Parser Discipline 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-parsers/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 reader contract (size bounded before allocation, contextual Err, no silent short decode, non-vacuous corpus gates), delta-first scoping, six-dimension sweep, and gate-matrix 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 keep the ownership split the canonical states: parsers owns the non-NIF/non-ESM readers, while NIF routes to /audit-nif, ESM to /audit-esm, .pex/.psc to /audit-papyrus, .spt to /audit-speedtree, HKX playback to /audit-scripting, and MenuXml rendering to /audit-ui.
• Prefer the repository's codebase-memory graph tools for callers, guard locations, and decode-to-consumer wiring; use text search for bounds checks, allocation sites, version gates, guard names, configs, docs, and graph gaps.
• Interpret Claude Task-agent instructions as Codex sub-agent delegation only when delegation is available and allowed. Otherwise execute the routed dimensions sequentially.
• Run --ignored real-data suites only with game data present under the strict lane, and never materialise the vanilla Starfield CDB through the full parse (stream it, per the canonical's Phase 1 warning).
• 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.