UI Audit: Scaleform (R4 + M48) and MenuXml (M48.4-M48.7)

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

Load the Canonical Command

1. Read .claude/commands/audit-ui/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 baseline, re-derived catalog counts and data-gated test caveats, per-dimension guards and first steps, and Host-Contract-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 routed sibling audit (parsers, concurrency, safety, gameplay) only when the handoff it owns needs its deeper checklist.
• Prefer the repository's codebase-memory graph tools for host-call, guard-symbol, and route discovery; use text search for SWF/ABC bytecode, eval/layout/raster details, Vulkan upload and presentation sites, and graph gaps.
• Interpret Claude Task-agent instructions as Codex sub-agent delegation only when delegation is available and allowed. Otherwise audit one dimension at a time.
• Treat data-gated tests that silently return early without game data as unverified unless run with data, per the canonical command.
• 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.