lint

name	lint
description	LLM-driven semantic wiki audit. Checks contradictions, stale claims, missing concepts, orphan clusters, and duplicates. Use when user says "lint", "语义检查", "audit wiki", or "deep check".

/lint — LLM Semantic Wiki Audit

Purpose

Complement the deterministic health check (health-check.py) with LLM-powered semantic analysis. Tier 1 (health-check) catches structural problems. Tier 2 (/lint) catches meaning-level problems.

Trigger: "lint", "语义检查", "audit wiki", "deep check", "run lint"

Execution

Step 1: Gather context

Read:

• wiki/purpose.md — what this wiki is for
• wiki/overview.md — current state snapshot
• wiki/log.md — recent activity

Step 2: Sample wiki pages

Read a representative sample:

• 5-10 permanent notes (spread across topics)
• 3-5 literature notes
• 2-3 MOC pages

The LLM decides which pages to read based on overview.md topic distribution and recent log activity.

Step 3: Semantic checks

For each of the following, find concrete issues with page references:

Contradictions: Two pages make conflicting factual claims. Example: page-A says "X was released in 2023" but page-B says "X launched in 2024."

Stale claims: A page asserts something that a more recent source corrects or refines. Check sources[] dates and cross-reference with newer literature notes.

Missing concepts: A term or concept appears frequently across pages (in wikilinks, in prose) but lacks its own permanent note.

Orphan clusters: A group of pages heavily interlinked with each other, but with no links to the rest of the wiki. These may represent knowledge that should be integrated.

Duplicates: Two pages covering substantially the same idea under different titles. Flag for potential merge.

Step 4: Generate report

Write wiki/lint-{date}.md:

---
type: lint
date: "YYYY-MM-DD"
issues_found: N
---
# Lint Report — {date}

## Contradictions
- `page-A` vs `page-B`: ...

## Stale Claims
- ...

## Missing Concepts
- ...

## Orphan Clusters
- ...

## Duplicates
- ...

## Recommendations
_(actionable next steps)_

Step 5: Log and report

Append to wiki/log.md:

## [YYYY-MM-DD HH:MM] lint | N issues found

Present a summary to the user. Ask which issues to fix.

Design note

This is complementary to health-check.py, not a replacement. health-check runs on every compile (deterministic, zero token cost). /lint runs on demand (LLM-powered, semantic depth).