Agent Skills: Code Quality Audit

Use for read-only code quality, correctness, security, maintainability, observability, and operational audits that produce a ranked PROGRESS.md backlog in an isolated audit worktree.

UncategorizedID: ingpoc/SKILLS/code-quality-audit

Install this agent skill to your local

pnpm dlx add-skill https://github.com/ingpoc/SKILLS/tree/HEAD/code-quality-audit

Skill Files

Browse the full folder contents for code-quality-audit.

Download Skill

Loading file tree…

code-quality-audit/SKILL.md

Skill Metadata

Name
code-quality-audit
Description
Use for read-only code quality, correctness, security, maintainability, observability, and operational audits that produce a ranked PROGRESS.md backlog in an isolated audit worktree.

Code Quality Audit

Use this skill when the user asks to audit code quality, code smells, maintainability, correctness, security, complexity hotspots, observability, operational risk, or agentic/LLM risk without editing code.

Method

  1. Run scripts/audit-start.sh from the target repo or pass --repo <path>.
  2. The script creates an audit branch, a Git worktree, GOAL.md, and PROGRESS.md.
  3. Start Codex in the audit worktree and set the active goal from GOAL.md.
  4. During audit mode, only PROGRESS.md may be written.
  5. Use read-only subagents only for bounded exploration in large or unfamiliar repos.
  6. The main thread owns synthesis, deduping, ranking, and the final PROGRESS.md write.

Commands

Setup and print the Codex launch command:

~/.codex/skills/code-quality-audit/scripts/audit-start.sh --repo .

Setup and launch Codex immediately:

~/.codex/skills/code-quality-audit/scripts/audit-start.sh --repo . --launch

Continue later with one fix item:

~/.codex/skills/code-quality-audit/scripts/audit-next.sh /path/to/audit-worktree --launch

If you are already inside an active Codex session, do not launch nested Codex. Use the printed prompt directly, or set the current thread goal from GOAL.md.

Audit Rules

  • Setup may create branch, worktree, GOAL.md, and PROGRESS.md.
  • Audit may write only PROGRESS.md.
  • Audit must not edit, refactor, format, or fix code.
  • Audit should avoid commands that mutate repo state, including formatters, generators, snapshot updates, package installs, and test commands known to write artifacts.
  • If a finding needs code, write a checklist item instead.
  • Preserve ## Completed and ## Rejected in existing PROGRESS.md.
  • Subagents must be read-only and must not write files.

Subagent Policy

Use subagents when the repo is large, unfamiliar, has many modules, or has files over 1000 lines. Do not require subagents for small repos.

Recommended read-only splits:

  • architecture and test map
  • correctness and failure modes
  • security and agentic/LLM risk
  • complexity hotspots and performance
  • maintainability, observability, and operations

Each subagent returns concise findings with path, line range, severity, evidence, proposed fix, verification, and affected files. The main thread waits for them, removes duplicates, resolves contradictions, and writes the ranked checklist to PROGRESS.md.

Complexity Hotspot Handling

Treat complexity analysis as an audit lens, not a separate backlog owner. Raw scanner output or subagent notes must be normalized into the existing PROGRESS.md sections:

  • DEFECTS for confirmed timeouts, memory blowups, UI freezes, or production failures.
  • RISKS for material O(n^2), O(n*m), N+1, repeated lookup, or render-heavy paths on user-sized data.
  • SMELLS for local low-risk inefficiencies.
  • CROSS-CUTTING for repeated complexity patterns across modules.

Every complexity item must include the data-size assumption, before/after complexity estimate, risk level, and tests needed. The main thread decides whether the item belongs in PROGRESS.md.

References

  • templates/GOAL.md - full audit-only goal contract.
  • templates/PROGRESS.md - durable audit backlog schema.
  • scripts/audit-start.sh - deterministic audit environment setup.
  • scripts/audit-next.sh - one-item executor launch helper.