Exterior Audit — EXAL / SKYAL / WATAL / Ground Cover

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

Load the Canonical Command

1. Read .claude/commands/audit-exterior/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 four tier invariants, per-dimension Paths: scope triage, live-guard rule, extra per-finding fields (Dimension, Tier Violated, Game Affected), Known-Open Register, 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 cross only at the canonical's declared seams: byte-level WTHR/WATR/LTEX/GRAS/LAND decode to /audit-esm, water shading and sky consumption to /audit-renderer, buoyancy to /audit-physics.
• Prefer the repository's codebase-memory graph tools for translation-boundary callers, producer/consumer chains, and impact analysis; use text search for translate_* call sites, GameKind tokens in Rust and GLSL, sentinel literals, shader text, 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 evidence to cargo tests and the canonical's captured-frame smoke harness; no windowed engine launch (feedback_no_parallel_engine_launch).
• 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.