
+            'more_body': False,
+        }
+
+    async def send(message: Message) -> None:
+        messages.append(message)
+
+    scope: Scope = {
+        'type': 'http',
+        'asgi': {'version': '3.0'},
+        'http_version': '1.1',
+        'method': method,
+        'scheme': 'http',
+        'path': path,
+        'raw_path': path.encode(),
+        'query_string': b'',
+        'root_path': '',
+        'headers': _build_headers(body, payload is not None),
+        'client': ('testclient', 50000),
+        'server': ('testserver', 80),
+        'state': {},
+    }
+
+    await app(scope, receive, send)
+
+    status = 500
+    response_body = b''
+    for message in messages:
+        if message['type'] == 'http.response.start':
+            status = int(message['status'])
+        if message['type'] == 'http.response.body':
+            response_body += bytes(message.get('body', b''))
+
+    if not response_body:
+        return status, {}
+
+    return status, json.loads(response_body.decode())
+
+
+def _build_headers(body: bytes, has_json_body: bool) -> list[tuple[bytes, bytes]]:
+    headers = [
+        (b'host', b'testserver'),
+        (b'content-length', str(len(body)).encode()),
+    ]
+    if has_json_body:
+        headers.append((b'content-type', b'application/json'))
+    return headers
+
+
+async def post_json(
+    app: FastAPI, path: str, payload: dict[str, str]
+) -> tuple[int, dict[str, object]]:
+    return await request_json(app, 'POST', path, payload)
+
+
+async def get_json(app: FastAPI, path: str) -> tuple[int, dict[str, object]]:
+    return await request_json(app, 'GET', path)
+
+
+async def exercise_create_request(
+    app: FastAPI,
+    payload: dict[str, str],
+) -> tuple[int, dict[str, object]]:
+    await app.router.startup()
+    try:
+        status, response = await post_json(app, '/api/v1/create', payload)
+        await asyncio.sleep(0)
+        return status, response
+    finally:
+        await app.router.shutdown()
+
+
+async def exercise_get_request(
+    app: FastAPI,
+    path: str,
+) -> tuple[int, dict[str, object]]:
+    await app.router.startup()
+    try:
+        status, response = await get_json(app, path)
+        await asyncio.sleep(0)
+        return status, response
+    finally:
+        await app.router.shutdown()
+
+
+def test_post_create_returns_session_with_canonical_chat_id(monkeypatch) -> None:
+    config = build_config()
+    expires_at = datetime(2026, 4, 2, 12, 5, tzinfo=UTC)
+    session = SandboxSession(
+        session_id=SESSION_ID,
+        chat_id=CHAT_ID,
+        container_id='container-123',
+        status=SandboxStatus.RUNNING,
+        created_at=expires_at - timedelta(minutes=5),
+        expires_at=expires_at,
+    )
+    logger = FakeLogger()
+    create_usecase = FakeCreateSandboxUsecase(session=session)
+    cleanup_usecase = FakeCleanupExpiredSandboxes()
+    docker_client = FakeDockerClient()
+    container = build_container(
+        config,
+        create_usecase,
+        cleanup_usecase,
+        logger,
+        docker_client,
+    )
+    monkeypatch.setattr(app_module, 'build_container', lambda **kwargs: container)
+    monkeypatch.setattr(
+        app_module.FastAPIInstrumentor, 'instrument_app', lambda *args, **kwargs: None
+    )
+
+    app = app_module.create_app(config=config)
+
+    status_code, response = asyncio.run(
+        exercise_create_request(app, {'chat_id': NON_CANONICAL_CHAT_ID})
+    )
+
+    assert status_code == 200
+    assert response == {
+        'session_id': str(SESSION_ID),
+        'chat_id': str(CHAT_ID),
+        'container_id': 'container-123',
+        'status': 'running',
+        'expires_at': '2026-04-02T12:05:00Z',
+    }
+    assert len(create_usecase.commands) == 1
+    assert create_usecase.commands[0].chat_id == CHAT_ID
+    assert cleanup_usecase.calls >= 1
+    assert any(
+        message == 'http_request'
+        and attrs is not None
+        and attrs['http.path'] == '/api/v1/create'
+        for _, message, attrs in logger.messages
+    )
+    assert docker_client.close_calls == 1
+
+
+def test_post_create_rejects_non_uuid_chat_id(monkeypatch) -> None:
+    config = build_config()
+    expires_at = datetime(2026, 4, 2, 12, 5, tzinfo=UTC)
+    session = SandboxSession(
+        session_id=SESSION_ID,
+        chat_id=CHAT_ID,
+        container_id='container-123',
+        status=SandboxStatus.RUNNING,
+        created_at=expires_at - timedelta(minutes=5),
+        expires_at=expires_at,
+    )
+    logger = FakeLogger()
+    create_usecase = FakeCreateSandboxUsecase(session=session)
+    cleanup_usecase = FakeCleanupExpiredSandboxes()
+    docker_client = FakeDockerClient()
+    container = build_container(
+        config,
+        create_usecase,
+        cleanup_usecase,
+        logger,
+        docker_client,
+    )
+    monkeypatch.setattr(app_module, 'build_container', lambda **kwargs: container)
+    monkeypatch.setattr(
+        app_module.FastAPIInstrumentor, 'instrument_app', lambda *args, **kwargs: None
+    )
+
+    app = app_module.create_app(config=config)
+
+    status_code, response = asyncio.run(
+        exercise_create_request(app, {'chat_id': 'x/../y'})
+    )
+
+    assert status_code == 422
+    assert 'detail' in response
+    assert create_usecase.commands == []
+    assert docker_client.close_calls == 1
+
+
+def test_post_create_maps_start_errors_to_service_unavailable(monkeypatch) -> None:
+    config = build_config()
+    logger = FakeLogger()
+    create_usecase = FakeCreateSandboxUsecase(error=SandboxStartError(str(CHAT_ID)))
+    cleanup_usecase = FakeCleanupExpiredSandboxes()
+    docker_client = FakeDockerClient()
+    container = build_container(
+        config,
+        create_usecase,
+        cleanup_usecase,
+        logger,
+        docker_client,
+    )
+    monkeypatch.setattr(app_module, 'build_container', lambda **kwargs: container)
+    monkeypatch.setattr(
+        app_module.FastAPIInstrumentor, 'instrument_app', lambda *args, **kwargs: None
+    )
+
+    app = app_module.create_app(config=config)
+
+    status_code, response = asyncio.run(
+        exercise_create_request(app, {'chat_id': str(CHAT_ID)})
+    )
+
+    assert status_code == 503
+    assert response == {'detail': 'sandbox_start_failed'}
+    assert docker_client.close_calls == 1
+
+
+def test_post_create_maps_generic_sandbox_errors_to_internal_error(monkeypatch) -> None:
+    config = build_config()
+    logger = FakeLogger()
+    create_usecase = FakeCreateSandboxUsecase(error=SandboxError('sandbox_broken'))
+    cleanup_usecase = FakeCleanupExpiredSandboxes()
+    docker_client = FakeDockerClient()
+    container = build_container(
+        config,
+        create_usecase,
+        cleanup_usecase,
+        logger,
+        docker_client,
+    )
+    monkeypatch.setattr(app_module, 'build_container', lambda **kwargs: container)
+    monkeypatch.setattr(
+        app_module.FastAPIInstrumentor, 'instrument_app', lambda *args, **kwargs: None
+    )
+
+    app = app_module.create_app(config=config)
+
+    status_code, response = asyncio.run(
+        exercise_create_request(app, {'chat_id': str(CHAT_ID)})
+    )
+
+    assert status_code == 500
+    assert response == {'detail': 'sandbox_broken'}
+    assert docker_client.close_calls == 1
+
+
+def test_startup_reconciliation_reuses_existing_container_after_restart(
+    monkeypatch,
+) -> None:
+    config = build_config()
+    created_at = datetime(2026, 4, 2, 12, 0, tzinfo=UTC)
+    restored_session = SandboxSession(
+        session_id=SESSION_ID,
+        chat_id=CHAT_ID,
+        container_id='container-123',
+        status=SandboxStatus.RUNNING,
+        created_at=created_at,
+        expires_at=created_at + timedelta(minutes=5),
+    )
+    logger = FakeLogger()
+    docker_client = FakeDockerClient()
+    runtime = FakeLifecycleRuntime([restored_session])
+    repository = InMemorySandboxSessionRepository()
+    observability = ObservabilityRuntime(
+        logger=logger,
+        metrics=NoopMetrics(),
+        tracer=NoopTracer(),
+    )
+    repositories = AppRepositories(sandbox_session=repository)
+    reconciler = SandboxSessionReconciler(
+        state_source=runtime,
+        registry=repository,
+        logger=logger,
+        metrics=observability.metrics,
+        tracer=observability.tracer,
+    )
+    usecases = AppUsecases(
+        create_sandbox=CreateSandbox(
+            repository=repository,
+            locker=ProcessLocalSandboxLifecycleLocker(),
+            runtime=runtime,
+            clock=FakeClock(created_at),
+            logger=logger,
+            metrics=NoopMetrics(),
+            tracer=NoopTracer(),
+            ttl=timedelta(minutes=5),
+        ),
+        cleanup_expired_sandboxes=CleanupExpiredSandboxes(
+            repository=repository,
+            locker=ProcessLocalSandboxLifecycleLocker(),
+            runtime=runtime,
+            clock=FakeClock(created_at),
+            logger=logger,
+            metrics=NoopMetrics(),
+            tracer=NoopTracer(),
+        ),
+    )
+    container = AppContainer(
+        config=config,
+        observability=observability,
+        repositories=repositories,
+        usecases=usecases,
+        sandbox_reconciler=reconciler,
+        _docker_client=docker_client,
+    )
+    monkeypatch.setattr(app_module, 'build_container', lambda **kwargs: container)
+    monkeypatch.setattr(
+        app_module.FastAPIInstrumentor, 'instrument_app', lambda *args, **kwargs: None
+    )
+
+    app = app_module.create_app(config=config)
+
+    status_code, response = asyncio.run(
+        exercise_create_request(app, {'chat_id': str(CHAT_ID)})
+    )
+
+    assert status_code == 200
+    assert response == {
+        'session_id': str(SESSION_ID),
+        'chat_id': str(CHAT_ID),
+        'container_id': 'container-123',
+        'status': 'running',
+        'expires_at': '2026-04-02T12:05:00Z',
+    }
+    assert runtime.list_calls == 1
+    assert runtime.create_calls == []
+    assert runtime.stop_calls == []
+    assert repository.get_active_by_chat_id(CHAT_ID) == restored_session
+    assert docker_client.close_calls == 1
+
+
+def test_removed_user_endpoint_returns_not_found(monkeypatch) -> None:
+    config = build_config()
+    expires_at = datetime(2026, 4, 2, 12, 5, tzinfo=UTC)
+    session = SandboxSession(
+        session_id=SESSION_ID,
+        chat_id=CHAT_ID,
+        container_id='container-123',
+        status=SandboxStatus.RUNNING,
+        created_at=expires_at - timedelta(minutes=5),
+        expires_at=expires_at,
+    )
+    logger = FakeLogger()
+    create_usecase = FakeCreateSandboxUsecase(session=session)
+    cleanup_usecase = FakeCleanupExpiredSandboxes()
+    docker_client = FakeDockerClient()
+    container = build_container(
+        config,
+        create_usecase,
+        cleanup_usecase,
+        logger,
+        docker_client,
+    )
+    monkeypatch.setattr(app_module, 'build_container', lambda **kwargs: container)
+    monkeypatch.setattr(
+        app_module.FastAPIInstrumentor, 'instrument_app', lambda *args, **kwargs: None
+    )
+
+    app = app_module.create_app(config=config)
+
+    status_code, response = asyncio.run(
+        exercise_get_request(app, '/api/v1/users/user-123')
+    )
+
+    assert status_code == 404
+    assert response == {'detail': 'Not Found'}
+    assert docker_client.close_calls == 1
+
+
+def test_reconciliation_uses_registry_backed_active_count_metric() -> None:
+    logger = FakeLogger()
+    metrics = RecordingMetrics()
+    tracer = RecordingTracer()
+    created_at = datetime(2026, 4, 2, 12, 0, tzinfo=UTC)
+    session = SandboxSession(
+        session_id=SESSION_ID,
+        chat_id=CHAT_ID,
+        container_id='container-123',
+        status=SandboxStatus.RUNNING,
+        created_at=created_at,
+        expires_at=created_at + timedelta(minutes=5),
+    )
+    registry = CountingRegistry(count_active_result=7)
+    reconciler = SandboxSessionReconciler(
+        state_source=FixedSandboxState([session]),
+        registry=registry,
+        logger=logger,
+        metrics=metrics,
+        tracer=tracer,
+    )
+
+    sessions = reconciler.execute()
+
+    assert sessions == [session]
+    assert registry.replaced_sessions == [session]
+    assert metrics.set_calls == [('sandbox.active.count', 7, None)]
+    assert tracer.spans[0][0] == 'adapter.sandbox.reconcile_sessions'
+    assert tracer.spans[0][2].attrs['sandbox.active_count'] == 7
+
+
+def test_reconciliation_records_error_when_state_source_fails() -> None:
+    logger = FakeLogger()
+    metrics = RecordingMetrics()
+    tracer = RecordingTracer()
+    state_error = RuntimeError('state_failed')
+    state_source = FailingSandboxState(state_error)
+    reconciler = SandboxSessionReconciler(
+        state_source=state_source,
+        registry=CountingRegistry(count_active_result=7),
+        logger=logger,
+        metrics=metrics,
+        tracer=tracer,
+    )
+
+    with pytest.raises(RuntimeError, match='state_failed') as excinfo:
+        reconciler.execute()
+
+    assert state_source.calls == 1
+    assert metrics.set_calls == []
+    spans = [
+        span
+        for name, _, span in tracer.spans
+        if name == 'adapter.sandbox.reconcile_sessions'
+    ]
+    assert spans
+    span = spans[0]
+    assert span.attrs['sandbox.result'] == 'error'
+    assert 'sandbox.discovered_count' not in span.attrs
+    assert 'sandbox.active_count' not in span.attrs
+    assert excinfo.value in span.errors
+
+
+def test_reconciliation_records_error_without_active_count_metric_on_registry_failure() -> (
+    None
+):
+    logger = FakeLogger()
+    metrics = RecordingMetrics()
+    tracer = RecordingTracer()
+    created_at = datetime(2026, 4, 2, 12, 0, tzinfo=UTC)
+    session = SandboxSession(
+        session_id=SESSION_ID,
+        chat_id=CHAT_ID,
+        container_id='container-123',
+        status=SandboxStatus.RUNNING,
+        created_at=created_at,
+        expires_at=created_at + timedelta(minutes=5),
+    )
+    registry_error = RuntimeError('replace_failed')
+    registry = FailingRegistry(registry_error)
+    reconciler = SandboxSessionReconciler(
+        state_source=FixedSandboxState([session]),
+        registry=registry,
+        logger=logger,
+        metrics=metrics,
+        tracer=tracer,
+    )
+
+    with pytest.raises(RuntimeError, match='replace_failed') as excinfo:
+        reconciler.execute()
+
+    assert registry.replaced_sessions == [session]
+    assert registry.count_calls == 0
+    assert metrics.set_calls == []
+    spans = [
+        span
+        for name, _, span in tracer.spans
+        if name == 'adapter.sandbox.reconcile_sessions'
+    ]
+    assert spans
+    span = spans[0]
+    assert span.attrs['sandbox.discovered_count'] == 1
+    assert span.attrs['sandbox.result'] == 'error'
+    assert 'sandbox.active_count' not in span.attrs
+    assert excinfo.value in span.errors
+
+
+def test_reconciliation_records_error_when_registry_count_active_fails() -> None:
+    logger = FakeLogger()
+    metrics = RecordingMetrics()
+    tracer = RecordingTracer()
+    created_at = datetime(2026, 4, 2, 12, 0, tzinfo=UTC)
+    session = SandboxSession(
+        session_id=SESSION_ID,
+        chat_id=CHAT_ID,
+        container_id='container-123',
+        status=SandboxStatus.RUNNING,
+        created_at=created_at,
+        expires_at=created_at + timedelta(minutes=5),
+    )
+    registry_error = RuntimeError('count_failed')
+    registry = FailingRegistry(registry_error, fail_on='count_active')
+    reconciler = SandboxSessionReconciler(
+        state_source=FixedSandboxState([session]),
+        registry=registry,
+        logger=logger,
+        metrics=metrics,
+        tracer=tracer,
+    )
+
+    with pytest.raises(RuntimeError, match='count_failed') as excinfo:
+        reconciler.execute()
+
+    assert registry.replaced_sessions == [session]
+    assert registry.count_calls == 1
+    assert metrics.set_calls == []
+    spans = [
+        span
+        for name, _, span in tracer.spans
+        if name == 'adapter.sandbox.reconcile_sessions'
+    ]
+    assert spans
+    span = spans[0]
+    assert span.attrs['sandbox.discovered_count'] == 1
+    assert 'sandbox.active_count' not in span.attrs
+    assert span.attrs['sandbox.result'] == 'error'
+    assert excinfo.value in span.errors
+
+
+def test_build_container_wires_observability_into_runtime_and_reconciler(
+    monkeypatch,
+) -> None:
+    logger = FakeLogger()
+    metrics = RecordingMetrics()
+    tracer = RecordingTracer()
+    observability = ObservabilityRuntime(
+        logger=logger,
+        metrics=metrics,
+        tracer=tracer,
+    )
+    docker_client = FakeDockerClient()
+    monkeypatch.setattr(
+        container_module, 'build_observability', lambda config: observability
+    )
+    monkeypatch.setattr(
+        container_module.docker,
+        'DockerClient',
+        lambda base_url: docker_client,
+    )
+
+    container = container_module.build_container(config=build_config())
+
+    runtime = container.sandbox_reconciler.state_source
+    assert isinstance(runtime, DockerSandboxRuntime)
+    assert runtime._metrics is metrics
+    assert runtime._tracer is tracer
+    assert container.sandbox_reconciler.metrics is metrics
+    assert container.sandbox_reconciler.tracer is tracer
+    assert container.usecases.create_sandbox._runtime is runtime
+    assert container.usecases.create_sandbox._metrics is metrics
+    assert container.usecases.create_sandbox._tracer is tracer
+    assert container.usecases.cleanup_expired_sandboxes._runtime is runtime
+    assert container.usecases.cleanup_expired_sandboxes._metrics is metrics
+    assert container.usecases.cleanup_expired_sandboxes._tracer is tracer
+    assert container._docker_client is docker_client
+
+    container.shutdown()
+
+    assert docker_client.close_calls == 1

test/test_docker_runtime.py

@@ -0,0 +1,732 @@
+from datetime import UTC, datetime, timedelta
+from pathlib import Path
+from types import TracebackType
+from typing import Any, TypedDict
+from uuid import UUID
+
+import pytest
+from docker import DockerClient
+from docker.errors import DockerException, NotFound
+from docker.types import Mount
+
+from adapter.config.model import SandboxConfig
+from adapter.docker.runtime import DockerSandboxRuntime
+from adapter.observability.noop import NoopMetrics, NoopTracer
+from domain.error import SandboxError, SandboxStartError
+from domain.sandbox import SandboxSession, SandboxStatus
+from usecase.interface import Attrs, AttrValue
+
+CHAT_ID = UUID('123e4567-e89b-12d3-a456-426614174000')
+NON_CANONICAL_CHAT_ID = '123E4567E89B12D3A456426614174000'
+SESSION_ID = UUID('00000000-0000-0000-0000-000000000010')
+
+
+class FakeContainer:
+    def __init__(self, container_id: str) -> None:
+        self.id = container_id
+        self.stop_calls = 0
+
+    def stop(self) -> None:
+        self.stop_calls += 1
+
+
+class FakeListedContainer(FakeContainer):
+    def __init__(
+        self,
+        container_id: str,
+        *,
+        labels: dict[str, str],
+        created_at: str,
+    ) -> None:
+        super().__init__(container_id)
+        self.labels = labels
+        self.attrs = {'Created': created_at}
+
+
+class FailingStopContainer(FakeListedContainer):
+    def __init__(
+        self,
+        container_id: str,
+        *,
+        labels: dict[str, str],
+        created_at: str,
+        error: Exception,
+    ) -> None:
+        super().__init__(
+            container_id,
+            labels=labels,
+            created_at=created_at,
+        )
+        self._error = error
+
+    def stop(self) -> None:
+        self.stop_calls += 1
+        raise self._error
+
+
+class RunKwargs(TypedDict):
+    detach: bool
+    labels: dict[str, str]
+    mounts: list[Mount]
+
+
+class RunCall(TypedDict):
+    args: tuple[str]
+    kwargs: RunKwargs
+
+
+class FakeContainers:
+    def __init__(self, run_result: FakeContainer | None = None) -> None:
+        self.run_calls: list[RunCall] = []
+        self.get_calls: list[str] = []
+        self.list_calls: list[dict[str, object]] = []
+        self.run_result = run_result or FakeContainer('container-123')
+        self.get_result: FakeContainer | Exception | None = None
+        self.list_result: list[object] = []
+        self.list_error: Exception | None = None
+
+    def run(
+        self,
+        image: str,
+        *,
+        detach: bool,
+        labels: dict[str, str],
+        mounts: list[Mount],
+    ) -> FakeContainer:
+        self.run_calls.append(
+            {
+                'args': (image,),
+                'kwargs': {
+                    'detach': detach,
+                    'labels': labels,
+                    'mounts': mounts,
+                },
+            }
+        )
+        return self.run_result
+
+    def get(self, container_id: str) -> FakeContainer:
+        self.get_calls.append(container_id)
+        if isinstance(self.get_result, Exception):
+            raise self.get_result
+        if self.get_result is None:
+            raise AssertionError('missing get result')
+        return self.get_result
+
+    def list(self, *, filters: dict[str, list[str]]) -> list[object]:
+        self.list_calls.append({'filters': filters})
+        if self.list_error is not None:
+            raise self.list_error
+        return self.list_result
+
+
+class FakeDockerClient(DockerClient):
+    def __init__(self, containers: FakeContainers) -> None:
+        self._containers = containers
+
+    @property
+    def containers(self) -> Any:
+        return self._containers
+
+
+class RecordingMetrics:
+    def __init__(self) -> None:
+        self.increment_calls: list[tuple[str, int, Attrs | None]] = []
+        self.record_calls: list[tuple[str, float, Attrs | None]] = []
+        self.set_calls: list[tuple[str, int | float, Attrs | None]] = []
+
+    def increment(
+        self,
+        name: str,
+        value: int = 1,
+        attrs: Attrs | None = None,
+    ) -> None:
+        self.increment_calls.append((name, value, attrs))
+
+    def record(
+        self,
+        name: str,
+        value: float,
+        attrs: Attrs | None = None,
+    ) -> None:
+        self.record_calls.append((name, value, attrs))
+
+    def set(
+        self,
+        name: str,
+        value: int | float,
+        attrs: Attrs | None = None,
+    ) -> None:
+        self.set_calls.append((name, value, attrs))
+
+
+class RecordingSpan:
+    def __init__(self) -> None:
+        self.attrs: dict[str, AttrValue] = {}
+        self.errors: list[Exception] = []
+
+    def set_attribute(self, name: str, value: AttrValue) -> None:
+        self.attrs[name] = value
+
+    def record_error(self, error: Exception) -> None:
+        self.errors.append(error)
+
+
+class RecordingSpanContext:
+    def __init__(self, span: RecordingSpan) -> None:
+        self._span = span
+
+    def __enter__(self) -> RecordingSpan:
+        return self._span
+
+    def __exit__(
+        self,
+        exc_type: type[BaseException] | None,
+        exc: BaseException | None,
+        traceback: TracebackType | None,
+    ) -> bool | None:
+        return None
+
+
+class RecordingTracer:
+    def __init__(self) -> None:
+        self.spans: list[tuple[str, Attrs | None, RecordingSpan]] = []
+
+    def start_span(
+        self,
+        name: str,
+        attrs: Attrs | None = None,
+    ) -> RecordingSpanContext:
+        span = RecordingSpan()
+        self.spans.append((name, attrs, span))
+        return RecordingSpanContext(span)
+
+
+def _attrs_include(
+    actual: Attrs | dict[str, AttrValue] | None,
+    expected: dict[str, AttrValue],
+) -> bool:
+    if actual is None:
+        return False
+
+    return all(actual.get(name) == value for name, value in expected.items())
+
+
+def _find_span(
+    tracer: RecordingTracer,
+    name: str,
+    attrs: dict[str, AttrValue] | None = None,
+    span_attrs: dict[str, AttrValue] | None = None,
+) -> RecordingSpan:
+    for recorded_name, recorded_attrs, span in tracer.spans:
+        if recorded_name != name:
+            continue
+        if attrs is not None and not _attrs_include(recorded_attrs, attrs):
+            continue
+        if span_attrs is not None and not _attrs_include(span.attrs, span_attrs):
+            continue
+        return span
+
+    raise AssertionError(f'missing span {name}')
+
+
+def _find_increment_call(
+    metrics: RecordingMetrics,
+    name: str,
+    *,
+    value: int = 1,
+    attrs: dict[str, AttrValue] | None = None,
+) -> tuple[str, int, Attrs | None]:
+    for recorded_name, recorded_value, recorded_attrs in metrics.increment_calls:
+        if recorded_name != name or recorded_value != value:
+            continue
+        if attrs is not None and not _attrs_include(recorded_attrs, attrs):
+            continue
+        return recorded_name, recorded_value, recorded_attrs
+
+    raise AssertionError(f'missing increment metric {name}')
+
+
+def _find_record_call(
+    metrics: RecordingMetrics,
+    name: str,
+    *,
+    attrs: dict[str, AttrValue] | None = None,
+) -> tuple[str, float, Attrs | None]:
+    for recorded_name, recorded_value, recorded_attrs in metrics.record_calls:
+        if recorded_name != name:
+            continue
+        if attrs is not None and not _attrs_include(recorded_attrs, attrs):
+            continue
+        return recorded_name, recorded_value, recorded_attrs
+
+    raise AssertionError(f'missing record metric {name}')
+
+
+def build_config(tmp_path: Path) -> SandboxConfig:
+    return SandboxConfig(
+        image='sandbox:latest',
+        ttl_seconds=300,
+        cleanup_interval_seconds=60,
+        chats_root=str(tmp_path / 'chats'),
+        dependencies_host_path=str(tmp_path / 'dependencies'),
+        lambda_tools_host_path=str(tmp_path / 'lambda-tools'),
+        chat_mount_path='/workspace/chat',
+        dependencies_mount_path='/workspace/dependencies',
+        lambda_tools_mount_path='/workspace/lambda-tools',
+    )
+
+
+def build_runtime(
+    config: SandboxConfig,
+    containers: FakeContainers,
+) -> DockerSandboxRuntime:
+    return DockerSandboxRuntime(
+        config,
+        FakeDockerClient(containers),
+        NoopMetrics(),
+        NoopTracer(),
+    )
+
+
+def test_runtime_create_applies_mount_policy_and_labels_with_canonical_chat_id(
+    tmp_path: Path,
+) -> None:
+    config = build_config(tmp_path)
+    (tmp_path / 'dependencies').mkdir()
+    (tmp_path / 'lambda-tools').mkdir()
+    containers = FakeContainers()
+    runtime = build_runtime(config, containers)
+    created_at = datetime(2026, 4, 2, 12, 0, tzinfo=UTC)
+    expires_at = created_at + timedelta(minutes=5)
+
+    session = runtime.create(
+        session_id=SESSION_ID,
+        chat_id=UUID(NON_CANONICAL_CHAT_ID),
+        created_at=created_at,
+        expires_at=expires_at,
+    )
+
+    assert session.session_id == SESSION_ID
+    assert session.chat_id == CHAT_ID
+    assert session.container_id == 'container-123'
+    assert session.status is SandboxStatus.RUNNING
+    assert session.created_at == created_at
+    assert session.expires_at == expires_at
+    assert (tmp_path / 'chats' / str(CHAT_ID)).is_dir()
+
+    call = containers.run_calls[0]
+    assert call['args'] == ('sandbox:latest',)
+    assert call['kwargs']['detach'] is True
+    assert call['kwargs']['labels'] == {
+        'session_id': str(SESSION_ID),
+        'chat_id': str(CHAT_ID),
+        'expires_at': expires_at.isoformat(),
+    }
+
+    mounts = call['kwargs']['mounts']
+    assert [dict(mount) for mount in mounts] == [
+        {
+            'Target': '/workspace/chat',
+            'Source': str((tmp_path / 'chats' / str(CHAT_ID)).resolve(strict=False)),
+            'Type': 'bind',
+            'ReadOnly': False,
+        },
+        {
+            'Target': '/workspace/dependencies',
+            'Source': str((tmp_path / 'dependencies').resolve(strict=False)),
+            'Type': 'bind',
+            'ReadOnly': True,
+        },
+        {
+            'Target': '/workspace/lambda-tools',
+            'Source': str((tmp_path / 'lambda-tools').resolve(strict=False)),
+            'Type': 'bind',
+            'ReadOnly': True,
+        },
+    ]
+
+
+def test_runtime_create_records_observability(tmp_path: Path) -> None:
+    config = build_config(tmp_path)
+    (tmp_path / 'dependencies').mkdir()
+    (tmp_path / 'lambda-tools').mkdir()
+    containers = FakeContainers()
+    metrics = RecordingMetrics()
+    tracer = RecordingTracer()
+    runtime = DockerSandboxRuntime(
+        config,
+        FakeDockerClient(containers),
+        metrics,
+        tracer,
+    )
+    created_at = datetime(2026, 4, 2, 12, 0, tzinfo=UTC)
+    expires_at = created_at + timedelta(minutes=5)
+
+    session = runtime.create(
+        session_id=SESSION_ID,
+        chat_id=CHAT_ID,
+        created_at=created_at,
+        expires_at=expires_at,
+    )
+
+    assert session.container_id == 'container-123'
+    duration_call = _find_record_call(
+        metrics,
+        'sandbox.runtime.create.duration_ms',
+        attrs={'operation': 'create', 'result': 'created'},
+    )
+    assert duration_call[1] >= 0
+    span = _find_span(
+        tracer,
+        'adapter.docker.create_sandbox',
+        {'chat.id': str(CHAT_ID), 'session.id': str(SESSION_ID)},
+        {
+            'container.id': 'container-123',
+            'sandbox.result': 'created',
+        },
+    )
+    assert not span.errors
+
+
+def test_runtime_create_raises_start_error_when_container_id_is_missing(
+    tmp_path: Path,
+) -> None:
+    config = build_config(tmp_path)
+    (tmp_path / 'dependencies').mkdir()
+    (tmp_path / 'lambda-tools').mkdir()
+    containers = FakeContainers(run_result=FakeContainer(''))
+    runtime = build_runtime(config, containers)
+
+    with pytest.raises(SandboxStartError) as excinfo:
+        runtime.create(
+            session_id=SESSION_ID,
+            chat_id=CHAT_ID,
+            created_at=datetime(2026, 4, 2, 12, 0, tzinfo=UTC),
+            expires_at=datetime(2026, 4, 2, 12, 5, tzinfo=UTC),
+        )
+
+    assert str(excinfo.value) == 'sandbox_start_failed'
+    assert excinfo.value.chat_id == str(CHAT_ID)
+
+
+def test_runtime_create_error_records_observability_when_container_id_missing(
+    tmp_path: Path,
+) -> None:
+    config = build_config(tmp_path)
+    (tmp_path / 'dependencies').mkdir()
+    (tmp_path / 'lambda-tools').mkdir()
+    containers = FakeContainers(run_result=FakeContainer(''))
+    metrics = RecordingMetrics()
+    tracer = RecordingTracer()
+    runtime = DockerSandboxRuntime(
+        config,
+        FakeDockerClient(containers),
+        metrics,
+        tracer,
+    )
+
+    with pytest.raises(SandboxStartError) as excinfo:
+        runtime.create(
+            session_id=SESSION_ID,
+            chat_id=CHAT_ID,
+            created_at=datetime(2026, 4, 2, 12, 0, tzinfo=UTC),
+            expires_at=datetime(2026, 4, 2, 12, 5, tzinfo=UTC),
+        )
+
+    assert str(excinfo.value) == 'sandbox_start_failed'
+    _find_increment_call(
+        metrics,
+        'sandbox.runtime.error.total',
+        attrs={'operation': 'create', 'error.type': 'SandboxStartError'},
+    )
+    duration_call = _find_record_call(
+        metrics,
+        'sandbox.runtime.create.duration_ms',
+        attrs={'operation': 'create', 'result': 'error'},
+    )
+    assert duration_call[1] >= 0
+    span = _find_span(
+        tracer,
+        'adapter.docker.create_sandbox',
+        {'chat.id': str(CHAT_ID), 'session.id': str(SESSION_ID)},
+        {'sandbox.result': 'error'},
+    )
+    assert excinfo.value in span.errors
+
+
+def test_runtime_stop_ignores_missing_container(tmp_path: Path) -> None:
+    config = build_config(tmp_path)
+    containers = FakeContainers()
+    containers.get_result = NotFound('missing')
+    metrics = RecordingMetrics()
+    tracer = RecordingTracer()
+    runtime = DockerSandboxRuntime(
+        config,
+        FakeDockerClient(containers),
+        metrics,
+        tracer,
+    )
+
+    runtime.stop('container-123')
+
+    assert containers.get_calls == ['container-123']
+    duration_call = _find_record_call(
+        metrics,
+        'sandbox.runtime.stop.duration_ms',
+        attrs={'operation': 'stop', 'result': 'not_found'},
+    )
+    assert duration_call[1] >= 0
+    span = _find_span(
+        tracer,
+        'adapter.docker.stop_sandbox',
+        {'container.id': 'container-123'},
+        {'sandbox.result': 'not_found'},
+    )
+    assert not span.errors
+    stop_error_calls = [
+        call
+        for call in metrics.increment_calls
+        if call[0] == 'sandbox.runtime.error.total'
+        and call[2] is not None
+        and call[2].get('operation') == 'stop'
+    ]
+    assert stop_error_calls == []
+
+
+def test_runtime_stop_wraps_docker_errors(tmp_path: Path) -> None:
+    config = build_config(tmp_path)
+    containers = FakeContainers()
+    containers.get_result = FailingStopContainer(
+        'container-123',
+        labels={
+            'session_id': str(SESSION_ID),
+            'chat_id': str(CHAT_ID),
+            'expires_at': '2026-04-02T12:05:00+00:00',
+        },
+        created_at='2026-04-02T12:00:00Z',
+        error=DockerException('boom'),
+    )
+    metrics = RecordingMetrics()
+    tracer = RecordingTracer()
+    runtime = DockerSandboxRuntime(
+        config,
+        FakeDockerClient(containers),
+        metrics,
+        tracer,
+    )
+
+    with pytest.raises(SandboxError) as excinfo:
+        runtime.stop('container-123')
+
+    assert str(excinfo.value) == 'sandbox_stop_failed'
+    _find_increment_call(
+        metrics,
+        'sandbox.runtime.error.total',
+        attrs={'operation': 'stop', 'error.type': 'DockerException'},
+    )
+    duration_call = _find_record_call(
+        metrics,
+        'sandbox.runtime.stop.duration_ms',
+        attrs={'operation': 'stop', 'result': 'error'},
+    )
+    assert duration_call[1] >= 0
+    span = _find_span(
+        tracer,
+        'adapter.docker.stop_sandbox',
+        {'container.id': 'container-123'},
+        {
+            'session.id': str(SESSION_ID),
+            'chat.id': str(CHAT_ID),
+            'sandbox.result': 'error',
+        },
+    )
+    cause = excinfo.value.__cause__
+    assert isinstance(cause, DockerException)
+    assert cause in span.errors
+
+
+def test_runtime_stop_records_observability_on_success(tmp_path: Path) -> None:
+    config = build_config(tmp_path)
+    containers = FakeContainers()
+    container = FakeListedContainer(
+        'container-123',
+        labels={
+            'session_id': str(SESSION_ID),
+            'chat_id': str(CHAT_ID),
+            'expires_at': '2026-04-02T12:05:00+00:00',
+        },
+        created_at='2026-04-02T12:00:00Z',
+    )
+    containers.get_result = container
+    metrics = RecordingMetrics()
+    tracer = RecordingTracer()
+    runtime = DockerSandboxRuntime(
+        config,
+        FakeDockerClient(containers),
+        metrics,
+        tracer,
+    )
+
+    runtime.stop('container-123')
+
+    assert container.stop_calls == 1
+    duration_call = _find_record_call(
+        metrics,
+        'sandbox.runtime.stop.duration_ms',
+        attrs={'operation': 'stop', 'result': 'stopped'},
+    )
+    assert duration_call[1] >= 0
+    span = _find_span(
+        tracer,
+        'adapter.docker.stop_sandbox',
+        {'container.id': 'container-123'},
+        {
+            'session.id': str(SESSION_ID),
+            'chat.id': str(CHAT_ID),
+            'sandbox.result': 'stopped',
+        },
+    )
+    assert not span.errors
+    stop_error_calls = [
+        call
+        for call in metrics.increment_calls
+        if call[0] == 'sandbox.runtime.error.total'
+        and call[2] is not None
+        and call[2].get('operation') == 'stop'
+    ]
+    assert stop_error_calls == []
+
+
+def test_runtime_list_active_sessions_reads_valid_labeled_containers(
+    tmp_path: Path,
+) -> None:
+    config = build_config(tmp_path)
+    containers = FakeContainers()
+    expires_at = datetime(2026, 4, 2, 12, 5, tzinfo=UTC)
+    containers.list_result = [
+        FakeListedContainer(
+            'container-123',
+            labels={
+                'session_id': str(SESSION_ID),
+                'chat_id': str(CHAT_ID),
+                'expires_at': expires_at.isoformat(),
+            },
+            created_at='2026-04-02T12:00:00Z',
+        ),
+        FakeListedContainer(
+            'container-bad',
+            labels={
+                'chat_id': str(CHAT_ID),
+                'expires_at': expires_at.isoformat(),
+            },
+            created_at='2026-04-02T12:01:00Z',
+        ),
+    ]
+    runtime = build_runtime(config, containers)
+
+    sessions = runtime.list_active_sessions()
+
+    assert sessions == [
+        SandboxSession(
+            session_id=SESSION_ID,
+            chat_id=CHAT_ID,
+            container_id='container-123',
+            status=SandboxStatus.RUNNING,
+            created_at=datetime(2026, 4, 2, 12, 0, tzinfo=UTC),
+            expires_at=expires_at,
+        )
+    ]
+    assert containers.list_calls == [
+        {'filters': {'label': ['session_id', 'chat_id', 'expires_at']}}
+    ]
+
+
+def test_runtime_list_active_records_observability(tmp_path: Path) -> None:
+    config = build_config(tmp_path)
+    containers = FakeContainers()
+    expires_at = datetime(2026, 4, 2, 12, 5, tzinfo=UTC)
+    containers.list_result = [
+        FakeListedContainer(
+            'container-123',
+            labels={
+                'session_id': str(SESSION_ID),
+                'chat_id': str(CHAT_ID),
+                'expires_at': expires_at.isoformat(),
+            },
+            created_at='2026-04-02T12:00:00Z',
+        ),
+        FakeListedContainer(
+            'container-bad',
+            labels={
+                'chat_id': str(CHAT_ID),
+                'expires_at': expires_at.isoformat(),
+            },
+            created_at='2026-04-02T12:01:00Z',
+        ),
+    ]
+    metrics = RecordingMetrics()
+    tracer = RecordingTracer()
+    runtime = DockerSandboxRuntime(
+        config,
+        FakeDockerClient(containers),
+        metrics,
+        tracer,
+    )
+
+    sessions = runtime.list_active_sessions()
+
+    assert len(sessions) == 1
+    duration_call = _find_record_call(
+        metrics,
+        'sandbox.runtime.list_active.duration_ms',
+        attrs={'operation': 'list_active', 'result': 'listed'},
+    )
+    assert duration_call[1] >= 0
+    span = _find_span(
+        tracer,
+        'adapter.docker.list_active_sandboxes',
+        span_attrs={
+            'sandbox.container_count': 2,
+            'sandbox.active_count': 1,
+            'sandbox.result': 'listed',
+        },
+    )
+    assert not span.errors
+
+
+def test_runtime_list_active_error_records_observability(tmp_path: Path) -> None:
+    config = build_config(tmp_path)
+    containers = FakeContainers()
+    containers.list_error = DockerException('boom')
+    metrics = RecordingMetrics()
+    tracer = RecordingTracer()
+    runtime = DockerSandboxRuntime(
+        config,
+        FakeDockerClient(containers),
+        metrics,
+        tracer,
+    )
+
+    with pytest.raises(SandboxError) as excinfo:
+        runtime.list_active_sessions()
+
+    assert str(excinfo.value) == 'sandbox_list_failed'
+    _find_increment_call(
+        metrics,
+        'sandbox.runtime.error.total',
+        attrs={'operation': 'list_active', 'error.type': 'DockerException'},
+    )
+    duration_call = _find_record_call(
+        metrics,
+        'sandbox.runtime.list_active.duration_ms',
+        attrs={'operation': 'list_active', 'result': 'error'},
+    )
+    assert duration_call[1] >= 0
+    span = _find_span(
+        tracer,
+        'adapter.docker.list_active_sandboxes',
+        span_attrs={'sandbox.result': 'error'},
+    )
+    assert isinstance(excinfo.value.__cause__, DockerException)
+    assert excinfo.value in span.errors

test/test_sandbox_lock.py

@@ -0,0 +1,93 @@
+import threading
+from uuid import UUID
+
+from repository.sandbox_lock import ProcessLocalSandboxLifecycleLocker
+
+CHAT_ID = UUID('77777777-7777-7777-7777-777777777777')
+
+
+class LockRace:
+    def __init__(self, locker: ProcessLocalSandboxLifecycleLocker) -> None:
+        self.locker = locker
+        self.entered_first = threading.Event()
+        self.second_requested = threading.Event()
+        self.second_entered = threading.Event()
+        self.release_first = threading.Event()
+        self.release_second = threading.Event()
+        self.errors: list[Exception] = []
+        self.order: list[str] = []
+        self.first_entry: object | None = None
+
+    def run_first(self) -> None:
+        try:
+            with self.locker.lock(CHAT_ID):
+                self.first_entry = self.locker._locks_by_chat_id[CHAT_ID]
+                self.order.append('first_entered')
+                self.entered_first.set()
+                assert self.release_first.wait(timeout=1)
+                self.order.append('first_releasing')
+        except Exception as exc:
+            self.errors.append(exc)
+
+    def run_second(self) -> None:
+        try:
+            assert self.entered_first.wait(timeout=1)
+            context = self.locker.lock(CHAT_ID)
+            self.second_requested.set()
+
+            with context:
+                self.order.append('second_entered')
+                self.second_entered.set()
+                assert self.release_second.wait(timeout=1)
+                self.order.append('second_releasing')
+        except Exception as exc:
+            self.errors.append(exc)
+
+
+def test_process_local_sandbox_lifecycle_locker_evicts_idle_lock() -> None:
+    locker = ProcessLocalSandboxLifecycleLocker()
+
+    with locker.lock(CHAT_ID):
+        assert CHAT_ID in locker._locks_by_chat_id
+        assert len(locker._locks_by_chat_id) == 1
+
+    assert CHAT_ID not in locker._locks_by_chat_id
+    assert len(locker._locks_by_chat_id) == 0
+
+
+def test_process_local_sandbox_lifecycle_locker_keeps_shared_lock_for_waiters() -> None:
+    locker = ProcessLocalSandboxLifecycleLocker()
+    race = LockRace(locker)
+    first_thread = threading.Thread(target=race.run_first)
+    second_thread = threading.Thread(target=race.run_second)
+
+    first_thread.start()
+    assert race.entered_first.wait(timeout=1)
+
+    second_thread.start()
+    assert race.second_requested.wait(timeout=1)
+    assert len(locker._locks_by_chat_id) == 1
+    assert locker._locks_by_chat_id[CHAT_ID] is race.first_entry
+    assert not race.second_entered.wait(timeout=0.1)
+
+    race.release_first.set()
+    assert race.second_entered.wait(timeout=1)
+    assert len(locker._locks_by_chat_id) == 1
+    assert locker._locks_by_chat_id[CHAT_ID] is race.first_entry
+
+    race.release_second.set()
+
+    first_thread.join(timeout=1)
+    second_thread.join(timeout=1)
+
+    assert not first_thread.is_alive()
+    assert not second_thread.is_alive()
+    assert race.errors == []
+    assert race.order == [
+        'first_entered',
+        'first_releasing',
+        'second_entered',
+        'second_releasing',
+    ]
+    assert CHAT_ID not in locker._locks_by_chat_id
+    assert len(locker._locks_by_chat_id) == 0

test/test_storage_metadata_repositories.py

@@ -0,0 +1,129 @@
+from datetime import UTC, datetime
+from uuid import UUID
+
+from domain.chat import Chat, ChatAttachmentName, ChatFile
+from domain.workspace import Workspace
+from repository.chat import InMemoryChatRepository
+from repository.chat_file import InMemoryChatFileRepository
+from repository.workspace import InMemoryWorkspaceRepository
+
+USER_A = UUID('aaaaaaaa-aaaa-aaaa-aaaa-aaaaaaaaaaaa')
+USER_B = UUID('bbbbbbbb-bbbb-bbbb-bbbb-bbbbbbbbbbbb')
+WS_A = UUID('11111111-1111-1111-1111-111111111111')
+WS_B = UUID('22222222-2222-2222-2222-222222222222')
+CHAT_A = UUID('33333333-3333-3333-3333-333333333333')
+CHAT_B = UUID('44444444-4444-4444-4444-444444444444')
+FILE_A = UUID('55555555-5555-5555-5555-555555555555')
+FILE_B = UUID('66666666-6666-6666-6666-666666666666')
+TS = datetime(2026, 4, 1, 12, 0, 0, tzinfo=UTC)
+TS_2 = datetime(2026, 4, 1, 13, 0, 0, tzinfo=UTC)
+
+
+def test_workspace_get_by_user_id() -> None:
+    repo = InMemoryWorkspaceRepository()
+    ws = Workspace(workspace_id=WS_A, user_id=USER_A, created_at=TS)
+    repo.save(ws)
+    assert repo.get(WS_A) == ws
+    assert repo.get_by_user_id(USER_A) == ws
+    assert repo.get_by_user_id(USER_B) is None
+
+
+def test_workspace_replace_for_user() -> None:
+    repo = InMemoryWorkspaceRepository()
+    ws = Workspace(workspace_id=WS_A, user_id=USER_A, created_at=TS)
+    repo.save(ws)
+    new_ws = Workspace(workspace_id=WS_B, user_id=USER_A, created_at=TS_2)
+    repo.save(new_ws)
+    assert repo.get_by_user_id(USER_A) == new_ws
+    assert repo.get(WS_A) is None
+    assert repo.get(WS_B) == new_ws
+
+
+def test_chat_crud_workspace_scope() -> None:
+    chat_repo = InMemoryChatRepository()
+    chat_a = Chat(chat_id=CHAT_A, workspace_id=WS_A, created_at=TS)
+    chat_b = Chat(chat_id=CHAT_B, workspace_id=WS_A, created_at=TS_2)
+    chat_repo.save(chat_a)
+    chat_repo.save(chat_b)
+
+    listed = chat_repo.list_by_workspace_id(WS_A)
+    assert listed == [chat_a, chat_b]
+
+    assert chat_repo.get(CHAT_A) == chat_a
+    chat_repo.delete(CHAT_A)
+    assert chat_repo.get(CHAT_A) is None
+    assert chat_repo.list_by_workspace_id(WS_A) == [chat_b]
+
+
+def test_chat_list_only_same_workspace() -> None:
+    chat_repo = InMemoryChatRepository()
+    chat_a = Chat(chat_id=CHAT_A, workspace_id=WS_A, created_at=TS)
+    chat_b = Chat(chat_id=CHAT_B, workspace_id=WS_B, created_at=TS_2)
+    chat_repo.save(chat_a)
+    chat_repo.save(chat_b)
+    assert chat_repo.list_by_workspace_id(WS_A) == [chat_a]
+    assert chat_repo.list_by_workspace_id(WS_B) == [chat_b]
+
+
+def test_chat_file_metadata_save_get_list_delete_clear() -> None:
+    name_a = ChatAttachmentName('doc.pdf')
+    name_b = ChatAttachmentName('x.png')
+
+    repo = InMemoryChatFileRepository()
+    f_a = ChatFile(
+        file_id=FILE_A,
+        chat_id=CHAT_A,
+        name=name_a,
+        content_type='application/pdf',
+        size_bytes=100,
+        created_at=TS,
+    )
+    f_b = ChatFile(
+        file_id=FILE_B,
+        chat_id=CHAT_A,
+        name=name_b,
+        content_type='image/png',
+        size_bytes=200,
+        created_at=TS_2,
+    )
+    repo.save(f_a)
+    repo.save(f_b)
+
+    assert repo.get(FILE_A) == f_a
+    assert repo.get_by_chat_id_and_name(CHAT_A, name_a) == f_a
+    listed = repo.list_by_chat_id(CHAT_A)
+    assert listed == [f_a, f_b]
+
+    repo.delete(FILE_A)
+    assert repo.get(FILE_A) is None
+    assert repo.get_by_chat_id_and_name(CHAT_A, name_a) is None
+
+    repo.save(f_a)
+    repo.delete_by_chat_id(CHAT_A)
+    assert repo.list_by_chat_id(CHAT_A) == []
+
+
+def test_chat_file_same_name_replaced_by_new_id() -> None:
+    name = ChatAttachmentName('a.txt')
+    repo = InMemoryChatFileRepository()
+    first = ChatFile(
+        file_id=FILE_A,
+        chat_id=CHAT_A,
+        name=name,
+        content_type='text/plain',
+        size_bytes=1,
+        created_at=TS,
+    )
+    second = ChatFile(
+        file_id=FILE_B,
+        chat_id=CHAT_A,
+        name=name,
+        content_type='text/plain',
+        size_bytes=2,
+        created_at=TS_2,
+    )
+    repo.save(first)
+    repo.save(second)
+    assert repo.get(FILE_A) is None
+    assert repo.get(FILE_B) == second
+    assert repo.get_by_chat_id_and_name(CHAT_A, name) == second

usecase/interface.py

@@ -1,8 +1,13 @@
 from collections.abc import Mapping
+from datetime import datetime
 from types import TracebackType
 from typing import Protocol, TypeAlias
+from uuid import UUID
 
+from domain.chat import Chat, ChatAttachmentName, ChatFile
+from domain.sandbox import SandboxSession
 from domain.user import User
+from domain.workspace import Workspace, WorkspaceUsage
 
 AttrValue: TypeAlias = str | int | float | bool
 Attrs: TypeAlias = Mapping[str, AttrValue]
@@ -16,6 +21,125 @@ class UserRepository(Protocol):
     def save(self, user: User) -> None: ...
 
 
+class WorkspaceRepository(Protocol):
+    def get(self, workspace_id: UUID) -> Workspace | None: ...
+
+    def get_by_user_id(self, user_id: UUID) -> Workspace | None: ...
+
+    def save(self, workspace: Workspace) -> None: ...
+
+
+class ChatRepository(Protocol):
+    def get(self, chat_id: UUID) -> Chat | None: ...
+
+    def list_by_workspace_id(self, workspace_id: UUID) -> list[Chat]: ...
+
+    def save(self, chat: Chat) -> None: ...
+
+    def delete(self, chat_id: UUID) -> None: ...
+
+
+class ChatFileRepository(Protocol):
+    def get(self, file_id: UUID) -> ChatFile | None: ...
+
+    def get_by_chat_id_and_name(
+        self,
+        chat_id: UUID,
+        name: ChatAttachmentName,
+    ) -> ChatFile | None: ...
+
+    def list_by_chat_id(self, chat_id: UUID) -> list[ChatFile]: ...
+
+    def save(self, chat_file: ChatFile) -> None: ...
+
+    def delete(self, file_id: UUID) -> None: ...
+
+    def delete_by_chat_id(self, chat_id: UUID) -> None: ...
+
+
+class ChatStorage(Protocol):
+    def ensure_chat(self, chat: Chat) -> None: ...
+
+    def read_history(self, chat: Chat) -> str: ...
+
+    def write_history(self, chat: Chat, content: str) -> None: ...
+
+    def delete_chat(self, chat: Chat) -> None: ...
+
+    def write_attachment(
+        self,
+        chat: Chat,
+        file_name: ChatAttachmentName,
+        content: bytes,
+    ) -> int: ...
+
+    def read_attachment(self, chat: Chat, file_name: ChatAttachmentName) -> bytes: ...
+
+    def delete_attachment(
+        self,
+        chat: Chat,
+        file_name: ChatAttachmentName,
+    ) -> None: ...
+
+    def clear_attachments(self, chat: Chat) -> None: ...
+
+
+class StorageUsageReader(Protocol):
+    def get_workspace_usage(
+        self,
+        workspace: Workspace,
+        chats: list[Chat],
+    ) -> WorkspaceUsage: ...
+
+
+class IdGenerator(Protocol):
+    def new(self) -> UUID: ...
+
+
+class SandboxSessionRepository(Protocol):
+    def get_active_by_chat_id(self, chat_id: UUID) -> SandboxSession | None: ...
+
+    def list_expired(self, now: datetime) -> list[SandboxSession]: ...
+
+    def count_active(self) -> int: ...
+
+    def save(self, session: SandboxSession) -> None: ...
+
+    def delete(self, session_id: UUID) -> None: ...
+
+
+class LockContext(Protocol):
+    def __enter__(self) -> None: ...
+
+    def __exit__(
+        self,
+        exc_type: type[BaseException] | None,
+        exc: BaseException | None,
+        traceback: TracebackType | None,
+    ) -> bool | None: ...
+
+
+class SandboxLifecycleLocker(Protocol):
+    def lock(self, chat_id: UUID) -> LockContext: ...
+
+
+class SandboxRuntime(Protocol):
+    def create(
+        self,
+        *,
+        session_id: UUID,
+        chat_id: UUID,
+        created_at: datetime,
+        expires_at: datetime,
+    ) -> SandboxSession: ...
+
+    def stop(self, container_id: str) -> None: ...
+
+
+class Clock(Protocol):
+    def now(self) -> datetime: ...
+
+
 class Logger(Protocol):
     def debug(self, message: str, attrs: Attrs | None = None) -> None: ...
 
@@ -41,6 +165,13 @@ class Metrics(Protocol):
         attrs: Attrs | None = None,
     ) -> None: ...
 
+    def set(
+        self,
+        name: str,
+        value: int | float,
+        attrs: Attrs | None = None,
+    ) -> None: ...
+
 
 class Span(Protocol):
     def set_attribute(self, name: str, value: AttrValue) -> None: ...

usecase/sandbox.py

@@ -0,0 +1,303 @@
+from dataclasses import dataclass
+from datetime import timedelta
+from uuid import UUID, uuid4
+
+from domain.sandbox import SandboxSession
+from usecase.interface import (
+    Clock,
+    Logger,
+    Metrics,
+    SandboxLifecycleLocker,
+    SandboxRuntime,
+    SandboxSessionRepository,
+    Tracer,
+)
+
+
+@dataclass(frozen=True, slots=True)
+class CreateSandboxCommand:
+    chat_id: UUID
+
+
+class CreateSandbox:
+    def __init__(
+        self,
+        repository: SandboxSessionRepository,
+        locker: SandboxLifecycleLocker,
+        runtime: SandboxRuntime,
+        clock: Clock,
+        logger: Logger,
+        metrics: Metrics,
+        tracer: Tracer,
+        ttl: timedelta,
+    ) -> None:
+        self._repository = repository
+        self._locker = locker
+        self._runtime = runtime
+        self._clock = clock
+        self._logger = logger
+        self._metrics = metrics
+        self._tracer = tracer
+        self._ttl = ttl
+
+    def execute(self, command: CreateSandboxCommand) -> SandboxSession:
+        chat_id = command.chat_id
+
+        with self._tracer.start_span(
+            'usecase.create_sandbox',
+            attrs={'chat.id': str(chat_id)},
+        ) as span:
+            try:
+                with self._locker.lock(chat_id):
+                    session = self._repository.get_active_by_chat_id(chat_id)
+                    now = self._clock.now()
+
+                    if session is not None and session.expires_at > now:
+                        span.set_attribute('session.id', str(session.session_id))
+                        span.set_attribute('container.id', session.container_id)
+                        span.set_attribute('sandbox.result', 'reused')
+                        self._metrics.increment(
+                            'sandbox.create.total',
+                            attrs=_result_metric_attrs('reused'),
+                        )
+                        self._logger.info(
+                            'sandbox_reused',
+                            attrs=_sandbox_attrs(session),
+                        )
+                        return session
+
+                    result = 'created'
+                    new_session_id: UUID | None = None
+                    if session is not None:
+                        result = 'replaced'
+                        new_session_id = _new_session_id()
+                        span.set_attribute(
+                            'sandbox.previous_session.id',
+                            str(session.session_id),
+                        )
+                        span.set_attribute(
+                            'sandbox.previous_container.id',
+                            session.container_id,
+                        )
+                        span.set_attribute(
+                            'sandbox.new_session.id',
+                            str(new_session_id),
+                        )
+                        self._logger.info(
+                            'sandbox_replaced',
+                            attrs=_sandbox_attrs(session),
+                        )
+                        self._runtime.stop(session.container_id)
+                        self._repository.delete(session.session_id)
+                        _set_active_count(self._metrics, self._repository)
+
+                    created_at = self._clock.now()
+                    expires_at = created_at + self._ttl
+                    if new_session_id is None:
+                        new_session_id = _new_session_id()
+                        span.set_attribute('session.id', str(new_session_id))
+                    new_session = self._runtime.create(
+                        session_id=new_session_id,
+                        chat_id=chat_id,
+                        created_at=created_at,
+                        expires_at=expires_at,
+                    )
+                    if result == 'replaced':
+                        span.set_attribute(
+                            'sandbox.new_container.id',
+                            new_session.container_id,
+                        )
+                    self._save_created_session(new_session)
+                    _set_active_count(self._metrics, self._repository)
+                    if result == 'replaced':
+                        span.set_attribute('session.id', str(new_session.session_id))
+                    span.set_attribute('container.id', new_session.container_id)
+                    span.set_attribute('sandbox.result', result)
+                    self._metrics.increment(
+                        'sandbox.create.total',
+                        attrs=_result_metric_attrs(result),
+                    )
+                    self._logger.info(
+                        'sandbox_created',
+                        attrs=_sandbox_attrs(new_session),
+                    )
+                    return new_session
+            except Exception as exc:
+                span.set_attribute('sandbox.result', 'error')
+                self._metrics.increment(
+                    'sandbox.create.total',
+                    attrs=_result_metric_attrs('error'),
+                )
+                span.record_error(exc)
+                raise
+
+    def _save_created_session(self, session: SandboxSession) -> None:
+        try:
+            self._repository.save(session)
+        except Exception as exc:
+            self._compensate_save_failure(session, exc)
+            raise
+
+    def _compensate_save_failure(
+        self,
+        session: SandboxSession,
+        error: Exception,
+    ) -> None:
+        try:
+            self._runtime.stop(session.container_id)
+        except Exception as stop_error:
+            _set_active_count(self._metrics, self._repository)
+            raise error from stop_error
+
+        _set_active_count(self._metrics, self._repository)
+
+
+class CleanupExpiredSandboxes:
+    def __init__(
+        self,
+        repository: SandboxSessionRepository,
+        locker: SandboxLifecycleLocker,
+        runtime: SandboxRuntime,
+        clock: Clock,
+        logger: Logger,
+        metrics: Metrics,
+        tracer: Tracer,
+    ) -> None:
+        self._repository = repository
+        self._locker = locker
+        self._runtime = runtime
+        self._clock = clock
+        self._logger = logger
+        self._metrics = metrics
+        self._tracer = tracer
+
+    def execute(self) -> list[SandboxSession]:
+        cleaned_sessions: list[SandboxSession] = []
+        error_count = 0
+
+        with self._tracer.start_span(
+            'usecase.cleanup_expired_sandboxes',
+        ) as span:
+            try:
+                expired_sessions = self._repository.list_expired(self._clock.now())
+            except Exception as exc:
+                span.set_attribute('sandbox.result', 'error')
+                self._metrics.increment(
+                    'sandbox.cleanup.error.total',
+                    attrs=_cleanup_error_metric_attrs(
+                        type(exc).__name__,
+                        'list_expired',
+                    ),
+                )
+                span.record_error(exc)
+                raise
+
+            span.set_attribute('sandbox.expired_count', len(expired_sessions))
+            for session in expired_sessions:
+                with self._tracer.start_span(
+                    'usecase.cleanup_expired_sandbox',
+                    attrs=_sandbox_span_attrs(session),
+                ) as cleanup_span:
+                    try:
+                        cleaned_session = self._cleanup_session(session)
+                    except Exception as exc:
+                        error_count += 1
+                        cleanup_span.set_attribute('sandbox.result', 'error')
+                        cleanup_span.record_error(exc)
+                        self._metrics.increment(
+                            'sandbox.cleanup.error.total',
+                            attrs=_error_metric_attrs(type(exc).__name__),
+                        )
+                        attrs = _sandbox_attrs(session)
+                        attrs['error'] = type(exc).__name__
+                        self._logger.error(
+                            'sandbox_clean_failed',
+                            attrs=attrs,
+                        )
+                        continue
+
+                    if cleaned_session is None:
+                        cleanup_span.set_attribute('sandbox.result', 'skipped')
+                        continue
+
+                    cleanup_span.set_attribute('sandbox.result', 'cleaned')
+                    cleaned_sessions.append(cleaned_session)
+                    self._metrics.increment(
+                        'sandbox.cleanup.total',
+                        attrs=_result_metric_attrs('cleaned'),
+                    )
+                    self._logger.info(
+                        'sandbox_cleaned',
+                        attrs=_sandbox_attrs(cleaned_session),
+                    )
+
+            span.set_attribute('sandbox.cleaned_count', len(cleaned_sessions))
+            span.set_attribute('sandbox.error_count', error_count)
+            span.set_attribute(
+                'sandbox.result',
+                'completed' if error_count == 0 else 'completed_with_errors',
+            )
+            return cleaned_sessions
+
+    def _cleanup_session(self, session: SandboxSession) -> SandboxSession | None:
+        with self._locker.lock(session.chat_id):
+            current_session = self._repository.get_active_by_chat_id(session.chat_id)
+            now = self._clock.now()
+            if current_session is None:
+                return None
+
+            if current_session.session_id != session.session_id:
+                return None
+
+            if current_session.expires_at > now:
+                return None
+
+            self._runtime.stop(current_session.container_id)
+            self._repository.delete(current_session.session_id)
+            _set_active_count(self._metrics, self._repository)
+            return current_session
+
+
+def _new_session_id() -> UUID:
+    return uuid4()
+
+
+def _sandbox_attrs(session: SandboxSession) -> dict[str, str]:
+    return {
+        'chat_id': str(session.chat_id),
+        'session_id': str(session.session_id),
+        'container_id': session.container_id,
+    }
+
+
+def _sandbox_span_attrs(session: SandboxSession) -> dict[str, str]:
+    return {
+        'chat.id': str(session.chat_id),
+        'session.id': str(session.session_id),
+        'container.id': session.container_id,
+    }
+
+
+def _result_metric_attrs(result: str) -> dict[str, str]:
+    return {'result': result}
+
+
+def _error_metric_attrs(error_type: str) -> dict[str, str]:
+    return {'error.type': error_type}
+
+
+def _cleanup_error_metric_attrs(
+    error_type: str,
+    reason: str,
+) -> dict[str, str]:
+    return {
+        'error.type': error_type,
+        'reason': reason,
+    }
+
+
+def _set_active_count(
+    metrics: Metrics,
+    repository: SandboxSessionRepository,
+) -> None:
+    metrics.set('sandbox.active.count', repository.count_active())

uv.lock

@@ -50,6 +50,51 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/9a/3c/c17fb3ca2d9c3acff52e30b309f538586f9f5b9c9cf454f3845fc9af4881/certifi-2026.2.25-py3-none-any.whl", hash = "sha256:027692e4402ad994f1c42e52a4997a9763c646b73e4096e4d5d6db8af1d6f0fa", size = 153684, upload-time = "2026-02-25T02:54:15.766Z" },
 ]
 
+[[package]]
+name = "cffi"
+version = "2.0.0"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "pycparser", marker = "implementation_name != 'PyPy'" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/eb/56/b1ba7935a17738ae8453301356628e8147c79dbb825bcbc73dc7401f9846/cffi-2.0.0.tar.gz", hash = "sha256:44d1b5909021139fe36001ae048dbdde8214afa20200eda0f64c068cac5d5529", size = 523588, upload-time = "2025-09-08T23:24:04.541Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/4b/8d/a0a47a0c9e413a658623d014e91e74a50cdd2c423f7ccfd44086ef767f90/cffi-2.0.0-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:00bdf7acc5f795150faa6957054fbbca2439db2f775ce831222b66f192f03beb", size = 185230, upload-time = "2025-09-08T23:23:00.879Z" },
+    { url = "https://files.pythonhosted.org/packages/4a/d2/a6c0296814556c68ee32009d9c2ad4f85f2707cdecfd7727951ec228005d/cffi-2.0.0-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:45d5e886156860dc35862657e1494b9bae8dfa63bf56796f2fb56e1679fc0bca", size = 181043, upload-time = "2025-09-08T23:23:02.231Z" },
+    { url = "https://files.pythonhosted.org/packages/b0/1e/d22cc63332bd59b06481ceaac49d6c507598642e2230f201649058a7e704/cffi-2.0.0-cp313-cp313-manylinux1_i686.manylinux2014_i686.manylinux_2_17_i686.manylinux_2_5_i686.whl", hash = "sha256:07b271772c100085dd28b74fa0cd81c8fb1a3ba18b21e03d7c27f3436a10606b", size = 212446, upload-time = "2025-09-08T23:23:03.472Z" },
+    { url = "https://files.pythonhosted.org/packages/a9/f5/a2c23eb03b61a0b8747f211eb716446c826ad66818ddc7810cc2cc19b3f2/cffi-2.0.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:d48a880098c96020b02d5a1f7d9251308510ce8858940e6fa99ece33f610838b", size = 220101, upload-time = "2025-09-08T23:23:04.792Z" },
+    { url = "https://files.pythonhosted.org/packages/f2/7f/e6647792fc5850d634695bc0e6ab4111ae88e89981d35ac269956605feba/cffi-2.0.0-cp313-cp313-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:f93fd8e5c8c0a4aa1f424d6173f14a892044054871c771f8566e4008eaa359d2", size = 207948, upload-time = "2025-09-08T23:23:06.127Z" },
+    { url = "https://files.pythonhosted.org/packages/cb/1e/a5a1bd6f1fb30f22573f76533de12a00bf274abcdc55c8edab639078abb6/cffi-2.0.0-cp313-cp313-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:dd4f05f54a52fb558f1ba9f528228066954fee3ebe629fc1660d874d040ae5a3", size = 206422, upload-time = "2025-09-08T23:23:07.753Z" },
+    { url = "https://files.pythonhosted.org/packages/98/df/0a1755e750013a2081e863e7cd37e0cdd02664372c754e5560099eb7aa44/cffi-2.0.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:c8d3b5532fc71b7a77c09192b4a5a200ea992702734a2e9279a37f2478236f26", size = 219499, upload-time = "2025-09-08T23:23:09.648Z" },
+    { url = "https://files.pythonhosted.org/packages/50/e1/a969e687fcf9ea58e6e2a928ad5e2dd88cc12f6f0ab477e9971f2309b57c/cffi-2.0.0-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:d9b29c1f0ae438d5ee9acb31cadee00a58c46cc9c0b2f9038c6b0b3470877a8c", size = 222928, upload-time = "2025-09-08T23:23:10.928Z" },
+    { url = "https://files.pythonhosted.org/packages/36/54/0362578dd2c9e557a28ac77698ed67323ed5b9775ca9d3fe73fe191bb5d8/cffi-2.0.0-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:6d50360be4546678fc1b79ffe7a66265e28667840010348dd69a314145807a1b", size = 221302, upload-time = "2025-09-08T23:23:12.42Z" },
+    { url = "https://files.pythonhosted.org/packages/eb/6d/bf9bda840d5f1dfdbf0feca87fbdb64a918a69bca42cfa0ba7b137c48cb8/cffi-2.0.0-cp313-cp313-win32.whl", hash = "sha256:74a03b9698e198d47562765773b4a8309919089150a0bb17d829ad7b44b60d27", size = 172909, upload-time = "2025-09-08T23:23:14.32Z" },
+    { url = "https://files.pythonhosted.org/packages/37/18/6519e1ee6f5a1e579e04b9ddb6f1676c17368a7aba48299c3759bbc3c8b3/cffi-2.0.0-cp313-cp313-win_amd64.whl", hash = "sha256:19f705ada2530c1167abacb171925dd886168931e0a7b78f5bffcae5c6b5be75", size = 183402, upload-time = "2025-09-08T23:23:15.535Z" },
+    { url = "https://files.pythonhosted.org/packages/cb/0e/02ceeec9a7d6ee63bb596121c2c8e9b3a9e150936f4fbef6ca1943e6137c/cffi-2.0.0-cp313-cp313-win_arm64.whl", hash = "sha256:256f80b80ca3853f90c21b23ee78cd008713787b1b1e93eae9f3d6a7134abd91", size = 177780, upload-time = "2025-09-08T23:23:16.761Z" },
+    { url = "https://files.pythonhosted.org/packages/92/c4/3ce07396253a83250ee98564f8d7e9789fab8e58858f35d07a9a2c78de9f/cffi-2.0.0-cp314-cp314-macosx_10_13_x86_64.whl", hash = "sha256:fc33c5141b55ed366cfaad382df24fe7dcbc686de5be719b207bb248e3053dc5", size = 185320, upload-time = "2025-09-08T23:23:18.087Z" },
+    { url = "https://files.pythonhosted.org/packages/59/dd/27e9fa567a23931c838c6b02d0764611c62290062a6d4e8ff7863daf9730/cffi-2.0.0-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:c654de545946e0db659b3400168c9ad31b5d29593291482c43e3564effbcee13", size = 181487, upload-time = "2025-09-08T23:23:19.622Z" },
+    { url = "https://files.pythonhosted.org/packages/d6/43/0e822876f87ea8a4ef95442c3d766a06a51fc5298823f884ef87aaad168c/cffi-2.0.0-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:24b6f81f1983e6df8db3adc38562c83f7d4a0c36162885ec7f7b77c7dcbec97b", size = 220049, upload-time = "2025-09-08T23:23:20.853Z" },
+    { url = "https://files.pythonhosted.org/packages/b4/89/76799151d9c2d2d1ead63c2429da9ea9d7aac304603de0c6e8764e6e8e70/cffi-2.0.0-cp314-cp314-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:12873ca6cb9b0f0d3a0da705d6086fe911591737a59f28b7936bdfed27c0d47c", size = 207793, upload-time = "2025-09-08T23:23:22.08Z" },
+    { url = "https://files.pythonhosted.org/packages/bb/dd/3465b14bb9e24ee24cb88c9e3730f6de63111fffe513492bf8c808a3547e/cffi-2.0.0-cp314-cp314-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:d9b97165e8aed9272a6bb17c01e3cc5871a594a446ebedc996e2397a1c1ea8ef", size = 206300, upload-time = "2025-09-08T23:23:23.314Z" },
+    { url = "https://files.pythonhosted.org/packages/47/d9/d83e293854571c877a92da46fdec39158f8d7e68da75bf73581225d28e90/cffi-2.0.0-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:afb8db5439b81cf9c9d0c80404b60c3cc9c3add93e114dcae767f1477cb53775", size = 219244, upload-time = "2025-09-08T23:23:24.541Z" },
+    { url = "https://files.pythonhosted.org/packages/2b/0f/1f177e3683aead2bb00f7679a16451d302c436b5cbf2505f0ea8146ef59e/cffi-2.0.0-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:737fe7d37e1a1bffe70bd5754ea763a62a066dc5913ca57e957824b72a85e205", size = 222828, upload-time = "2025-09-08T23:23:26.143Z" },
+    { url = "https://files.pythonhosted.org/packages/c6/0f/cafacebd4b040e3119dcb32fed8bdef8dfe94da653155f9d0b9dc660166e/cffi-2.0.0-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:38100abb9d1b1435bc4cc340bb4489635dc2f0da7456590877030c9b3d40b0c1", size = 220926, upload-time = "2025-09-08T23:23:27.873Z" },
+    { url = "https://files.pythonhosted.org/packages/3e/aa/df335faa45b395396fcbc03de2dfcab242cd61a9900e914fe682a59170b1/cffi-2.0.0-cp314-cp314-win32.whl", hash = "sha256:087067fa8953339c723661eda6b54bc98c5625757ea62e95eb4898ad5e776e9f", size = 175328, upload-time = "2025-09-08T23:23:44.61Z" },
+    { url = "https://files.pythonhosted.org/packages/bb/92/882c2d30831744296ce713f0feb4c1cd30f346ef747b530b5318715cc367/cffi-2.0.0-cp314-cp314-win_amd64.whl", hash = "sha256:203a48d1fb583fc7d78a4c6655692963b860a417c0528492a6bc21f1aaefab25", size = 185650, upload-time = "2025-09-08T23:23:45.848Z" },
+    { url = "https://files.pythonhosted.org/packages/9f/2c/98ece204b9d35a7366b5b2c6539c350313ca13932143e79dc133ba757104/cffi-2.0.0-cp314-cp314-win_arm64.whl", hash = "sha256:dbd5c7a25a7cb98f5ca55d258b103a2054f859a46ae11aaf23134f9cc0d356ad", size = 180687, upload-time = "2025-09-08T23:23:47.105Z" },
+    { url = "https://files.pythonhosted.org/packages/3e/61/c768e4d548bfa607abcda77423448df8c471f25dbe64fb2ef6d555eae006/cffi-2.0.0-cp314-cp314t-macosx_10_13_x86_64.whl", hash = "sha256:9a67fc9e8eb39039280526379fb3a70023d77caec1852002b4da7e8b270c4dd9", size = 188773, upload-time = "2025-09-08T23:23:29.347Z" },
+    { url = "https://files.pythonhosted.org/packages/2c/ea/5f76bce7cf6fcd0ab1a1058b5af899bfbef198bea4d5686da88471ea0336/cffi-2.0.0-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:7a66c7204d8869299919db4d5069a82f1561581af12b11b3c9f48c584eb8743d", size = 185013, upload-time = "2025-09-08T23:23:30.63Z" },
+    { url = "https://files.pythonhosted.org/packages/be/b4/c56878d0d1755cf9caa54ba71e5d049479c52f9e4afc230f06822162ab2f/cffi-2.0.0-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:7cc09976e8b56f8cebd752f7113ad07752461f48a58cbba644139015ac24954c", size = 221593, upload-time = "2025-09-08T23:23:31.91Z" },
+    { url = "https://files.pythonhosted.org/packages/e0/0d/eb704606dfe8033e7128df5e90fee946bbcb64a04fcdaa97321309004000/cffi-2.0.0-cp314-cp314t-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:92b68146a71df78564e4ef48af17551a5ddd142e5190cdf2c5624d0c3ff5b2e8", size = 209354, upload-time = "2025-09-08T23:23:33.214Z" },
+    { url = "https://files.pythonhosted.org/packages/d8/19/3c435d727b368ca475fb8742ab97c9cb13a0de600ce86f62eab7fa3eea60/cffi-2.0.0-cp314-cp314t-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:b1e74d11748e7e98e2f426ab176d4ed720a64412b6a15054378afdb71e0f37dc", size = 208480, upload-time = "2025-09-08T23:23:34.495Z" },
+    { url = "https://files.pythonhosted.org/packages/d0/44/681604464ed9541673e486521497406fadcc15b5217c3e326b061696899a/cffi-2.0.0-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:28a3a209b96630bca57cce802da70c266eb08c6e97e5afd61a75611ee6c64592", size = 221584, upload-time = "2025-09-08T23:23:36.096Z" },
+    { url = "https://files.pythonhosted.org/packages/25/8e/342a504ff018a2825d395d44d63a767dd8ebc927ebda557fecdaca3ac33a/cffi-2.0.0-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:7553fb2090d71822f02c629afe6042c299edf91ba1bf94951165613553984512", size = 224443, upload-time = "2025-09-08T23:23:37.328Z" },
+    { url = "https://files.pythonhosted.org/packages/e1/5e/b666bacbbc60fbf415ba9988324a132c9a7a0448a9a8f125074671c0f2c3/cffi-2.0.0-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:6c6c373cfc5c83a975506110d17457138c8c63016b563cc9ed6e056a82f13ce4", size = 223437, upload-time = "2025-09-08T23:23:38.945Z" },
+    { url = "https://files.pythonhosted.org/packages/a0/1d/ec1a60bd1a10daa292d3cd6bb0b359a81607154fb8165f3ec95fe003b85c/cffi-2.0.0-cp314-cp314t-win32.whl", hash = "sha256:1fc9ea04857caf665289b7a75923f2c6ed559b8298a1b8c49e59f7dd95c8481e", size = 180487, upload-time = "2025-09-08T23:23:40.423Z" },
+    { url = "https://files.pythonhosted.org/packages/bf/41/4c1168c74fac325c0c8156f04b6749c8b6a8f405bbf91413ba088359f60d/cffi-2.0.0-cp314-cp314t-win_amd64.whl", hash = "sha256:d68b6cef7827e8641e8ef16f4494edda8b36104d79773a334beaa1e3521430f6", size = 191726, upload-time = "2025-09-08T23:23:41.742Z" },
+    { url = "https://files.pythonhosted.org/packages/ae/3a/dbeec9d1ee0844c679f6bb5d6ad4e9f198b1224f4e7a32825f47f6192b0c/cffi-2.0.0-cp314-cp314t-win_arm64.whl", hash = "sha256:0a1527a803f0a659de1af2e1fd700213caba79377e27e4693648c2923da066f9", size = 184195, upload-time = "2025-09-08T23:23:43.004Z" },
+]
+
 [[package]]
 name = "charset-normalizer"
 version = "3.4.6"
@@ -128,6 +173,73 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/d1/d6/3965ed04c63042e047cb6a3e6ed1a63a35087b6a609aa3a15ed8ac56c221/colorama-0.4.6-py2.py3-none-any.whl", hash = "sha256:4f1d9991f5acc0ca119f9d443620b77f9d6b33703e51011c16baf57afb285fc6", size = 25335, upload-time = "2022-10-25T02:36:20.889Z" },
 ]
 
+[[package]]
+name = "cryptography"
+version = "46.0.6"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "cffi", marker = "platform_python_implementation != 'PyPy'" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/a4/ba/04b1bd4218cbc58dc90ce967106d51582371b898690f3ae0402876cc4f34/cryptography-46.0.6.tar.gz", hash = "sha256:27550628a518c5c6c903d84f637fbecf287f6cb9ced3804838a1295dc1fd0759", size = 750542, upload-time = "2026-03-25T23:34:53.396Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/47/23/9285e15e3bc57325b0a72e592921983a701efc1ee8f91c06c5f0235d86d9/cryptography-46.0.6-cp311-abi3-macosx_10_9_universal2.whl", hash = "sha256:64235194bad039a10bb6d2d930ab3323baaec67e2ce36215fd0952fad0930ca8", size = 7176401, upload-time = "2026-03-25T23:33:22.096Z" },
+    { url = "https://files.pythonhosted.org/packages/60/f8/e61f8f13950ab6195b31913b42d39f0f9afc7d93f76710f299b5ec286ae6/cryptography-46.0.6-cp311-abi3-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:26031f1e5ca62fcb9d1fcb34b2b60b390d1aacaa15dc8b895a9ed00968b97b30", size = 4275275, upload-time = "2026-03-25T23:33:23.844Z" },
+    { url = "https://files.pythonhosted.org/packages/19/69/732a736d12c2631e140be2348b4ad3d226302df63ef64d30dfdb8db7ad1c/cryptography-46.0.6-cp311-abi3-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:9a693028b9cbe51b5a1136232ee8f2bc242e4e19d456ded3fa7c86e43c713b4a", size = 4425320, upload-time = "2026-03-25T23:33:25.703Z" },
+    { url = "https://files.pythonhosted.org/packages/d4/12/123be7292674abf76b21ac1fc0e1af50661f0e5b8f0ec8285faac18eb99e/cryptography-46.0.6-cp311-abi3-manylinux_2_28_aarch64.whl", hash = "sha256:67177e8a9f421aa2d3a170c3e56eca4e0128883cf52a071a7cbf53297f18b175", size = 4278082, upload-time = "2026-03-25T23:33:27.423Z" },
+    { url = "https://files.pythonhosted.org/packages/5b/ba/d5e27f8d68c24951b0a484924a84c7cdaed7502bac9f18601cd357f8b1d2/cryptography-46.0.6-cp311-abi3-manylinux_2_28_ppc64le.whl", hash = "sha256:d9528b535a6c4f8ff37847144b8986a9a143585f0540fbcb1a98115b543aa463", size = 4926514, upload-time = "2026-03-25T23:33:29.206Z" },
+    { url = "https://files.pythonhosted.org/packages/34/71/1ea5a7352ae516d5512d17babe7e1b87d9db5150b21f794b1377eac1edc0/cryptography-46.0.6-cp311-abi3-manylinux_2_28_x86_64.whl", hash = "sha256:22259338084d6ae497a19bae5d4c66b7ca1387d3264d1c2c0e72d9e9b6a77b97", size = 4457766, upload-time = "2026-03-25T23:33:30.834Z" },
+    { url = "https://files.pythonhosted.org/packages/01/59/562be1e653accee4fdad92c7a2e88fced26b3fdfce144047519bbebc299e/cryptography-46.0.6-cp311-abi3-manylinux_2_31_armv7l.whl", hash = "sha256:760997a4b950ff00d418398ad73fbc91aa2894b5c1db7ccb45b4f68b42a63b3c", size = 3986535, upload-time = "2026-03-25T23:33:33.02Z" },
+    { url = "https://files.pythonhosted.org/packages/d6/8b/b1ebfeb788bf4624d36e45ed2662b8bd43a05ff62157093c1539c1288a18/cryptography-46.0.6-cp311-abi3-manylinux_2_34_aarch64.whl", hash = "sha256:3dfa6567f2e9e4c5dceb8ccb5a708158a2a871052fa75c8b78cb0977063f1507", size = 4277618, upload-time = "2026-03-25T23:33:34.567Z" },
+    { url = "https://files.pythonhosted.org/packages/dd/52/a005f8eabdb28df57c20f84c44d397a755782d6ff6d455f05baa2785bd91/cryptography-46.0.6-cp311-abi3-manylinux_2_34_ppc64le.whl", hash = "sha256:cdcd3edcbc5d55757e5f5f3d330dd00007ae463a7e7aa5bf132d1f22a4b62b19", size = 4890802, upload-time = "2026-03-25T23:33:37.034Z" },
+    { url = "https://files.pythonhosted.org/packages/ec/4d/8e7d7245c79c617d08724e2efa397737715ca0ec830ecb3c91e547302555/cryptography-46.0.6-cp311-abi3-manylinux_2_34_x86_64.whl", hash = "sha256:d4e4aadb7fc1f88687f47ca20bb7227981b03afaae69287029da08096853b738", size = 4457425, upload-time = "2026-03-25T23:33:38.904Z" },
+    { url = "https://files.pythonhosted.org/packages/1d/5c/f6c3596a1430cec6f949085f0e1a970638d76f81c3ea56d93d564d04c340/cryptography-46.0.6-cp311-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:2b417edbe8877cda9022dde3a008e2deb50be9c407eef034aeeb3a8b11d9db3c", size = 4405530, upload-time = "2026-03-25T23:33:40.842Z" },
+    { url = "https://files.pythonhosted.org/packages/7e/c9/9f9cea13ee2dbde070424e0c4f621c091a91ffcc504ffea5e74f0e1daeff/cryptography-46.0.6-cp311-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:380343e0653b1c9d7e1f55b52aaa2dbb2fdf2730088d48c43ca1c7c0abb7cc2f", size = 4667896, upload-time = "2026-03-25T23:33:42.781Z" },
+    { url = "https://files.pythonhosted.org/packages/ad/b5/1895bc0821226f129bc74d00eccfc6a5969e2028f8617c09790bf89c185e/cryptography-46.0.6-cp311-abi3-win32.whl", hash = "sha256:bcb87663e1f7b075e48c3be3ecb5f0b46c8fc50b50a97cf264e7f60242dca3f2", size = 3026348, upload-time = "2026-03-25T23:33:45.021Z" },
+    { url = "https://files.pythonhosted.org/packages/c3/f8/c9bcbf0d3e6ad288b9d9aa0b1dee04b063d19e8c4f871855a03ab3a297ab/cryptography-46.0.6-cp311-abi3-win_amd64.whl", hash = "sha256:6739d56300662c468fddb0e5e291f9b4d084bead381667b9e654c7dd81705124", size = 3483896, upload-time = "2026-03-25T23:33:46.649Z" },
+    { url = "https://files.pythonhosted.org/packages/01/41/3a578f7fd5c70611c0aacba52cd13cb364a5dee895a5c1d467208a9380b0/cryptography-46.0.6-cp314-cp314t-macosx_10_9_universal2.whl", hash = "sha256:2ef9e69886cbb137c2aef9772c2e7138dc581fad4fcbcf13cc181eb5a3ab6275", size = 7117147, upload-time = "2026-03-25T23:33:48.249Z" },
+    { url = "https://files.pythonhosted.org/packages/fa/87/887f35a6fca9dde90cad08e0de0c89263a8e59b2d2ff904fd9fcd8025b6f/cryptography-46.0.6-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:7f417f034f91dcec1cb6c5c35b07cdbb2ef262557f701b4ecd803ee8cefed4f4", size = 4266221, upload-time = "2026-03-25T23:33:49.874Z" },
+    { url = "https://files.pythonhosted.org/packages/aa/a8/0a90c4f0b0871e0e3d1ed126aed101328a8a57fd9fd17f00fb67e82a51ca/cryptography-46.0.6-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:d24c13369e856b94892a89ddf70b332e0b70ad4a5c43cf3e9cb71d6d7ffa1f7b", size = 4408952, upload-time = "2026-03-25T23:33:52.128Z" },
+    { url = "https://files.pythonhosted.org/packages/16/0b/b239701eb946523e4e9f329336e4ff32b1247e109cbab32d1a7b61da8ed7/cryptography-46.0.6-cp314-cp314t-manylinux_2_28_aarch64.whl", hash = "sha256:aad75154a7ac9039936d50cf431719a2f8d4ed3d3c277ac03f3339ded1a5e707", size = 4270141, upload-time = "2026-03-25T23:33:54.11Z" },
+    { url = "https://files.pythonhosted.org/packages/0f/a8/976acdd4f0f30df7b25605f4b9d3d89295351665c2091d18224f7ad5cdbf/cryptography-46.0.6-cp314-cp314t-manylinux_2_28_ppc64le.whl", hash = "sha256:3c21d92ed15e9cfc6eb64c1f5a0326db22ca9c2566ca46d845119b45b4400361", size = 4904178, upload-time = "2026-03-25T23:33:55.725Z" },
+    { url = "https://files.pythonhosted.org/packages/b1/1b/bf0e01a88efd0e59679b69f42d4afd5bced8700bb5e80617b2d63a3741af/cryptography-46.0.6-cp314-cp314t-manylinux_2_28_x86_64.whl", hash = "sha256:4668298aef7cddeaf5c6ecc244c2302a2b8e40f384255505c22875eebb47888b", size = 4441812, upload-time = "2026-03-25T23:33:57.364Z" },
+    { url = "https://files.pythonhosted.org/packages/bb/8b/11df86de2ea389c65aa1806f331cae145f2ed18011f30234cc10ca253de8/cryptography-46.0.6-cp314-cp314t-manylinux_2_31_armv7l.whl", hash = "sha256:8ce35b77aaf02f3b59c90b2c8a05c73bac12cea5b4e8f3fbece1f5fddea5f0ca", size = 3963923, upload-time = "2026-03-25T23:33:59.361Z" },
+    { url = "https://files.pythonhosted.org/packages/91/e0/207fb177c3a9ef6a8108f234208c3e9e76a6aa8cf20d51932916bd43bda0/cryptography-46.0.6-cp314-cp314t-manylinux_2_34_aarch64.whl", hash = "sha256:c89eb37fae9216985d8734c1afd172ba4927f5a05cfd9bf0e4863c6d5465b013", size = 4269695, upload-time = "2026-03-25T23:34:00.909Z" },
+    { url = "https://files.pythonhosted.org/packages/21/5e/19f3260ed1e95bced52ace7501fabcd266df67077eeb382b79c81729d2d3/cryptography-46.0.6-cp314-cp314t-manylinux_2_34_ppc64le.whl", hash = "sha256:ed418c37d095aeddf5336898a132fba01091f0ac5844e3e8018506f014b6d2c4", size = 4869785, upload-time = "2026-03-25T23:34:02.796Z" },
+    { url = "https://files.pythonhosted.org/packages/10/38/cd7864d79aa1d92ef6f1a584281433419b955ad5a5ba8d1eb6c872165bcb/cryptography-46.0.6-cp314-cp314t-manylinux_2_34_x86_64.whl", hash = "sha256:69cf0056d6947edc6e6760e5f17afe4bea06b56a9ac8a06de9d2bd6b532d4f3a", size = 4441404, upload-time = "2026-03-25T23:34:04.35Z" },
+    { url = "https://files.pythonhosted.org/packages/09/0a/4fe7a8d25fed74419f91835cf5829ade6408fd1963c9eae9c4bce390ecbb/cryptography-46.0.6-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:8e7304c4f4e9490e11efe56af6713983460ee0780f16c63f219984dab3af9d2d", size = 4397549, upload-time = "2026-03-25T23:34:06.342Z" },
+    { url = "https://files.pythonhosted.org/packages/5f/a0/7d738944eac6513cd60a8da98b65951f4a3b279b93479a7e8926d9cd730b/cryptography-46.0.6-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:b928a3ca837c77a10e81a814a693f2295200adb3352395fad024559b7be7a736", size = 4651874, upload-time = "2026-03-25T23:34:07.916Z" },
+    { url = "https://files.pythonhosted.org/packages/cb/f1/c2326781ca05208845efca38bf714f76939ae446cd492d7613808badedf1/cryptography-46.0.6-cp314-cp314t-win32.whl", hash = "sha256:97c8115b27e19e592a05c45d0dd89c57f81f841cc9880e353e0d3bf25b2139ed", size = 3001511, upload-time = "2026-03-25T23:34:09.892Z" },
+    { url = "https://files.pythonhosted.org/packages/c9/57/fe4a23eb549ac9d903bd4698ffda13383808ef0876cc912bcb2838799ece/cryptography-46.0.6-cp314-cp314t-win_amd64.whl", hash = "sha256:c797e2517cb7880f8297e2c0f43bb910e91381339336f75d2c1c2cbf811b70b4", size = 3471692, upload-time = "2026-03-25T23:34:11.613Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/cc/f330e982852403da79008552de9906804568ae9230da8432f7496ce02b71/cryptography-46.0.6-cp38-abi3-macosx_10_9_universal2.whl", hash = "sha256:12cae594e9473bca1a7aceb90536060643128bb274fcea0fc459ab90f7d1ae7a", size = 7162776, upload-time = "2026-03-25T23:34:13.308Z" },
+    { url = "https://files.pythonhosted.org/packages/49/b3/dc27efd8dcc4bff583b3f01d4a3943cd8b5821777a58b3a6a5f054d61b79/cryptography-46.0.6-cp38-abi3-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:639301950939d844a9e1c4464d7e07f902fe9a7f6b215bb0d4f28584729935d8", size = 4270529, upload-time = "2026-03-25T23:34:15.019Z" },
+    { url = "https://files.pythonhosted.org/packages/e6/05/e8d0e6eb4f0d83365b3cb0e00eb3c484f7348db0266652ccd84632a3d58d/cryptography-46.0.6-cp38-abi3-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:ed3775295fb91f70b4027aeba878d79b3e55c0b3e97eaa4de71f8f23a9f2eb77", size = 4414827, upload-time = "2026-03-25T23:34:16.604Z" },
+    { url = "https://files.pythonhosted.org/packages/2f/97/daba0f5d2dc6d855e2dcb70733c812558a7977a55dd4a6722756628c44d1/cryptography-46.0.6-cp38-abi3-manylinux_2_28_aarch64.whl", hash = "sha256:8927ccfbe967c7df312ade694f987e7e9e22b2425976ddbf28271d7e58845290", size = 4271265, upload-time = "2026-03-25T23:34:18.586Z" },
+    { url = "https://files.pythonhosted.org/packages/89/06/fe1fce39a37ac452e58d04b43b0855261dac320a2ebf8f5260dd55b201a9/cryptography-46.0.6-cp38-abi3-manylinux_2_28_ppc64le.whl", hash = "sha256:b12c6b1e1651e42ab5de8b1e00dc3b6354fdfd778e7fa60541ddacc27cd21410", size = 4916800, upload-time = "2026-03-25T23:34:20.561Z" },
+    { url = "https://files.pythonhosted.org/packages/ff/8a/b14f3101fe9c3592603339eb5d94046c3ce5f7fc76d6512a2d40efd9724e/cryptography-46.0.6-cp38-abi3-manylinux_2_28_x86_64.whl", hash = "sha256:063b67749f338ca9c5a0b7fe438a52c25f9526b851e24e6c9310e7195aad3b4d", size = 4448771, upload-time = "2026-03-25T23:34:22.406Z" },
+    { url = "https://files.pythonhosted.org/packages/01/b3/0796998056a66d1973fd52ee89dc1bb3b6581960a91ad4ac705f182d398f/cryptography-46.0.6-cp38-abi3-manylinux_2_31_armv7l.whl", hash = "sha256:02fad249cb0e090b574e30b276a3da6a149e04ee2f049725b1f69e7b8351ec70", size = 3978333, upload-time = "2026-03-25T23:34:24.281Z" },
+    { url = "https://files.pythonhosted.org/packages/c5/3d/db200af5a4ffd08918cd55c08399dc6c9c50b0bc72c00a3246e099d3a849/cryptography-46.0.6-cp38-abi3-manylinux_2_34_aarch64.whl", hash = "sha256:7e6142674f2a9291463e5e150090b95a8519b2fb6e6aaec8917dd8d094ce750d", size = 4271069, upload-time = "2026-03-25T23:34:25.895Z" },
+    { url = "https://files.pythonhosted.org/packages/d7/18/61acfd5b414309d74ee838be321c636fe71815436f53c9f0334bf19064fa/cryptography-46.0.6-cp38-abi3-manylinux_2_34_ppc64le.whl", hash = "sha256:456b3215172aeefb9284550b162801d62f5f264a081049a3e94307fe20792cfa", size = 4878358, upload-time = "2026-03-25T23:34:27.67Z" },
+    { url = "https://files.pythonhosted.org/packages/8b/65/5bf43286d566f8171917cae23ac6add941654ccf085d739195a4eacf1674/cryptography-46.0.6-cp38-abi3-manylinux_2_34_x86_64.whl", hash = "sha256:341359d6c9e68834e204ceaf25936dffeafea3829ab80e9503860dcc4f4dac58", size = 4448061, upload-time = "2026-03-25T23:34:29.375Z" },
+    { url = "https://files.pythonhosted.org/packages/e0/25/7e49c0fa7205cf3597e525d156a6bce5b5c9de1fd7e8cb01120e459f205a/cryptography-46.0.6-cp38-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:9a9c42a2723999a710445bc0d974e345c32adfd8d2fac6d8a251fa829ad31cfb", size = 4399103, upload-time = "2026-03-25T23:34:32.036Z" },
+    { url = "https://files.pythonhosted.org/packages/44/46/466269e833f1c4718d6cd496ffe20c56c9c8d013486ff66b4f69c302a68d/cryptography-46.0.6-cp38-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:6617f67b1606dfd9fe4dbfa354a9508d4a6d37afe30306fe6c101b7ce3274b72", size = 4659255, upload-time = "2026-03-25T23:34:33.679Z" },
+    { url = "https://files.pythonhosted.org/packages/0a/09/ddc5f630cc32287d2c953fc5d32705e63ec73e37308e5120955316f53827/cryptography-46.0.6-cp38-abi3-win32.whl", hash = "sha256:7f6690b6c55e9c5332c0b59b9c8a3fb232ebf059094c17f9019a51e9827df91c", size = 3010660, upload-time = "2026-03-25T23:34:35.418Z" },
+    { url = "https://files.pythonhosted.org/packages/1b/82/ca4893968aeb2709aacfb57a30dec6fa2ab25b10fa9f064b8882ce33f599/cryptography-46.0.6-cp38-abi3-win_amd64.whl", hash = "sha256:79e865c642cfc5c0b3eb12af83c35c5aeff4fa5c672dc28c43721c2c9fdd2f0f", size = 3471160, upload-time = "2026-03-25T23:34:37.191Z" },
+]
+
+[[package]]
+name = "docker"
+version = "7.1.0"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "pywin32", marker = "sys_platform == 'win32'" },
+    { name = "requests" },
+    { name = "urllib3" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/91/9b/4a2ea29aeba62471211598dac5d96825bb49348fa07e906ea930394a83ce/docker-7.1.0.tar.gz", hash = "sha256:ad8c70e6e3f8926cb8a92619b832b4ea5299e2831c14284663184e200546fa6c", size = 117834, upload-time = "2024-05-23T11:13:57.216Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/e3/26/57c6fb270950d476074c087527a558ccb6f4436657314bfb6cdf484114c4/docker-7.1.0-py3-none-any.whl", hash = "sha256:c96b93b7f0a746f9e77d325bcfb87422a3d8bd4f03136ae8a85b37f1898d5fc0", size = 147774, upload-time = "2024-05-23T11:13:55.01Z" },
+]
+
 [[package]]
 name = "fastapi"
 version = "0.135.1"
@@ -247,6 +359,7 @@ name = "master"
 version = "0.0.1"
 source = { virtual = "." }
 dependencies = [
+    { name = "docker" },
     { name = "fastapi" },
     { name = "opentelemetry-api" },
     { name = "opentelemetry-exporter-otlp-proto-http" },
@@ -262,11 +375,13 @@ dev = [
     { name = "mypy" },
     { name = "pytest" },
     { name = "ruff" },
+    { name = "types-docker" },
     { name = "types-pyyaml" },
 ]
 
 [package.metadata]
 requires-dist = [
+    { name = "docker", specifier = ">=7.1.0" },
     { name = "fastapi", specifier = ">=0.116.1" },
     { name = "opentelemetry-api", specifier = ">=1.31.1" },
     { name = "opentelemetry-exporter-otlp-proto-http", specifier = ">=1.31.1" },
@@ -282,6 +397,7 @@ dev = [
     { name = "mypy", specifier = ">=1.18.2" },
     { name = "pytest", specifier = ">=8.4.2" },
     { name = "ruff", specifier = ">=0.13.1" },
+    { name = "types-docker", specifier = ">=7.1.0.20260402" },
     { name = "types-pyyaml", specifier = ">=6.0.12.20250915" },
 ]
 
@@ -501,6 +617,15 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/c4/72/02445137af02769918a93807b2b7890047c32bfb9f90371cbc12688819eb/protobuf-6.33.6-py3-none-any.whl", hash = "sha256:77179e006c476e69bf8e8ce866640091ec42e1beb80b213c3900006ecfba6901", size = 170656, upload-time = "2026-03-18T19:04:59.826Z" },
 ]
 
+[[package]]
+name = "pycparser"
+version = "3.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/1b/7d/92392ff7815c21062bea51aa7b87d45576f649f16458d78b7cf94b9ab2e6/pycparser-3.0.tar.gz", hash = "sha256:600f49d217304a5902ac3c37e1281c9fe94e4d0489de643a9504c5cdfdfc6b29", size = 103492, upload-time = "2026-01-21T14:26:51.89Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/0c/c3/44f3fbbfa403ea2a7c779186dc20772604442dde72947e7d01069cbe98e3/pycparser-3.0-py3-none-any.whl", hash = "sha256:b727414169a36b7d524c1c3e31839a521725078d7b2ff038656844266160a992", size = 48172, upload-time = "2026-01-21T14:26:50.693Z" },
+]
+
 [[package]]
 name = "pydantic"
 version = "2.12.5"
@@ -603,6 +728,19 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/0b/d7/1959b9648791274998a9c3526f6d0ec8fd2233e4d4acce81bbae76b44b2a/python_dotenv-1.2.2-py3-none-any.whl", hash = "sha256:1d8214789a24de455a8b8bd8ae6fe3c6b69a5e3d64aa8a8e5d68e694bbcb285a", size = 22101, upload-time = "2026-03-01T16:00:25.09Z" },
 ]
 
+[[package]]
+name = "pywin32"
+version = "311"
+source = { registry = "https://pypi.org/simple" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/a5/be/3fd5de0979fcb3994bfee0d65ed8ca9506a8a1260651b86174f6a86f52b3/pywin32-311-cp313-cp313-win32.whl", hash = "sha256:f95ba5a847cba10dd8c4d8fefa9f2a6cf283b8b88ed6178fa8a6c1ab16054d0d", size = 8705700, upload-time = "2025-07-14T20:13:26.471Z" },
+    { url = "https://files.pythonhosted.org/packages/e3/28/e0a1909523c6890208295a29e05c2adb2126364e289826c0a8bc7297bd5c/pywin32-311-cp313-cp313-win_amd64.whl", hash = "sha256:718a38f7e5b058e76aee1c56ddd06908116d35147e133427e59a3983f703a20d", size = 9494700, upload-time = "2025-07-14T20:13:28.243Z" },
+    { url = "https://files.pythonhosted.org/packages/04/bf/90339ac0f55726dce7d794e6d79a18a91265bdf3aa70b6b9ca52f35e022a/pywin32-311-cp313-cp313-win_arm64.whl", hash = "sha256:7b4075d959648406202d92a2310cb990fea19b535c7f4a78d3f5e10b926eeb8a", size = 8709318, upload-time = "2025-07-14T20:13:30.348Z" },
+    { url = "https://files.pythonhosted.org/packages/c9/31/097f2e132c4f16d99a22bfb777e0fd88bd8e1c634304e102f313af69ace5/pywin32-311-cp314-cp314-win32.whl", hash = "sha256:b7a2c10b93f8986666d0c803ee19b5990885872a7de910fc460f9b0c2fbf92ee", size = 8840714, upload-time = "2025-07-14T20:13:32.449Z" },
+    { url = "https://files.pythonhosted.org/packages/90/4b/07c77d8ba0e01349358082713400435347df8426208171ce297da32c313d/pywin32-311-cp314-cp314-win_amd64.whl", hash = "sha256:3aca44c046bd2ed8c90de9cb8427f581c479e594e99b5c0bb19b29c10fd6cb87", size = 9656800, upload-time = "2025-07-14T20:13:34.312Z" },
+    { url = "https://files.pythonhosted.org/packages/c0/d2/21af5c535501a7233e734b8af901574572da66fcc254cb35d0609c9080dd/pywin32-311-cp314-cp314-win_arm64.whl", hash = "sha256:a508e2d9025764a8270f93111a970e1d0fbfc33f4153b388bb649b7eec4f9b42", size = 8932540, upload-time = "2025-07-14T20:13:36.379Z" },
+]
+
 [[package]]
 name = "pyyaml"
 version = "6.0.3"
@@ -691,6 +829,32 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/81/0d/13d1d239a25cbfb19e740db83143e95c772a1fe10202dda4b76792b114dd/starlette-0.52.1-py3-none-any.whl", hash = "sha256:0029d43eb3d273bc4f83a08720b4912ea4b071087a3b48db01b7c839f7954d74", size = 74272, upload-time = "2026-01-18T13:34:09.188Z" },
 ]
 
+[[package]]
+name = "types-docker"
+version = "7.1.0.20260402"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "types-paramiko" },
+    { name = "types-requests" },
+    { name = "urllib3" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/1f/b6/e9dd7b51f5db0df219aed496e5f1fe4e4e17828cd9c354fc6a98a4454ea1/types_docker-7.1.0.20260402.tar.gz", hash = "sha256:2df72c6268a815f4ba28fe2556072a99b8acffe46370fa4d27c485355af2e37e", size = 32974, upload-time = "2026-04-02T04:22:54.318Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/90/8c/5056b3f6ba818dc9f677f9e7aa9044113657a8b67871ed30921d3712b2f2/types_docker-7.1.0.20260402-py3-none-any.whl", hash = "sha256:88055c13ad43a3d13415db42d180193e812e6a25d0b50710d6559e86cb9345e5", size = 47455, upload-time = "2026-04-02T04:22:53.426Z" },
+]
+
+[[package]]
+name = "types-paramiko"
+version = "4.0.0.20260402"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "cryptography" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/c6/91/63f4fa68a7d563ca54f62c013aeefb480a2e18a0ee0623a56119f1662bf2/types_paramiko-4.0.0.20260402.tar.gz", hash = "sha256:a9287bdb78cb67c8e79897dc98b761900968cd2de288f72cc298c1a25cde6a38", size = 29105, upload-time = "2026-04-02T04:20:26.809Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/8d/5e/d650556433e9b745fb85676f4382b7ec0f35156696d946b9bf5ac470358b/types_paramiko-4.0.0.20260402-py3-none-any.whl", hash = "sha256:38ef646f54d5410012d8607b9f023a355ecaec12b749d0325bce5f16706f6183", size = 38816, upload-time = "2026-04-02T04:20:25.788Z" },
+]
+
 [[package]]
 name = "types-pyyaml"
 version = "6.0.12.20250915"
@@ -700,6 +864,18 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/bd/e0/1eed384f02555dde685fff1a1ac805c1c7dcb6dd019c916fe659b1c1f9ec/types_pyyaml-6.0.12.20250915-py3-none-any.whl", hash = "sha256:e7d4d9e064e89a3b3cae120b4990cd370874d2bf12fa5f46c97018dd5d3c9ab6", size = 20338, upload-time = "2025-09-15T03:00:59.218Z" },
 ]
 
+[[package]]
+name = "types-requests"
+version = "2.33.0.20260402"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "urllib3" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/c1/7b/a06527d20af1441d813360b8e0ce152a75b7d8e4aab7c7d0a156f405d7ec/types_requests-2.33.0.20260402.tar.gz", hash = "sha256:1bdd3ada9b869741c5c4b887d2c8b4e38284a1449751823b5ebbccba3eefd9da", size = 23851, upload-time = "2026-04-02T04:19:55.942Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/51/65/3853bb6bac5ae789dc7e28781154705c27859eccc8e46282c3f36780f5f5/types_requests-2.33.0.20260402-py3-none-any.whl", hash = "sha256:c98372d7124dd5d10af815ee25c013897592ff92af27b27e22c98984102c3254", size = 20739, upload-time = "2026-04-02T04:19:54.955Z" },
+]
+
 [[package]]
 name = "typing-extensions"
 version = "4.15.0"