NIF 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-nif/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 stream-drift-first dimension order, docs/engine/nif-parser.md / nif.xml reference policy, regression pins (verify still fixed, never re-report), opt-in corpus-gate rules, per-finding Dimension + Game Affected fields, and Block Type Coverage 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 load a sibling audit only at the canonical's declared seams — archive readers and previs → parsers, per-game material translation → nifal, SpeedTree .spt → speedtree, Havok behaviour → physics (only the constraint CInfo decode is a NIF seam).
• Prefer the repository's codebase-memory graph tools for symbol discovery, callers, and impact analysis; use text search for raw version literals, bsver bands, nif.xml since=/until= gates, block_sizes reads, and dispatch/shader arms.
• Interpret Claude Task-agent instructions as Codex sub-agent delegation only when delegation is available and allowed. Otherwise audit one dimension at a time.
• Corpus gates are opt-in #[ignore] tests needing installed game data — check the latest per-title gate result before trusting a green default cargo test -p byroredux-nif for parse rates.
• 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.