skillopt-sleep
/skillopt-sleep
Uruchamia Microsoft SkillOpt-Sleep w OMP, uczy się z dawnych sesji OMP, przegląda i przyjmuje przygotowane ulepszenia skilli oraz planuje offline'owy cykl samodoskonalenia.
- Autor
- edmundmiller ↗
- Repozytorium
- dotfiles ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 25.10.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 25.10.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Uruchamia Microsoft SkillOpt-Sleep w OMP, uczy się z dawnych sesji OMP, przegląda i przyjmuje przygotowane ulepszenia skilli oraz planuje offline'owy cykl samodoskonalenia.
Oryginalny opis (EN)
Use when the user wants OMP to run Microsoft SkillOpt-Sleep, learn from past OMP sessions, review or adopt staged skill improvements, or schedule an offline sleep/dream self-improvement cycle.
Jak działa
- When to use
- Mechanism
- Commands
- Recommended workflow
- Safety rules
Podgląd SKILL.md
--- name: skillopt-sleep description: Use when the user wants OMP to run Microsoft SkillOpt-Sleep, learn from past OMP sessions, review or adopt staged skill improvements, or schedule an offline sleep/dream self-improvement cycle. --- # SkillOpt-Sleep for OMP This is the OMP thin shell for Microsoft SkillOpt-Sleep. It follows the upstream Claude Code, Codex, and Devin pattern: harvest local agent sessions, mine recurring tasks, replay them offline, gate proposed memory/skill edits on a held-out score, and stage improvements for review before adoption. ## When to use Use this when the user asks OMP to: - learn from prior OMP sessions or repeated feedback; - run a nightly/offline sleep, dream, or self-improvement cycle; - inspect SkillOpt-Sleep status, harvest, dry-run, run, adopt, schedule, or unschedule; - refine reusable Agent Skills from real OMP usage with validation gates. ## Mechanism The upstream `skillopt_sleep` engine currently knows Claude Code and Codex transcript sources. The bundled wrapper mirrors OMP JSONL sessions into a Claude-compatible, sanitized mirror at `~/.skillopt-sleep/omp/claude-home`, then runs `python -m skillopt_sleep` with `--source claude --claude-home <mirror>`. - Source sessions: `~/.omp/agent/sessions/**/*.jsonl` - Mirror state: `~/.skillopt-sleep/omp/claude-home` - SkillOpt state/staging: `~/.skillopt-sleep/omp/.skillopt-sleep/state.json` and `~/.skillopt-sleep/omp/staging/` - Live changes: none until `adopt`; every adoption is backed up by the engine Tool outputs are not mirrored. The wrapper keeps user/assistant text and assistant tool names, enough for SkillOpt to mine recurring task patterns without copying raw tool results. ## Commands Set `SKILLOPT_SLEEP_REPO` when SkillOpt is cloned somewhere the wrapper cannot auto-detect: ```bash export SKILLOPT_SLEEP_REPO=/path/to/SkillOpt ``` Run from the project whose memory/skills should evolve: ```bash # Safe smoke check, no proposal staged /skillopt-sleep dry-run \ --backend mock --max-sessions 5 --max-tasks 3 --progress # Full cycle; stages a proposal only /skillopt-sleep run \ --backend codex --max-sessions 10 --max-tasks 5 --progress # Inspect staged proposals and history /skillopt-sleep status # Adopt only after explicit review/approval /skillopt-sleep adopt # Nightly wrapper schedule; this installs cron for the OMP wrapper, not the # upstream bare `python -m skillopt_sleep run`. /skillopt-sleep schedule \ --hour 3 --minute 17 --backend codex --max-sessions 10 --max-tasks 5 /skillopt-sleep unschedule ``` Actions are `status`, `harvest`, `dry-run`, `run`, `adopt`, `schedule`, and `unschedule`. `schedule` / `unschedule` are handled by this OMP wrapper so the nightly job refreshes the OMP transcript mirror before invoking SkillOpt-Sleep. Other actions pass through to upstream `python -m skillopt_sleep` with `--source claude` and the mirrored OMP home injected by default. Upstream flags such as `--target-skill-path`, `--tasks-file`, `--backend`, `--model`, `--edit-budget`, `--lookback-hours`, `--max-sessions`, `--max-tasks`, `--progress`, and `--json` pass through unchanged. Slash command usage is `/skillopt-sleep <action> [flags...]`; the extension also exposes the `skillopt_sleep_omp` tool for structured calls. Wrapper-only flags: | Flag | Meaning | | ----------------------- | -------------------------------------------------------------------------------------------- | | `--omp-sessions DIR` | Override the OMP session root. Defaults to `~/.omp/agent/sessions`. | | `--omp-mirror-home DIR` | Override the Claude-compatible mirror home. Defaults to `~/.skillopt-sleep/omp/claude-home`. | | `--help-omp-wrapper` | Show wrapper-specific help. | | `--self-test` | Validate the OMP-to-Claude transcript translator without SkillOpt. | ## Recommended workflow 1. Start with `dry-run --backend mock --json` to validate harvesting and task mining without API spend. 2. For real improvement, use `run --backend codex` or another upstream-supported backend. Keep `--max-sessions`, `--max-tasks`, and budget flags bounded. 3. Read the staged `report.md` and proposed edits before summarizing. 4. Adopt only after explicit user approval. Do not hand-edit memory or skills as a substitute for `adopt`. ## Safety rules - OMP session harvest is read-only; never edit `~/.omp/agent/sessions`. - Keep raw secrets, private transcript contents, and raw tool outputs out of chat, logs, commits, and generated skill text. - Treat generated edits as proposals until the held-out gate and human review approve them. - Prefer the `mock` backend for plumbing checks; real backends spend the user's own agent budget. - If using `--tasks-file` with a real backend, review and mark the task file as reviewed according to upstream SkillOpt-Sleep rules.
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · edmundmiller/dotfiles | 100 | MIT |
| microsoft/SkillOpt | 100 | MIT |
| microsoft/SkillOpt | 100 | MIT |
| pleaseai/claude-code-plugins | 100 | MIT |
| microsoft/SkillOpt | 100 | MIT |
| alirezarezvani/claude-skills | 100 | MIT |
| microsoft/SkillOpt | 98 | MIT |
| microsoft/SkillOpt | 95 | MIT |
Migawka bazy z 13.11.2026 · wersja paczki 89b2b42aea70 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 90skillopt-setup
Instaluje i konfiguruje Microsoft SkillOpt do optymalizacji skilli: podziały ewaluacyjne, Codex CLI i import zmian. - 96omp-long-tasks
Dzieli długie zadania na porcje wielkości pull requesta, gdy uruchomienie omp zbliża się do limitu czasu. - 100omp-delegate
Deleguje zadanie kodowania do Oh My Pi (omp) jako implementera w tle, po czym przegląda jego diff i samodzielnie wprowadza zmiany do repozytorium. - 100omp-cli
Uruchamia agenta OMP w Herdr, przekazuje mu zadanie lub umiejętność, czyta jego wynik i obsługuje sytuację, gdy agent jest zablokowany. - 100sleep
Prowadzi pamięć agenta przez cykl T0, T1 i T2: zapisuje obserwacje, sprawdza kandydatów i awansuje zatwierdzone wpisy. - 100omp-delegate
Przekazuje zadania kodowania do Oh My Pi (omp) tylko na wyraźne żądanie użytkownika, a orkiestrator odpowiada za przegląd i włączenie wyników.