CLAUDE.md

May 21, 2026 · View on GitHub

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

What this repo is

A single Claude Code skill (ios-code-audit) that produces a CODE_AUDIT.md report for an iOS/macOS Swift project. There is no application code, build system, or test suite here — the "deliverables" are the prompts that drive an audit run against some other Swift codebase.

When the user invokes this skill, Claude is operating in the user's iOS project directory, not this one. Edits in this repo only change future audit behavior.

File layout

  • SKILL.md — the skill definition and the 6-step workflow Claude follows when invoked. This is the source of truth; everything else is referenced from it.
  • references/agent-prompts.md — copy-paste briefs for the three parallel Explore agents (concurrency/API modernity, dead code/duplication, bugs/security/performance). Contains a placeholder table at the bottom showing how each {PLACEHOLDER} is discovered.
  • references/report-template.md — the CODE_AUDIT.md skeleton with section numbering. Section numbers (3.1, 3.2, …) are stable across edits by design so user notes referencing "§5.4" don't shift.

The three files form a single pipeline: SKILL.md drives the workflow, the agent prompts produce raw findings, and the report template structures them.

Editing rules specific to this skill

  • Severity discipline is the load-bearing invariant. SKILL.md enforces a four-level guide (Critical/High/Medium/Low) and requires spot-verification of every Critical claim before it lands in the report. Don't loosen this — it's a direct response to past runs where agents overstated severity.
  • The three agents must run in parallel. agent-prompts.md repeats this: "send all three in a single message with multiple Agent tool calls." If you edit the workflow, preserve the parallel-launch requirement.
  • Per-finding template is identical across all three agent briefs. If you change the format (Location/What/Why/Action/Severity), update it in all three briefs and the report template in lockstep — synthesis depends on it being mechanical.
  • Dead/ (or any explicitly-archived directory) is excluded by default. The skill expects to read the audited project's own CLAUDE.md to discover excluded paths.

What the skill explicitly does not cover

Listed in SKILL.md under "What this skill does NOT cover" — Metal kernel correctness, Xcode project structure, third-party dependencies, deep test coverage, localization wording, Instruments profiling. Keep that list in sync if scope changes.