Skip to content
Research
Command

/audit

Truth-seeking umbrella audit for llm-wiki. Combines active wiki maintenance, output drift checks, provenance review, and fresh research when trust is in doubt.

BOOST
From plugin
nvk-llm-wiki
1.4k28 skills28 commands
Install
> /plugin marketplace add nvk/llm-wiki
> /plugin install wiki@llm-wiki

How it fires

How this command gets triggered: by you, by Claude, or both.

  • Fires itselfClaude auto-loads it when your prompt matches the work.
  • You can call itInvoke it directly when you want it.
  • Slash command/audit

Context preview

What this command does when you run it.

Truth-seeking umbrella audit for llm-wiki. Combines active wiki maintenance, output drift checks, provenance review, and fresh research when trust is in doubt.

Command definition

audit.md
description: "Truth-seeking umbrella audit for llm-wiki. Combines active wiki maintenance, output drift checks, provenance review, and fresh research when trust is in doubt."
argument-hint: "scan [--artifact <path>] [--project <slug>] [--wiki-only] [--outputs-only] [--quick] [--fresh] [--include-archived] | report [--wiki <name>] [--local]"
allowed-tools: Read, Write, Edit, Glob, Grep, Bash(ls:*), Bash(wc:*), Bash(date:*), Bash(mkdir:*), WebFetch, WebSearch, Agent

Your task

**Resolve the wiki.** Do NOT search the filesystem or read reference files until the wiki location is known — follow these steps: 1. Read `$HOME/.config/llm-wiki/config.json`. If it has `hub_path`, expand leading `~` only (not tildes in `com~apple~CloudDocs`) and prefer that path; use `resolved_path` only as a fallback cache when the expanded `hub_path` is unavailable and `resolved_path` is initialized. If config has only `resolved_path`, use it. If the configured path can be statted but reading `wikis.json` or listing `topics/` fails with `Operation not permitted`, stop and ask the user to grant Full Disk Access/iCloud Drive access to the launcher; do not fall back to `~/wiki` or `resolved_path`. Do not write machine-specific `resolved_path` into shared configs. 2. If no config -> read `$HOME/wiki/_index.md`. If it exists -> HUB = `$HOME/wiki`. If nothing found, ask the user where to create the wiki. 3. **Wiki location** (first match): `--local` -> `.wiki/` in CWD; `--wiki <name>` -> `HUB/wikis.json` lookup with portable path resolution (`<HUB>`, `~`, absolute, or HUB-relative); if the registry path is stale, fall back to `HUB/topics/<name>`; CWD has `.wiki/` -> use it; else -> HUB. 4. Read `<wiki>/_index.md` to verify. If missing -> stop with "No wiki found. Run `/wiki init` first."

Read the audit reference at `skills/wiki-manager/references/audit.md`. If you need a fresh wiki-only pass, also read `commands/librarian.md` and reuse its scan protocol rather than inventing a parallel one.

Inventory awareness: audit findings that require future action should be offered as inventory candidates, not only buried in `.audit/REPORT.md`. Examples include blocked verification work, stale source queues, watch items, and important missing corpora. Show a small sample before creating more than a few records, and never treat inventory records as evidence for factual verdicts.

Parse $ARGUMENTS

Default subcommand is `scan`.

  • **scan**: Run the umbrella trust audit
  • **report**: Display the latest `.audit/REPORT.md`

Flags:

  • **--artifact <path>**: Audit one artifact relative to the wiki root. Supports `wiki/...` articles or `output/...` artifacts.
  • **--project <slug>**: Audit outputs in `output/projects/<slug>/`
  • **--wiki-only**: Only run the wiki-content pass
  • **--outputs-only**: Skip the wiki-wide pass and focus on outputs plus their dependency chains
  • **--quick**: Local-only audit. Skip fresh web research unless the user explicitly demands it.
  • **--fresh**: Ignore cached `.librarian/scan-results.json` and run a fresh wiki-content pass first
  • **--include-archived**: Explicitly include archived topic wikis. Default full

audits skip archived material unless the target artifact depends on it.

  • **--wiki <name>**: Target a specific topic wiki
  • **--local**: Use project-local `.wiki/`

Scan Protocol

0. Initialize scope

1. Create `.audit/` if it does not exist. 2. Derive the audit scope:

  • `--artifact` -> just that artifact
  • `--project` -> markdown deliverables under `output/projects/<slug>/`, excluding `WHY.md`
  • `--wiki-only` -> all wiki articles
  • `--outputs-only` -> all markdown outputs under `output/`, excluding `_index.md` and `WHY.md`
  • default -> full umbrella audit

3. Record the scope, timestamp, and selected flags for the report.

Archive rule: default full-scope audit skips archived topic wikis. If the targeted artifact/source chain explicitly cites archived material, follow that dependency and label it as archived. For broad archived audits, require `--include-archived`.

1. Wiki-content pass

Skip this pass only when `--outputs-only` is set and the targeted artifact does not depend on wiki articles.

1. If `.librarian/scan-results.json` exists and is recent enough for the current task, reuse it unless `--fresh` is set. 2. Otherwise run the librarian scan protocol from `commands/librarian.md`. 3. Pull forward the findings that matter for trust:

  • stale wiki articles
  • low-quality wiki articles
  • weak or broken source chains

2. Output dependency and drift pass

Skip this pass only when `--wiki-only` is set.

1. Build the output artifact list from scope. 2. For each artifact:

  • Read frontmatter and capture `sources:`, `generated:`, `project:`
  • Flag `missing-provenance` if `sources:` is absent or empty
  • Resolve every dependency path
  • Flag `broken-source-ref` if a dependency does not exist
  • Compare dependency dates (`updated:`, `ingested:`, `generated:`) against the artifact's `generated:` date
  • Flag `drifted-dependency` if any dependency is newer than the artifact
  • If a dependency is a wiki article, inherit relevant librarian findings
  • If a dependency is another output artifact, recurse one hop into its `sources:` chain

3. Classify each artifact as `clean`, `drifted`, `provenance-gap`, `weak-evidence`, `contradicted`, or `unresolved`.

3. Truth escalation pass

This is the defining behavior of `/wiki:audit`. Do not stop at "it looks stale" if the truth can be resolved with more investigation.

Escalate when:

  • the user explicitly asks whether they can trust an artifact
  • an output is drifted or has a provenance gap
  • upstream wiki material is stale, weak, or contradictory
  • the topic is volatile or the stakes are high

For each escalated item:

1. Re-read the artifact and identify the specific claims under scrutiny. 2. Re-read local dependencies and any cited raw sources. 3. If raw sources map to live URLs, fetch th

Read more
Ships withnvk-llm-wiki

LLM-compiled knowledge bases for any AI agent. Parallel multi-agent research, thesis-driven investigation, source ingestion, wiki compilation, querying, and artifact generation.

Get the whole plugin
Stats
1,394
Stars
131
Forks
Active
Maintenance
Python
Language
MIT
License
6d ago
Last commit
6mo ago
Created
1d ago
Added

Repo: nvk/llm-wiki

Other commands on nvk-llm-wiki.