Zorille: Auditoria e Higiene de Codebase com Claude
boinger/zorille
Auditoria, planos de fix, deps e forensics de issues para codebases
Ver no GitHubConjunto de quatro skills TypeScript para diagnóstico profundo de codebases desconhecidas: auditoria estruturada de bugs e débito técnico, planejamento de correções em escala de PR, remediação de dependências com triagem por risco (Go, Python, Node, C#, etc.), e rigor forense para contribuições upstream. Transforma achados brutos em planos acionáveis e relatórios de qualidade.
Instalação
npx convez-ai boinger/zorille1 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
Casos de uso
- Diagnóstico cold-start de codebase herdada antes de mudanças
- Conversão de findings CodeQL/Semgrep em PRs planejadas e dimensionadas
- Auditoria e remediação de dependências com triagem de CVEs por risco
- Redação rigorosa de issues e PRs para contribuições upstream
Veredito

Ferramenta madura e bem-estruturada para times que herdam ou mantêm codebases complexas. Cobre o ciclo completo de higiene: detecção → planejamento → remediação → comunicação upstream. Excelente para automação de code review e CI/CD; requer familiaridade com SARIF e ecossistema de scanners.
README
A tradução para o português deste README ainda está na fila.
zorille
Four Claude Code skills for finding what's wrong in a codebase you don't know and doing something useful about it: bugs, fix plans, dependency hygiene, and upstream-contribution forensics.
Four skills, one repo
/codebase-auditfinds problems. Cold-start audit of any codebase: bugs, security issues, architectural problems, tech debt, test gaps. Writes a structured report and abaseline.jsonfor regression tracking. Read-only by default; opt in to mechanical fixes with--quick-fix./plan-fixesturns problems into fix plans. Reads the baseline/codebase-auditwrites — or any SARIF 2.1.0 source (CodeQL, ESLint, Semgrep, Sonar, GitHub Code Scanning). Groups findings into PR-sized plans with depth-aware investigation: callers, tests, and adjacent context./depshandles dependency hygiene. Audit, update, and CVE remediation across Go, Python, Swift, Dart/Flutter, C#/.NET, and Node.js. Risk-tiered updates (critical → security → patch → minor) with test verification; never auto-bumps majors without approval. Standalone — doesn't require the audit/plan-fixes flow, but complements/codebase-audit's quick CVE scan with deeper investigation and remediation./issue-forensicsapplies investigative rigor to a non-trivial finding before you file an upstream issue or PR. Four-question entry gate routes routine fixes to a quick-report template; substantive findings get a five-pillar playbook: SHA-pinned permalinks, structural twins, history with stated-intent-vs-side-effect discipline, disproof of current design, exhaustive caller trace. Produces a structured draft modelled on a gold-standard exemplar. Companion to give-back for anti-squatting hold-and-release timing.
The slash commands are /codebase-audit, /plan-fixes, /deps, and /issue-forensics regardless of where you cloned the repo. (For the story behind the repo name, see the bottom of this file.)
What this solves
You inherited a codebase you don't know. You need to know what's in there before you start changing it — what's broken, what's risky, what's tech debt, what's a security problem. /codebase-audit reads the code cold and produces a structured report you can act on. Run it again later and it tells you what's fixed, what's new, and where the score moved.
/plan-fixes is the bridge from "I have a list of findings" to "I have a PR-sized plan I can hand to a coding agent." It works against /codebase-audit's output, but it also reads SARIF from any external scanner — so existing CodeQL or Semgrep results can become reviewable fix plans without re-scanning.
And /deps covers the dependency-hygiene layer of "what's wrong" — outdated packages, known CVEs, deprecated libraries — across the major language ecosystems, with risk-tiered updates that won't silently bump a major version on you.
/issue-forensics is the one that turns toward someone else's codebase: once you've found something worth reporting upstream, it applies the same rigor as the audit tools to the write-up itself. Four-question gate to short-circuit routine fixes; full playbook for the non-trivial ones. Produces a structured draft that reads like the issue you wish every contributor filed.
What a fix plan looks like
/plan-fixes compresses raw findings into reviewable, PR-sized plans:
Compressed 217 CodeQL findings into 14 reviewable plans (avg 15.5 findings per plan).
Action needed: 9 plans (3 high-risk, 4 medium, 2 low)
Already applied: 5 plans (auto-fixed by /codebase-audit --quick-fix)
Sources: codeql: 217 findings
Each plan is a markdown file with Context, a Findings table, per-finding Approach (caller analysis, test coverage gaps, contextual notes), Files to Modify, Risk assessment, Verify & Rollback steps, and cross-plan Dependencies.
Install
git clone https://github.com/boinger/zorille ~/Projects/zorille
cd ~/Projects/zorille
./setup
After ./setup, all four slash commands (/codebase-audit, /plan-fixes, /deps, /issue-forensics) are available in Claude Code. The setup script is idempotent — re-run it to upgrade.
Usage
/codebase-audit Full audit (10-30 min depending on codebase size)
/codebase-audit --quick 2-minute smoke check
/codebase-audit --suggest-fixes Full audit with inline fix diffs per finding
/codebase-audit --quick-fix Audit + auto-apply mechanical fixes
/codebase-audit --plan-fixes [--thorough] Audit + dispatch to /plan-fixes (alias)
/plan-fixes Plan from latest baseline (auto-discovers)
/plan-fixes --from baseline.json Plan from explicit baseline file
/plan-fixes --from results.sarif Plan from any SARIF 2.1.0 source
/plan-fixes --sarif results.sarif Same, explicit format
/plan-fixes --thorough Auto-investigate all substantive findings
/plan-fixes --show-applied Expand already-applied plans in the menu
/deps Audit dependencies (default subcommand)
/deps update Apply non-major updates with test verification
/deps cve Apply security fixes only
/deps triage Prioritize open Dependabot/Renovate PRs
Audit modes: Full (default — all 4 phases, full report). Quick (--quick — phase 1 only + top 10 patterns, ~2 min). Regression (automatic — diffs against the previous baseline if one exists). Suggest fixes (--suggest-fixes — adds inline diffs to each finding). Quick fix (--quick-fix — auto-applies high-confidence mechanical fixes).
Configuration
Audit storage. Reports and baselines are saved to ~/.codebase-audits/<project>/audits/. Override the location with CODEBASE_AUDIT_HOME:
export CODEBASE_AUDIT_HOME="$HOME/.local/share/codebase-audits"
Custom checklist patterns. Drop a .codebase-audit/checklist.md in your project root to add project-specific audit rules. Custom patterns run after the built-in ones, in the same format as the built-in checklist.md in this repo.
Use in CI
zorille is also a published GitHub Action. The simplest case:
- uses: boinger/zorille@v1
with:
anthropic-api-key: ${{ secrets.ANTHROPIC_API_KEY }}
For SARIF output (fed to GitHub Code Scanning), baseline-first onboarding for legacy codebases, PR-scoped checks via changed-only, and the full inputs/outputs reference, see CI.md.
Upgrade
cd ~/Projects/zorille && ./setup
Idempotent — works for fresh installs and upgrades.
About the repo name
zorille is the alternate French spelling of zorilla — the African striped polecat (Ictonyx striatus), a small carnivore in family Mustelidae. The mustelids are stoats, martens, weasels, polecats, otters, badgers, ferrets: curious, smart little animals.
I felt like being named after a mustelid fit the work — small, sharp-nosed creatures that find things hidden under rocks, which is roughly what /codebase-audit, /plan-fixes, /deps, and /issue-forensics do for codebases you don't yet know (your own or someone else's). And I picked "zorille" specifically (not the more common English "zorilla") because there weren't any other projects named that.
A literal name like code-tools or audit-suite would have been boring. The skills inside have those descriptive names because that's what skills need; the repo gets to have personality.
License
MIT