Starfield Compatibility 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-starfield/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 owner-audit scope split (this audit owns Starfield's data through shared mechanisms), its measured status authority, the six risk-ordered dimensions with per-dimension scratch outputs, and the Phase 3 merge (CRC32 flag table, remaining-work chain, 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 route shared-mechanism defects to their owners: audit-parsers (BA2/CDB readers), audit-nif (block parsing), audit-esm (ESM walker), audit-nifal (canonical material invariants), audit-ui (HUD), audit-character (NPC stat model).
• Prefer the repository's codebase-memory graph tools for symbol discovery, callers, and impact analysis; use text search for guard test names, FourCCs, corpus gate commands, docs, and graph gaps.
• Interpret Claude Task-agent instructions as Codex sub-agent delegation only when delegation is available and allowed; the canonical runs each dimension as an agent (max 3 concurrent), otherwise audit the dimensions one at a time in risk order.
• 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.