Papyrus Frontend 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-papyrus/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-first scoping, untrusted-input severity escalations, per-dimension guard verification, and the report's untrusted-input and decompile-rate verdicts.

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 respect the boundary the canonical states: papyrus owns the .pex/.psc frontends up to the AST, while /audit-scripting owns recognizers and the runtime (translate_pex, the panic net, the extender preflight) and /audit-esm owns VMAD decode.
• Prefer the repository's codebase-memory graph tools for callers, guard-pin locations, and consumer impact analysis; use text search for bounds checks, allocation sites, depth constants, guard names, docs, 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.
• Keep the canonical's escalation in force: any panic, OOB, unbounded allocation or recursion reachable from raw .pex/.psc bytes is at least HIGH, and so is a wrong AST a recognizer accepts.
• 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.