teknium1 d6c710706f docs: add real-world testing findings to OBLITERATUS skill ... Added pitfalls discovered during live abliteration testing: - Models < 1B have fragmented refusal, respond poorly (0.5B: 60%→20%) - Models 3B+ work much better (3B: 75%→0% with advanced defaults) - aggressive method can backfire on small models (made it worse) - Spectral certification RED is common even when refusal rate is 0% - Fixed torch property: total_mem → total_memory		2026-03-09 02:52:54 -07:00
..
references	feat: update OBLITERATUS skill to v2.0 — match current repo state	2026-03-09 02:39:03 -07:00
templates	fix: restore all removed bundled skills + fix skills sync system	2026-03-06 15:57:30 -08:00
SKILL.md	docs: add real-world testing findings to OBLITERATUS skill	2026-03-09 02:52:54 -07:00