retro
/retro
Wyciąga wnioski z zakończonego zgłoszenia po scaleniu przez człowieka, porównuje wynik potoku z tym, co weszło, i koryguje szacunki.
- Autor
- CristianPGit ↗
- Repozytorium
- claude-qa-architect ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 04.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 04.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Wyciąga wnioski z zakończonego zgłoszenia po scaleniu przez człowieka, porównuje wynik potoku z tym, co weszło, i koryguje szacunki.
Wyzwalacze z opisu
Oryginalny opis (EN)
Learn from a finished ticket after the human reviewed/merged it. Use when the user says "retro PROJ-123", "the PR merged", "here's what I changed after the pipeline", or periodically after a few tickets. Compares the pipeline's output with what actually shipped, extracts lessons into the pipeline memory, and calibrates estimates. Read-only on code; writes only to .ticket-work/.
Jak działa
- Steps
- Pipeline memory format (.ticket-work/memory.md)
Podgląd SKILL.md
---
name: retro
description: Learn from a finished ticket after the human reviewed/merged it. Use when the user says "retro PROJ-123", "the PR merged", "here's what I changed after the pipeline", or periodically after a few tickets. Compares the pipeline's output with what actually shipped, extracts lessons into the pipeline memory, and calibrates estimates. Read-only on code; writes only to .ticket-work/.
---
# Ticket Pilot — retro (close the learning loop)
The pipeline's output was reviewed by a human, possibly corrected, and merged.
Every difference between what the pipeline produced and what actually shipped
is a lesson. Capture it so the next run is better.
`$ARGUMENTS`: a ticket key, or empty → offer retro candidates (tickets in
`.ticket-work/` whose branches are merged or gone).
## Steps
1. **Gather the delta.** Find what changed after the pipeline finished:
- If the branch merged: diff the pipeline's last commit (from `state.md`)
against the merge result of those files on the default branch.
- If the human amended/rebased: `git log` the branch for commits after the
pipeline's last recorded one.
- Also ask the user directly: "what did you have to fix or wish it had
done differently?" — their answer outranks the diff.
2. **Diagnose each difference** — classify, don't just list:
- wrong assumption (check `assumptions.md` — which one, and what the right
default would have been)
- missed requirement (why did the analyst/panel miss it?)
- style/convention mismatch (what convention, citable for next time?)
- human preference with no general rule (note it, low importance)
3. **Calibrate estimates**: if `triage.md` exists, compare the S/M/L/XL guess
with reality (commits, files touched, human-reported effort). Note the
miss direction.
4. **Write lessons to the pipeline memory** (`.ticket-work/memory.md` — see
format below): only lessons that would CHANGE a future run's behavior.
"Be more careful" is not a lesson; "the payments module has a second
consumer in cron-jobs/, always check it" is.
5. **Prune while you're there**: merge duplicate lessons, delete ones
invalidated by newer experience, demote stale ones. Memory that grows
unboundedly stops being read.
6. Report to the user: lessons added (verbatim), estimate calibration, and
any pattern emerging across the last few retros (same assumption wrong
twice = the config or CLAUDE.md should change — propose the exact edit).
## Pipeline memory format (`.ticket-work/memory.md`)
One lesson per bullet, grouped under stable headings, each tagged with
importance and date:
```markdown
# Pipeline memory — learned across tickets, pruned on retro
## Repo gotchas
- [high, 2026-07-03] `npm run test:all` is flaky in CI-mode; re-run once before calling it red (PROJ-101)
## Conventions the analyst/implementer must know
- [high, 2026-07-03] error responses use ProblemDetail, never bare strings (PROJ-98)
## Reviewer calibration
- [med, 2026-07-03] security-auditor flags our internal URL-builder as SSRF every time — pre-cleared, see ADR-14 (PROJ-99)
## Estimate calibration
- [med, 2026-07-03] tickets touching the billing module: bump one size (PROJ-97)
## Human preferences
- [low, 2026-07-03] prefers small commits over one squashed commit
```
The `ticket`, `triage`, and `preflight` skills read this file at startup and
weight lessons by importance and recency — high-importance old lessons beat
low-importance new ones. Keep it under ~60 lessons; prune on every retro.
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · CristianPGit/claude-qa-architect | 100 | MIT |
| jongio/skills | 100 | MIT |
| stefanoskarakasis/product-marketing-skills | 100 | MIT |
| bartolli/kmd | 100 | MIT |
| gethamster/plugin | 100 | MIT |
| cookys/autopilot | 100 | MIT |
| rizukirr/vibekit | 100 | MIT |
| rjmurillo/ai-agents | 100 | MIT |
| rjmurillo/ai-agents | 100 | MIT |
Migawka bazy z 13.11.2026 · wersja paczki 82ef2c8078bb (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
- 100retrospective
Prowadzi retrospektywę po scaleniu PR: czyta komentarze i diff, porównuje z poprzednimi retro i zamienia wnioski w reguły projektu. - 100ticket
Prowadzi zgłoszenie z Jira od analizy wymagań, przez plan i kod na osobnej gałęzi, do przeglądu przed wysłaniem. - 100closing-out-a-design-doc
Zamyka dokument projektowy po zakończeniu lub porzuceniu prac, zapisując, co faktycznie weszło na produkcję względem celów. - 90insight-retro
Prowadzi retrospektywę po dużym wdrożeniu: mapuje zasięg zmian, sprawdza dokumentację, wyciąga nowe skille i spisuje wnioski. - 95solo-retro
Ocenia jakość przebiegu potoku po jego zakończeniu: analizuje logi, wskazuje, co poszło źle, i podsumowuje wnioski. - 100ticket-flow
Przeprowadza cały ticket od zrozumienia po testy i dokumentację, zatrzymując się tylko na bramkach dla człowieka.