auditlisted

# Security Audit Audit Claude Code extension surfaces (skills, MCP servers, hooks, CLAUDE.md) for security risks. Quick pattern scan first, then optional parallel deep analysis. ## Phase 1 — Parse & route **Prereq:** verify Python 3 with `python3 --version 2>/dev/null || python --version 2>/dev/null`. Store the working command as `{PY}`. If neither works or it's Python 2.x, stop and tell the user to install Python 3. **Parse `$ARGUMENTS`** for `<mode> [target]`: - Modes: `skill`, `mcp`, `hooks`, `claudemd`, `all`. - Targets (skill/mcp only): local path, GitHub URL, GitHub subdirectory URL (e.g. `.../tree/main/skills/x`), or npm package `@scope/pkg` (mcp only). - If mode is missing/invalid, ask via `AskUserQuestion` with the 5 modes. - If `skill` or `mcp` with remote target and no local target given, ask for one. **Resolve target** (skill/mcp with target): ```bash {PY} "${CLAUDE_SKILL_DIR}/scripts/resolve-target.py" "<target>" --type <skill|mcp> ``` Returns `{source, resolved_path, is_temporary, temp_root, error}`. On `error`, stop. If `is_temporary`, remember `temp_root` for Phase 4 cleanup and tell the user the target was downloaded. For `mcp` without target: run `gather-mcp-config.py` — if no servers, stop with "No MCP servers configured". For `hooks`/`claudemd`: no resolution needed. ## Phase 2 — Quick scan Run the relevant gather + pattern-scan scripts in parallel (multiple `Bash` calls in one message): | Mode | Gather | Pattern scan | |---|---|---| | skill | `