Przejdź do treści
Biblioteka skilli
AI i agenci · Automatyzacja

self-improve

/self-improve

Uruchamia sesję samodoskonalenia ChaosEngine: lekcje o środowisku pracy trafiają do zgłoszeń GitHub, a produktowe do kolejki.

Repozytorium
shaft_engine ↗
Licencja
MIT
Ostatnia zmiana repozytorium
26.10.2026

Ocena i bezpieczeństwo

100/ 100Klasa: dobry
Format25 / 25
Frontmatter z nazwą i opisem, poprawna nazwa, opis do 1024 znaków, nazwa zgodna z folderem.
Odkrywalność25 / 25
Opis mówi, kiedy użyć skilla, ma sensowną długość i jest w 3. osobie.
Treść35 / 35
Co najmniej 150 słów i do 500 linii, nagłówki, listy kroków, przykłady albo kod.
Higiena15 / 15
Bez zaślepek typu TODO, bez ścieżek Windows, z licencją.
Ryzykobrak sygnałów w skanie
Pliki w paczce
7
Skrypty
0
Komentarze HTML
0
Adresy URL w SKILL.md
0
Przestarzałość

Brak sygnałów. Repozytorium zmienione 26.10.2026, bez odwołań do wycofanych modeli i API.

Licencja

MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.

Co robi i kiedy

Uruchamia sesję samodoskonalenia ChaosEngine: lekcje o środowisku pracy trafiają do zgłoszeń GitHub, a produktowe do kolejki.

Oryginalny opis (EN)

Use when running ChaosEngine Learning Session self-improve. Harness lessons are GitHub issues only. Product lessons may queue after delivery or on request.

Jak działa

  1. When
  2. Adopter instructions and upstream tickets
  3. Dual track
  4. How (wraps learning.py)
  5. Metrics / verify (zero-LLM)
  6. Local smoke
  7. Retrospective
Sekcje z nagłówków SKILL.md, w oryginale · 140 linii, 850 słów
Podgląd SKILL.md
---
name: self-improve
description: >-
  Use when running ChaosEngine Learning Session self-improve. Harness lessons
  are GitHub issues only. Product lessons may queue after delivery or on request.
---

# self-improve — ChaosEngine learning & adapting

Lean CE-native skill informed by Task Observer methodology
(Eoghan Henn / rebelytics, **CC BY 4.0** — see [LICENSE](LICENSE) and
[UPSTREAM.md](UPSTREAM.md)). Not a blind clone.

## When

- **Primary:** root-owned Learning Session after every final delivery (owner approval, publish done, final PR merged), automatic: lessons diff against the skills, one spec issue, ONE PR (`skip-release-notes`, auto-merge MERGE) babysat to merged.
- **Secondary:** explicit operator request mid-session.
- **Not:** every casual turn, and never the Learning Session PR's own merge.
- A lesson already shipped in the delivery pull request is not a new issue. Finalize with nothing durable.

If `assess` refuses `novel_success` under the one-incident rule, still
finalize: file or update the durable lesson tickets, then complete finalize
without inventing a second incident for the same success signal.

## Adopter instructions and upstream tickets

A project lesson updates that adopter's specialized instructions (pack,
profile, or project rules) for token use, accuracy, or performance in the
learning change. A harness lesson is redacted and filed on the configured
upstream (`scripts/ci/chaos-engine-upstream.txt` or
`CHAOS_ENGINE_UPSTREAM_REPO`) with the user's `gh` account only after machine
consent. `python3 .chaos-engine/learning_session.py consent allow` records
the standing approval. `consent deny` stops it. Unknown consent asks once,
records pending, and does not file again until the user answers.

## Dual track

1. **Harness** — skills, hooks, MemPalace, Graphify, installer/doctor. GitHub
   issue only. Never a local queue and never chat.
2. **Product** — enhancements for the product under development (queued issues).

Details: [references/observation-taxonomy.md](references/observation-taxonomy.md).
Product track: [references/product-track.md](references/product-track.md)
(product issues/evals + ChaosGauge link; CLI/doctor/silent-verify over essay tickets).
Activation: [references/activation.md](references/activation.md).
Adopt/reject research: [references/research-adopt-reject.md](references/research-adopt-reject.md).
Roadmap: [../../references/self-improve-master-plan.md](../../references/self-improve-master-plan.md).
Research/explore isolation (harness): [../../references/context-firewall.md](../../references/context-firewall.md).

## How (wraps learning.py)

1. Classify each finding (harness vs product; category allow-list).
2. Write minimal fields only: `category`, `title`, `lesson`, `proposedChange`,
   `benefit`, `estimatedTokens`.
3. File a harness lesson as a GitHub issue. Queue only a product lesson
   through `learning.py --track product`.
4. Never auto-install skill patches; stage proposals for human/CI review.
5. "Nothing durable" is a valid outcome only when no actionable retrospective line remains. File that line remotely first. The chat note is not the record.
6. A lessons pull request lands only when it states the concrete outcome the
   lesson would have changed: time, a defect, or a failure avoided. The body
   has a `## Outcome changed` section with that statement. A title starting
   with `Lesson` or `Lessons` is a lessons PR. The PR body gate rejects one
   that lacks the section. A candidate that fails the bar is one append-only
   line in `.chaos-engine-state/lesson-notes.log` (`record_lesson_note`), not
   a pull request. Zero-trigger rule retirement is
   [rule-retirement](../../references/rule-retirement.md).
7. Refine this protocol when traces show skipped Learning Sessions, local
   writers reviewing their own diffs, or queue-only lessons that never became
   issues. Self-development has no cap.

Example:

A harness example is a GitHub issue, not a queue file. A product example:

```bash
python3 .chaos-engine/learning.py queue \
  --track product \
  --state .chaos-engine-state/learning \
  --upstream Owner/ExampleRepo \
  --candidate .chaos-engine-state/learning-candidate.json
```

## Metrics / verify (zero-LLM)

```bash
python3 .chaos-engine/learning.py metrics
python3 .chaos-engine/silent_verify.py session-start-budget
python3 .chaos-engine/retrieve.py heuristics --top 3
python3 .chaos-engine/significance.py list
python3 .chaos-engine/skill_compress_audit.py audit --skill self-improve
python3 .chaos-engine/meta_optimize.py review
python3 .chaos-engine/draft_skill_pr.py status
```

SessionStart stays locator-only; prefer CLI over MCP for the same job.
Mid-session: mark **significant** friction only via `significance.py` (soft
fail/deny hooks already write tiny state notes) — never Task Observer.
Skill bodies: `skill_compress_audit.py` proposes compress diffs only; never auto-apply.
Periodic: `meta_optimize.py review` (offline cadence — not continuous).
Draft skill PRs: `draft_skill_pr.py` opt-in only (default OFF; never auto-merge).
That covers the automated drafter only; the Learning Session's own lessons PR
runs the normal PR gates and auto-merges.
See [meta-optimize](../../references/meta-optimize.md) and
[draft-skill-pr](../../references/draft-skill-pr.md).

## Local smoke

```bash
# From a temp project with ChaosEngine installed, or the source tree:
python3 -c "from pathlib import Path; assert Path('chaos-engine/skills/self-improve/SKILL.md').is_file()"
# Confirm learning.py queue --track harness exits non-zero and writes no queue.json.
```


## Retrospective

One closing note. Derby and Larsen: gather what happened, say why, then decide.
Start-Stop-Continue is the decision: each line is a behavior, and the note ends
with one next action.

- what happened, with the external proof
- why it happened
- continue doing
- stop doing
- start doing
- next action: one change, or nothing actionable

An actionable line would change a later session and is not already shipped on
the default branch or already an open issue. File it with `gh` before the note,
then cite the issue URL on that line. Cite an existing issue instead of opening
a duplicate.

- **Upstream:** a ChaosEngine change (skills, hooks, installer, roles, identity,
  companions, or overlay references). File it on the configured upstream
  repository (`scripts/ci/chaos-engine-upstream.txt` or
  `CHAOS_ENGINE_UPSTREAM_REPO`).
- **Private:** a change only for the project that uses ChaosEngine (product
  code, project rules, or adopter workflow). File it on that project's
  repository.

`nothing actionable` is the only close with no URL. Local memory, a lesson
note, and this chat note are not tracking. Do not estimate tokens, model
price, or currency in this note. A host meter and an empty local ledger are
not the same number. `session_token_usage.py` stays the ledger for deja
measurements and local dispatch records. Finalize does not attach it.
Pliki w paczce (7)
  • LICENSE 1 KB
  • SKILL.md 7 KB
  • UPSTREAM.md 1 KB
  • references/activation.md 3 KB
  • references/observation-taxonomy.md 2 KB
  • references/product-track.md 3 KB
  • references/research-adopt-reject.md 2 KB

Wersje

RepozytoriumWynikLicencja
Ta strona · shafthq/shaft_engine100MIT
zereight/gitlab-mcp100MIT
grandamenium/self-improve-skill100MIT
kennykennethkkang/switchboard100Apache-2.0
randomittin/heimdall100MIT
itestflow/itestflow-agent100MIT
nitinjain999/platform-skills98Apache-2.0
Hrutin-Suriya-Kotni/until-hired90MIT
eai-org/agent-toolkit90MIT

Migawka bazy z 13.11.2026 · wersja paczki 5fd0643e1428 (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

  • 90improve
    Stosuje zebrane sygnały samodoskonalenia: czyta refleksje, wprowadza sprawdzone zmiany i aktualizuje lekcje w pliku pamięci.
  • 100meta-improve
    Wdraża zatwierdzone przez operatora kandydatów z Dream w kanonicznym protokole samodoskonalenia, bez samodzielnego zatwierdzania i scalania.
  • 100maestro-improve
    Zamienia zapisane lekcje w najmniejszą zmianę doktryny: grupuje oczekujące lekcje, robi jeden commit z dowodami na cel i zwraca go do niezależnej weryfikacji.
  • 100harness-self-improvement
    Ocenia i poprawia instrukcje oraz skille agenta wyłącznie wtedy, gdy użytkownik wyraźnie o to prosi, bez działania w tle.
  • 90make-skill
    Tworzy, sprawdza i naprawia skille Agent Skills dla GitHub Copilot wraz z ewaluacją w środowisku testowym.
  • 100lessons
    Zapisuje trwałe reguły z korekt i nietypowych błędów do lessons.md, aktualizuje wpisy i awansuje powtarzające się lekcje do skilli.

Uwagi i zgłoszenia

Wiadomość trafi do Bartosza Sokolińskiego. Adres e-mail jest potrzebny tylko do odpowiedzi. Szczegóły w polityce prywatności.