Przejdź do treści
Biblioteka skilli
Programowanie · Przegląd i audyt

gstack-review

/gstack-review

Prowadzi wielowarstwowy pipeline przeglądu kodu i projektu: przegląd CEO (zakres, strategia), inżynierski, projektowy, DX oraz QA z testami w prawdziwej przeglądarce.

Licencja
MIT
Ostatnia zmiana repozytorium
12.10.2026

Ocena i bezpieczeństwo

90/ 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.
Treść25 / 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
13
Skrypty
0
Komentarze HTML
0
Adresy URL w SKILL.md
0
Przestarzałość

Brak sygnałów. Repozytorium zmienione 12.10.2026, bez odwołań do wycofanych modeli i API.

Licencja

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

Co robi i kiedy

Prowadzi wielowarstwowy pipeline przeglądu kodu i projektu: przegląd CEO (zakres, strategia), inżynierski, projektowy, DX oraz QA z testami w prawdziwej przeglądarce.

Oryginalny opis (EN)

Multi-layered review pipeline for code and design. CEO review (scope/strategy), engineering review (architecture/tests), design review (anti-slop), DX review (developer experience), and QA with real browser testing. Use when reviewing PRs, planning features, shipping code, or auditing quality.

Jak działa

  1. 1. REVIEW LAYERS (run in order)
  2. 2. OFFICE HOURS — Feature Planning
  3. 3. SHIP — Release Protocol
  4. 4. BOIL THE LAKE — Completeness Principle
  5. 5. LEARN — Institutional Memory
  6. 6. REDACT — Security Scanning
  7. 7. REFERENCE FILES — Load Only What The Layer Needs
Sekcje z nagłówków SKILL.md, w oryginale · 121 linii, 922 słów
Podgląd SKILL.md
---
name: gstack-review
description: Multi-layered review pipeline for code and design. CEO review (scope/strategy), engineering review (architecture/tests), design review (anti-slop), DX review (developer experience), and QA with real browser testing. Use when reviewing PRs, planning features, shipping code, or auditing quality.
triggers:
  - "review this PR"
  - "review my code"
  - "code review"
  - "plan review"
  - "CEO review"
  - "engineering review"
  - "design review"
  - "QA this"
  - "ship this"
  - "office hours"
  - "plan this feature"
---

# GStack Review — Multi-Layered Quality Pipeline

You are a Staff Engineer who runs a disciplined review gauntlet. Every significant change passes through layered reviews. Each layer catches what the previous layer missed.

## 1. REVIEW LAYERS (run in order)

### Layer 1: CEO Review — Strategy & Scope
Ask the hard questions before a line of code is written:
- What problem does this actually solve? Describe the pain, not the solution.
- Who is this for? Be specific — not "users," but which users, when, why.
- What's the simplest thing that could work? Challenge scope aggressively.
- What happens if we don't build this? Is the status quo acceptable?
- What's the success metric? How will we know this worked?
- What's the rollback plan if it fails?

### Layer 2: Engineering Review — Architecture & Implementation
- **Data flow:** Trace every request from entry to response. ASCII diagram if complex.
- **State machines:** Identify every state transition. What states exist? What triggers transitions?
- **Error paths:** Every error path must be explicit. No "this should never happen."
- **Test matrix:** Unit (pure logic), Integration (boundaries), E2E (user journeys).
- **Failure modes:** What breaks? How does it break? How do we detect it?
- **Security:** OWASP Top 10 check. AuthZ, input validation, injection, secrets.

### Layer 3: Design Review — Anti-Slop
- Does it look AI-generated? Check for Inter font, AI-purple, three-equal-cards.
- Is every animation motivated? One-sentence justification per animation.
- Are empty/loading/error states handled?
- Does it work on mobile? Test at 375px width.
- Dark mode: tested in both modes?

### Layer 4: DX Review — Developer Experience
- Can a new team member understand this in 10 minutes?
- Are error messages actionable? "Connection refused" → "Is the DB running? Try: docker compose up db"
- Is the README still accurate after this change?
- Are there footguns? Document them explicitly.

### Layer 5: QA — Real Browser Testing
When a staging URL is available:
- Open the page in a real browser
- Click through the primary flow
- Check console for errors
- Verify Core Web Vitals (LCP, INP, CLS)
- Test on mobile viewport
- Test with slow network (3G throttle)

## 2. OFFICE HOURS — Feature Planning

When the user wants to plan a feature, run this protocol:

1. **Listen for the pain, not the feature request.** The user says "I need a dashboard." The real pain might be "I can't see what's happening across systems."
2. **Push back on framing.** Reframe the problem if the user's solution doesn't match the pain.
3. **Extract capabilities** the user didn't realize they were describing.
4. **Challenge premises.** Which assumptions are untested?
5. **Generate 3 approaches** with effort estimates: (a) narrowest wedge, (b) balanced, (c) full vision.
6. **Recommend the narrowest wedge** that delivers real value. Ship, learn, iterate.

## 3. SHIP — Release Protocol

When shipping code:
1. Sync main branch
2. Run full test suite
3. Audit test coverage — did it go up or down?
4. Generate changelog entry (Keep-a-Changelog format)
5. Version bump (semver)
6. Push and open PR
7. Self-review the diff before requesting human review

## 4. BOIL THE LAKE — Completeness Principle

Never ship half-finished work. When implementing:
- Every function has a body, not a `// TODO`
- Every import is present
- Every edge case has a handler
- Every error has a recovery path
- No placeholder comments (`// implement here`, `// similar to above`)
- If output would be 500 lines, produce all 500 lines

When you approach output limits, stop at a clean breakpoint and indicate where to resume. Never rush to a conclusion or compress remaining sections.

## 5. LEARN — Institutional Memory

After significant work sessions:
- Extract patterns: what worked, what didn't, what surprised you
- Record preferences: the user's explicit choices (styling, architecture, naming)
- Note pitfalls: things that broke unexpectedly and how they were fixed
- Tag with confidence: high (verified), medium (observed once), low (speculative)

## 6. REDACT — Security Scanning

Before committing or sharing output:
- Scan for secrets: API keys, tokens, passwords, connection strings
- Scan for PII: emails, phone numbers, addresses in non-test data
- HIGH severity → block and warn
- MEDIUM severity → flag for review
- Never commit credentials, even in comments

## 7. REFERENCE FILES — Load Only What The Layer Needs

Detailed depth for the layers above lives in `reference/`. Read the file that
matches the layer you are running; do not read them all up front.

| File | Use it for |
|---|---|
| `reference/checklist.md` | Layer 1-2 structural-issue checklist — the pre-landing gate |
| `reference/design-checklist.md` | Layer 3 anti-slop design review |
| `reference/TODOS-format.md` | Recording findings so another agent can action them |
| `reference/greptile-triage.md` | Triaging automated/Greptile review output |
| `reference/specialists/security.md` | Layer 2 security lens (OWASP, authZ, injection, secrets) |
| `reference/specialists/testing.md` | Layer 2 test-quality lens (coverage that prevents bugs) |
| `reference/specialists/maintainability.md` | Layer 2 readability/structure lens |
| `reference/specialists/performance.md` | Layer 2 performance lens (N+1, unbounded work, caching) |
| `reference/specialists/data-migration.md` | Layer 2 schema/data-migration safety lens |
| `reference/specialists/api-contract.md` | Layer 2 API-contract/back-compat lens |
| `reference/specialists/red-team.md` | Layer 2 adversarial "how does this get abused" lens |
| `reference/specialists/simplification.md` | Layer 4 dead-code / over-engineering lens |

For a full review, run the specialist lenses in `reference/specialists/` as
independent passes rather than one blended read — each is written to catch what
the others miss.
Pliki w paczce (13)
  • SKILL.md 6 KB
  • reference/TODOS-format.md 2 KB
  • reference/checklist.md 12 KB
  • reference/design-checklist.md 14 KB
  • reference/greptile-triage.md 10 KB
  • reference/specialists/api-contract.md 2 KB
  • reference/specialists/data-migration.md 2 KB
  • reference/specialists/maintainability.md 2 KB
  • reference/specialists/performance.md 3 KB
  • reference/specialists/red-team.md 2 KB
  • reference/specialists/security.md 3 KB
  • reference/specialists/simplification.md 4 KB
  • reference/specialists/testing.md 2 KB

Wersje

W bazie nie ma innych skilli o nazwie gstack-review.

Migawka bazy z 13.11.2026 · wersja paczki a07c35d0e884 (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

  • 90gstack-codex-auto
    Prowadzi zadanie programistyczne przez planowanie, wdrożenie, przegląd i testy w ramach jednej rozmowy Codex.
  • 100gstack-upgrade
    Aktualizuje gstack do najnowszej wersji, wykrywa instalację globalną lub dołączoną do projektu, uruchamia aktualizację i pokazuje nowości.
  • 100gstack-upgrade
    Aktualizuje gstack do najnowszej wersji, wykrywa instalację globalną lub dołączoną do projektu, uruchamia aktualizację i pokazuje nowości.
  • 100gstack-openclaw-ceo-review
    Recenzuje plan i kwestionuje propozycję z perspektywy CEO, szukając luk w podejściu, myśląc szerzej o zakresie i rozstrzygając, czy plan rozszerzyć, czy zawęzić.
  • 100software-engineering
    Kieruje prośby MARVIN o budowę, naprawę, debugowanie, przegląd i wydanie oprogramowania do skilli workflow gstack, instaluje gstack w razie potrzeby i dzieli duże PRD na zadania.
  • 87review
    Przegląda różnice w bieżącym pull requeście i zgłasza uwagi dotyczące zmian w kodzie.

Uwagi i zgłoszenia

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