skillopt-sleep

1# SkillOpt-Sleep: usage-driven self-evolution for a local Claude agent
2 
3SkillOpt-Sleep gives the user's agent a **sleep cycle**. On demand or on a
4nightly schedule, it reviews real past Claude Code sessions, re-runs recurring
5tasks through the selected backend, and consolidates what it
6learns into **memory** (`CLAUDE.md`) and **skills** (`SKILL.md`). With the
7default validation gate enabled, it keeps only changes that improve a held-out
8score. Live files change only through explicit adoption or a user-requested
9`--auto-adopt`. It aims to improve this user's recurring work, while making
10each accepted proposal measurable on the run's held-out tasks,
11with no model-weight training. It is the deployment-time analogue of training:
12short-term experience → long-term competence.
13 
14It synthesizes three ideas:
15- **SkillOpt** — the skill/memory doc is trainable text; bounded add/delete/replace
16  edits; accepted only through a held-out gate; rejected edits are recorded in
17  the run report for review.
18- **Claude Dreams** — consolidation that reads past sessions and proposes changes
19  inside protected learned blocks; the input is never mutated, and output is
20  reviewed before adoption.
21- **Agent sleep** — periodic background replay turns episodes into durable skill.
22 
23## When to use this skill
24 
25Trigger when the user wants any of:
26- "make my agent learn from how I use it" / "get better the more I use it" / "remember my preferences across sessions"
27…