LLM Wiki — Audit Skill (wiki_audit)

Resolving script paths (read first): Commands below invoke scripts as <BIN>/X.py (and a few as <SKILLS>/...). Resolve these to absolute paths once before running anything:

>

- <SKILL_DIR> = the directory this SKILL.md lives in. - <SKILLS> = the skills/ folder containing this skill = <SKILL_DIR>/.. - <BIN> = the bin/ folder beside it = <SKILL_DIR>/../../bin

>

Do not hardcode a fixed prefix like .agents/bin or ../bin: shell relative paths resolve against the current working directory (usually the topic root), not this skill's location. Once resolved, <BIN> is typically .agents/bin when invoked from the hub root, or .claude/bin from inside a topic directory.

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 script to get a deterministic file listing — do NOT manually browse or rely on grep keywords alone:

python <BIN>/llm-wiki.py 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.
• Graph Analysis (MANDATORY): Run python <BIN>/llm-wiki.py graph to ensure the local graph database is strictly up to date. Do NOT skip this, otherwise you will read stale data!
• Then query the knowledge graph using python <BIN>/query-graph.py "<SQL>". Do not use direct sqlite3 command line execution.

Graph DB Schema:

• nodes(id TEXT PRIMARY KEY, path TEXT, title TEXT, type TEXT, category TEXT, summary TEXT, created TEXT, updated TEXT)
• edges(source_id TEXT, target_id TEXT, type TEXT)
• tags(node_id TEXT, tag TEXT)
• aliases(node_id TEXT, alias TEXT)

Example Queries:

• SELECT path FROM nodes WHERE category='reference' AND id IN (SELECT node_id FROM tags WHERE tag='quantum-mechanics')
• SELECT n.path, e.type FROM nodes n JOIN edges e ON n.id = e.target_id WHERE e.source_id = 'some-concept-id'
• Use the inventory and graph results to select the files most relevant to the user's audit query. Then use python <BIN>/search-wiki.py "<regex>" <files...> for targeted keyword searches within those specific files.
• Do NOT attempt to read all compiled cards manually.

1. Reduce (Subagent Phase):

• Use your agent's sub-agent / parallel-task tool to spawn one or more "Audit Subagents". Assign each subagent a specific subset of the relevant files. (If no sub-agent tool exists, audit each subset sequentially yourself.)
• Subagent Output Contract (MANDATORY): Each subagent MUST structure findings as:

        CLAIM: "<exact quote from conflicting file>"
        EVIDENCE: "<exact quote from file>"
        SOURCE_TYPE: local_wiki
        SOURCE: <wiki file path>
        CONTRADICTS_SOURCE: <conflicting wiki file path>
        SEVERITY: high|medium|low
        EXPLANATION: <why these claims conflict>

• If a subagent fails or times out, log the failure and proceed with available results.

1. Verify Citations (MANDATORY):

• Save all subagent outputs to scratch/temp_claims.txt.
• Run python <BIN>/verify_claims.py scratch/temp_claims.txt --topic-dir "<TOPIC_DIR>"
• Discard any finding that is reported as [UNVERIFIED]. Log discarded findings separately.

1. Synthesize: Merge the verified findings into a structured investigation report (Thesis).

1. Produce Theses:

• Save the compiled Thesis report under wiki/theses/YYYY-MM-DD-<slug>.md with proper YAML frontmatter:

        ---
        title: "Thesis: <descriptive title>"
        type: thesis
        category: reference
        created: YYYY-MM-DD
        sources:
          - <list of wiki files examined>
        tags: [audit, thesis]
        confidence: <high|medium|low>
        summary: "<1-2 sentence summary of findings>"
        ---

• Post-Write Validation (MANDATORY): Run:

python <BIN>/validate-output.py "<thesis_file>" --schema thesis --wiki-root "<TOPIC_DIR>" If validation fails, fix the reported issues before proceeding.

• Run: python <BIN>/llm-wiki.py stats <TOPIC_DIR> verify-refs "<thesis_file>"

to ensure all [[references]] in the thesis point to existing files.

1. Log: Update the activity log log.md with: audit query, files examined count, findings count, findings discarded count.