convez.ai
Skill
y-a-v-a

stern-code-review

y-a-v-a/stern-code-review

Ver no GitHub
InicialPluginScore · de 0 a 100

What I most desire doesn’t exist on the market yet: a coding agent model, but it’s only been trained on code reviews by Eastern Europeans. - vicki on X

Instalação

$npx convez-ai y-a-v-a/stern-code-review

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 4 meses
Versão
—
Issues abertas
0
Contribuidores
1

README

A tradução para o português deste README ainda está na fila.

stern-code-review

A Claude Code skill that performs a blunt, senior-level code review focused on production risk — correctness, data safety, security, operational reliability — not style nits.

Purpose

Use it when you want a rigorous reviewer who assumes the code will run in production, treats clever code as guilty until proven useful, and treats missing tests as unprotected behavior. It is not for brainstorming, greenfield implementation, or purely stylistic feedback.

Usage

/stern-code-review <files, a diff, or a description>

Or just ask for "a stern code review" of something in context. Arguments are treated as the review target; if scope is unclear the skill asks before reviewing everything, and for large changesets it starts with the riskiest files.

How it works

The skill steers the review with a few rules:

  • Ranked priorities — correctness → data safety → security → reliability → simplicity → testability → maintainability → performance → style. Formatting nits are ignored unless they hide a deeper problem.

  • Verify before filing — a finding must be confirmed in the code shown, or marked suspected with the assumption stated. Empty sections are a valid result; the skill does not invent issues to fill them.

  • Mechanical verdict — the verdict is derived from the highest-severity finding, not chosen by feel:

    Highest findingVerdict
    any blockerrequest changes
    design structurally unsafereject
    only majorapprove with fixes
    only minor, or noneapprove
  • Conditional output — only the verdict and a one-line closing sentence are mandatory. Other sections (Findings, Suspicious choices, Tests I expect, Minimal acceptable fix, Optional cleaner version) appear only when they have content, so a trivial clean change yields a two-line review.

Each finding carries a severity (blocker / major / minor), a confidence (confirmed / suspected), a location, the problem, why it matters, and a suggested fix.

Ver o original no GitHub

Bom para

Relacionados