TsukiNi22/ai-utils / skills/routers/audit
audit
FlaggedRouter running the audits of the current project - asks which ones to trigger (checklist - bugs/UB, quality/conventions, dependencies licenses + vulnerabilities, tests coverage, performance - coverage and benchmark checked and run automatically) and the report format, then runs the selected audit skills one after the other and writes a summary report linking them. Invoke manually with /audit.
Audit router
Request given with /audit: $ARGUMENTS
Load the skills with the Skill tool (never the other routers: they are manual-only). Never start fork agents; an audit that offers parallel agents asks the user first (its own rule).
1. Context
- Project root:
git rev-parse --show-toplevel(or the current folder outside a repository), languages / build system,git log --oneline -1(the commit audited). - Output folder:
audit/at that root (created if missing), neverdocs/unless asked.
2. Ask (one AskUserQuestion call, French questions)
- Quels audits lancer ? (
multiSelect: true, all checked = recommended):Bugs / UB->audit-bugs(sanitizers, tests, review; slow on big projects)Sécurité / failles->audit-security(memory errors, injection, auth, secrets, network...: fixed in passes until a clean one; its report is always a PDF)Propreté / conventions->audit-qualityDépendances->audit-deps(licenses to credit / restrictive, known vulnerabilities, transitive)Légal / conformité->legal-check(GDPR, cookies, legal notice, terms, accessibility, licenses, mentions; its scope global / web asked by it)Couverture des tests->coverage(C++ with tests: measured with llvm-cov) ortests(other languages)Performance->benchmark(profiling of the main scenario of the project)
- Format du rendu:
Markdown + PDF (Recommandé)·PDF seulement·Markdown seulement(passed to every audit, they don't ask again). If the request already names the audits / format, skip what is known. Audit specific questions (ex:audit-bugsscope, mode, iterations, fixes) are asked by that audit when it starts, in one call.
coverage and benchmark are checked by default and run automatically like the others, in audit mode:
coverage: measure +uncovered.md+ its report only (no test written during an audit: the "fill" part of the skill is proposed in the summary, its end-of-run question is answered by the audit's format choice); skipped with the reason when the project has no tests or no clang / llvm-cov.benchmark: the obvious scenario of the project without asking (the main binary with a typical input found in the README / tests / examples, else the test suite in an optimized build), quick depth (perf + time), report only, no optimization applied; skipped with the reason when nothing can be run (library without benchmark target or tests).
xstyle (when installed: command -v xstyle) is always run with --rtk by the audits (compact output made for
the assistant; -o <file>.json for counters kept in a report): audit-quality uses all its rules, audit-bugs the ones
that hide bugs. Not installed: say once that curl -fsSL https://tsukini22.github.io/my-universe/install.sh | bash -s then sudo dnf install xstyle (TsukiNi22/xstyle)
(no curl: wget -qO- instead of curl -fsSL) would add those checks.
3. Run (in this order, sequentially)
audit-depsfirst when selected: its JSON (/tmp/deps.json,/tmp/vulns.json) is reused byaudit-qualityandaudit-bugsinstead of running the scripts again.legal-check(reuses theaudit-depsJSON for the licenses), 3.audit-quality, 4. tests coverage, 5.benchmark, 6.audit-bugs(the longest, it may fix things only after confirmation), 7.audit-securitylast (it fixes as it goes and loops until a clean pass: the other reports describe the code before its fixes). Each audit writes its own reportaudit/<YYYY-MM-DD>-<audit>.{md,pdf}(English unless asked,reportstyle). An audit that fails or can't run (no build system, missing tool) is reported as such, the others continue.
4. Summary report
audit/<YYYY-MM-DD>-summary.{md,pdf} through report: Summary table | Audit | Verdict | High | Medium | Low | Report | (link to each report), the top 10 actions across all the audits (deduplicated, highest gain
first), what was not run and why, the commit audited. In the chat: the verdict per audit and the paths only.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit. | Stars: 466k | Updated | Claude CodeCodex | Verified | Install |
| rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure. | Stars: 466k | Updated | Claude CodeCodex | Verified | Install |
| rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics. | Stars: 466k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first. | Stars: 193k | Updated | Claude CodeCodex | Verified | Install |