critique
/critique
Przegląda własną gałąź z czystym okiem: sprawdza poprawność, prostotę, zasady projektu i pokrycie testami, z listą poprawek.
- Autor
- mo-cmyk ↗
- Repozytorium
- mazarbul ↗
- Licencja
- Apache-2.0
- Ostatnia zmiana repozytorium
- 31.10.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 31.10.2026, bez odwołań do wycofanych modeli i API.
Apache-2.0, otwarta: można używać i zmieniać, z zachowaniem licencji i informacji o zmianach.
Co robi i kiedy
Przegląda własną gałąź z czystym okiem: sprawdza poprawność, prostotę, zasady projektu i pokrycie testami, z listą poprawek.
Wyzwalacze z opisu
Oryginalny opis (EN)
Self-review of your own branch diff with clean eyes (forked Explore subagent). USE WHEN /go-strict requests a critique pass, you finished implementing >50 LoC, before opening a PR, or when user says 'critique this' or 'review my own diff'. Diffs main..HEAD, audits across 4 dimensions (correctness, simplicity, Mazarbul-specifics, test coverage), produces a punch list with file:line + suggested fix. NOT a substitute for /pr-review (which reviews OTHER PRs).
Jak działa
- When to invoke
- How it runs
- What this skill deliberately does NOT do
- Gotchas
- Pairs well with
Podgląd SKILL.md
---
name: critique
description: Self-review of your own branch diff with clean eyes (forked Explore subagent). USE WHEN /go-strict requests a critique pass, you finished implementing >50 LoC, before opening a PR, or when user says 'critique this' or 'review my own diff'. Diffs main..HEAD, audits across 4 dimensions (correctness, simplicity, Mazarbul-specifics, test coverage), produces a punch list with file:line + suggested fix. NOT a substitute for /pr-review (which reviews OTHER PRs).
context: fork
allowed-tools: Bash(git diff *), Bash(git log *), Bash(git status), Read, Grep, Glob
---
# /critique , Self-review of own diff
Critiquing your own code with the same model that wrote it leaks implementation bias. This skill runs in a **forked Explore subagent context** (`context: fork`) , fresh eyes, no memory of why you made each choice.
## When to invoke
- After implementing a feature (>50 LoC), before `/preflight` or `/ship`
- Inside the `/go-strict` auto-iteration loop
- When you're about to ship and want a sanity check
- When the user asks "critique this" / "find the holes" / "what did I miss?"
## How it runs
### Step 1 , Read the diff
```
git diff main..HEAD
git log main..HEAD --oneline
```
If the branch hasn't diverged from main (no commits, only working-tree changes), use `git diff HEAD`.
### Step 2 , Four-dimension audit
For each changed file, ask:
**1. Correctness**
- Does the code do what the commit message / PR description claims?
- Are error paths handled, or only the happy path?
- Are there off-by-ones, nil pointers, race conditions?
- Are panics possible? Should they be?
- Are types tight (no `interface{}` / `any` smuggling)?
**2. Simplicity**
- Could 2 layers of abstraction be 1? (Karpathy principle 2)
- Are there single-use helpers that should be inlined?
- Are options structs with default-only fields hiding a 2-arg function?
- Is anything speculative , handling a case that can't happen?
**3. Mazarbul-specific invariants**
- **Event log is append-only**, no UPDATE or DELETE on event rows.
- **Provenance** populated on every event (source, confidence, timestamp).
- **Scope enforcement** on every read path that surfaces user data.
- **No proto**, types are Go structs with `json:` tags; schemas are JSON Schema.
- **Timestamps**: UTC `time.Time`, JSON RFC 3339, never `*_at`.
- **TUI colors** via `internal/tui/style.Theme` only, never raw hex.
**4. Test coverage**
- Is there a test for the failing case (the one that motivated the change)?
- Are integration tests using testcontainers, not mocks of DB?
- Do tests use the canonical naming pattern (per `scripts/check-test-names.sh`)?
- Has `go test -race -short` been run on changed packages?
### Step 3 , Produce a punch list
Markdown table with file:line, issue, and concrete suggested fix:
```
File:Line | Dimension | Issue | Fix
internal/mcp/tools/relate.go:142 | Mazarbul | New tool field uses createdAt, not createTime | Rename per AIP-142
internal/kernel/dispatcher.go:88 | Simplicity | unused options struct field RetryCount | Inline default , remove struct
internal/cli/log.go:33 | Correctness | Formats timestamps in local tz, not UTC | Use .UTC() before formatting
internal/sync/reducer.go:215 | Coverage | supersede path has no test; convergence is critical | Add table-driven test
```
Group by severity:
- **MUST FIX** , compliance, correctness, security
- **SHOULD FIX** , simplicity, readability
- **NICE TO HAVE** , style, naming
### Step 4 , Report back
The critique is read-only. Return the punch list to the calling context. The caller (Claude in main context, or `/go-strict`) decides which findings to act on.
## What this skill deliberately does NOT do
- Modify code (read-only audit)
- Run lint or tests (that's `/verify`)
- Block the agent from continuing , outputs guidance only
- Repeat findings that lint or `mise run ci` already caught (focus on judgment-level issues)
- Pretend to be neutral when the diff is genuinely clean , say so
## Gotchas
- **`context: fork` means no shared state with caller** , pass diff info explicitly via the prompt, not via globals.
- **Don't over-critique tiny diffs** , for <30 LoC patches, a 4-dimension audit is overkill. Say "diff is too small to critique meaningfully" and exit.
- **MUST FIX should be RARE** , if you're flagging 5+ "must fix" items per skill invocation, the bar is too low. Reserve for compliance/correctness only.
- **Mazarbul-specific issues are highest-value** , generic JS/Go advice can be found in any linter. Focus on rules that lint can't catch (event-log append-only invariant, provenance on every fact, scope boundaries, AIP naming).
## Pairs well with
- `/go-strict` , invokes `/critique` between `/verify` and `/dead`
- `/preflight` , runs critique implicitly via golangci-lint + test suite
- `/pr-review` , for OTHER people's PRs; this skill is for YOUR OWN
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · mo-cmyk/mazarbul | 100 | Apache-2.0 |
| nigelng/nng-kit | 100 | MIT |
| synaptiai/agent-capability-standard | 100 | Apache-2.0 |
| nexu-io/open-design | 100 | Apache-2.0 |
| The-Utopia-Studio/skills | 100 | Apache-2.0 |
| enjalot/moonshine | 100 | MIT |
| fengshao1227/ccg-workflow | 100 | MIT |
| bengous/claude-code-plugins | 100 | MIT |
| akarachen/aghub | 100 | MIT |
Migawka bazy z 13.11.2026 · wersja paczki 461f3e0bed94 (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
- 100review
Przegląda repozytorium, PR-y, gałęzie lub lokalne zmiany od podstaw: konwencje, prostotę, projekt, poprawność, testy, dokumentację i bezpieczeństwo. - 90peer-review
Przegląda wskazany plik, diff albo gałąź kodu pod kątem poprawności, bezpieczeństwa i wydajności, z uwagami od krytycznych do drobnych. - 90code-review
Przegląda zmiany w EF Core pod kątem poprawności, zgodności, wydajności, bezpieczeństwa i pokrycia testami. - 90dev10x:review
Przegląda własne zmiany w gałęzi przed utworzeniem PR: porównuje diff z bazą i zwraca ustrukturyzowane uwagi. - 90review
Przegląda zmiany w kodzie pod kątem poprawności, bezpieczeństwa, prostoty, testów i zgodności ze specyfikacją. - 100dev-review
Przegląda diff, gałąź lub pull request przed scaleniem pod kątem poprawności, standardów i zgodności ze specyfikacją.