Audio Subsystem Audit

Use the Claude Code command as the canonical workflow instead of maintaining a duplicated snapshot.

Load the Canonical Command

1. Read .claude/commands/audit-audio/SKILL.md completely.
2. Read .claude/commands/_audit-common.md and .claude/commands/_audit-severity.md completely before auditing.
3. Follow the canonical command's delta-first scoping, regression guards, evidence rules (flag unsourced constants, never invent a rationale), deduplication, and report format.

Adapt Claude Conventions to Codex

• Treat the user's text after the skill request as the command arguments described by the canonical skill's argument-hint.
• Resolve each /audit-<name> reference to that audit's SKILL.md under .claude/commands/, and load a sibling audit (/audit-physics, /audit-ecs, /audit-concurrency) only at the canonical's handoff boundaries — water submersion state, stage/lock shape — reporting audio-specific ordering only.
• Prefer the repository's codebase-memory graph tools for tracing audio-API callers and the footstep/water/reverb data flow; use text search for play_oneshot/play_music/set_underwater call sites, guard test names, docs, and graph gaps.
• Disclose the canonical's #[ignore]d audio-device lifecycle guards: run them with -- --ignored or state that they were skipped.
• Interpret Claude Task-agent instructions as Codex sub-agent delegation only when delegation is available and allowed. Otherwise audit one dimension at a time.
• Preserve the canonical output path and do not create GitHub issues; publishing remains a separate audit-publish action.

Do not copy the canonical command body into this skill. The Claude command is the single source of truth.