wiki_auditlisted

# LLM Wiki — Audit Skill (wiki_audit) > **CLI (read first):** This skill drives the `magi` CLI (MAGI research workspace tool, assumed installed on PATH). If unsure of your surroundings, run `magi sync` first to locate the workspace. For the full syntax of any command: `magi <command> --help`. This skill handles factual auditing, truth-seeking evaluations, and thesis-driven investigations across the compiled knowledge base. To prevent context window limits on large vaults, it strictly uses a Map-Reduce architecture. > **Tooling (framework-agnostic):** This skill is written tool-agnostic. Map each capability to your own agent's tool — *read-file* (`Read` in Claude Code, `view_file` in Antigravity), *sub-agent / parallel task* (`Task`/`Agent` in Claude Code, `invoke_subagent` in Antigravity), *shell* (`Bash`/`PowerShell`). Use the closest equivalent your framework provides; if a parallel sub-agent tool is unavailable, audit each file subset sequentially yourself. When the user asks to perform an audit or truth check on their vault: 1. **Map (Deterministic Inventory — SCRIPT FIRST)**: * Run the wiki inventory command to get a deterministic file listing — do NOT manually browse or rely on grep keywords alone: `magi stats <TOPIC_DIR> wiki-summary` * Parse the JSON output to understand the vault structure: total files, per-directory counts, file titles, and which files have sources. * Before graph/regex retrieval, run `magi search "<claim or topic>" -