ESM / Plugin Parser 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-esm/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 phase structure (setup, eight dimensions, merge, cleanup), per-dimension guard verification, xEdit citation requirements, the Record Coverage Matrix, 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 only to route its out-of-scope territory (per-game slices, /audit-gameplay consumption, /audit-exterior translation), but always deduplicate against the latest per-game audit reports, where this crate's findings historically live.
• Prefer the repository's codebase-memory graph tools for parser call paths, dispatch sites, and consumer-handoff discovery; use text search for 4-char record codes, sub-record literals, GameKind arms, xEdit citations in docstrings, docs, and graph gaps.
• Interpret the canonical orchestrator (one Task agent per dimension, max 3 concurrent) as Codex sub-agent delegation only when delegation is available and allowed. Otherwise audit the dimensions sequentially.
• Keep the test surface to the canonical's allowed set (plain cargo tests, example probes); never run this crate's --ignored real-data tests, which have OOM-killed sessions.
• 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.