skillopt-sleep

1# SkillOpt-Sleep: usage-driven self-evolution for the dsh agent
2 
3SkillOpt-Sleep is Microsoft's [SkillOpt](https://github.com/microsoft/SkillOpt)
4deployment-time companion engine: it reviews your past sessions (harvest), mines
5recurring tasks (mine), replays them through a selected backend (replay), and
6consolidates what it learns into skill documents behind a **held-out validation
7gate** (consolidate).
8 
9This skill drives the engine through the 7 `skillopt_*` tools exposed by the
10dsh-skillopt plugin. The default `mock` backend makes no model calls, which is
11useful for verifying the plumbing; a real backend consumes your API budget.
12 
13## When to use
14 
15- "make my agent better the more I use it" / "learn my preferences across sessions"
16- a one-off **offline self-evolution / sleep / dream** run (immediate or scheduled)
17- review past sessions/trajectories and distill recurring tasks
18- consolidate feedback into `AGENTS.md` / `SKILL.md` / managed skills
19- schedule (cron) the cycle, or adopt a staged proposal
20 
21## The cycle (six stages)
22 
231. **Harvest** — read-only scan of supported local session records → digests
242. **Mine** — digests → recurring task records (intent + outcome labels + checkable refs)
253. **Replay** — re-run tasks under the current skill+memory with the selected backend → (hard, soft) scores
264. **Consolidate** — reflect on failures → propose bounded edits → **validation gate** on a held-out slice (default: accept only on strict improvement)
27…