Knowledge Health Checker#

Knowledge Health Checker audits a Markdown-based knowledge base as a living system, not a folder full of files.

It detects whether the knowledge garden is:

• connected or fragmented
• dense or hollow
• current or stale
• navigable or full of dead links
• safe to auto-fix or requiring human review

The goal is not only to find problems, but to produce a prioritized, safe, actionable health report.

---

When to Use#

Use this skill for:

• Obsidian vault cleanup
• Logseq / Notion Markdown export review
• documentation repository health checks
• wiki linting before migration or publishing
• broken link detection
• empty placeholder / TODO note detection
• orphan note and graph fragmentation analysis
• content density and structure quality review
• periodic knowledge-base maintenance

Do not use it for semantic fact-checking. This skill checks structure, links, density, freshness, and maintainability, not whether every claim is true.

---

Core Principle#

A healthy knowledge base has four properties:

1. Substance — notes contain enough content to be useful.
2. Connectivity — important notes are linked into the graph.
3. Navigability — links, headings, and structure help readers move through knowledge.
4. Maintainability — stale, broken, duplicate, or low-value content is visible and repairable.

A knowledge base can be large and still unhealthy. Size is not health.

---

Default Workflow#

Step 1: Confirm scope and safety#

Before scanning, identify:

text

Target path:
Formats: markdown / wiki links / relative links
External URL check: yes/no
Generate fix script: yes/no
Auto-apply fixes: no by default
Exclude directories:
Estimated file count:

Safe default:

text

scan only → report only → generate fix plan → user reviews → user applies

Never delete, rename, rewrite, or auto-apply fixes without explicit confirmation.

Step 2: Build file and heading index#

Index:

• .md files
• normalized filenames and aliases
• headings / anchors
• relative paths
• wiki links such as [[note]] and [[note#heading]]
• markdown links such as [text](path.md)

Exclude by default:

text

.git/
node_modules/
__pycache__/
.obsidian/
.trash/
dist/
build/

Step 3: Detect hollow or low-value notes#

Flag likely hollow notes when they match one or more:

• fewer than 200 characters
• no heading
• only TODO / placeholder text
• image-heavy with very little explanation
• template content not filled in
• empty exported page from Notion/Logseq

Classify severity:

SeverityMeaningTypical actionP0Empty or pure placeholderdelete, archive, or fill immediatelyP1Too thin to be usefulexpand with definition, context, examplesP2Usable but weakimprove structure or add links

Step 4: Detect broken links#

Check:

• wiki file links: [[filename]]
• wiki heading links: [[filename#heading]]
• local markdown links: [text](../path/file.md)
• image/embed paths
• optional external URLs, only with user confirmation because it can be slow/noisy

For each broken link, report:

text

source file
link text
target
link type
probable fix if a similar file exists

Step 5: Analyze content density and structure#

Measure:

• word/character count
• heading depth and hierarchy
• list/table/code-block usage
• internal link count
• external link count
• last modified time
• very long files that may need splitting
• files with no inbound or outbound links

Suggested ranges:

SignalHealthy rangeWarningShort note300+ words or intentionally atomic<200 charactersLong notestill navigable with headings>3000 words without structureInternal linksat least 1-3 for durable noteszero links = possible orphanFreshnessdepends on domainstale if >90 days and marked active

Step 6: Analyze knowledge graph health#

Build a graph:

text

node = markdown file
edge = internal link

Report:

• total nodes
• total edges
• orphan nodes
• central nodes
• weakly connected components
• one-way links
• fragmented topic clusters

A perfect graph is not required. The goal is to identify the highest-value repair points.

Step 7: Score health#

Default scoring:

DimensionWeightGood stateHollow note rate25%few or no empty placeholdersBroken link rate30%no broken internal linksContent density25%most notes have useful substance and structureNetwork connectivity20%important notes are connected; few accidental orphans

Health score:

text

health = weighted score from 0 to 100

Use labels:

ScoreLabel90-100Excellent75-89Healthy60-74Needs maintenance40-59Fragile0-39Critical

Step 8: Generate report and fix plan#

Return a concise summary first. For large scans, provide a full report path.

Fix plans must be safe:

• generate proposed changes
• group by risk
• include reason for each fix
• require user review before applying destructive changes

Never silently delete or rewrite knowledge files.

---

Output Format#

Use this format:

markdown

## Knowledge Health Summary
- Target:
- Files scanned:
- Health score:
- Label:
- Top risks:

## Findings
| Category | Count | Severity | Notes |
|---|---:|---|---|
| Hollow notes |  |  |  |
| Broken links |  |  |  |
| Orphan notes |  |  |  |
| Overlong notes |  |  |  |
| Stale active notes |  |  |  |

## Highest-Impact Fixes
1. P0:
2. P1:
3. P2:

## Safe Fix Plan
- Auto-safe fixes:
- Needs human review:
- Do not auto-apply:

## Artifacts
- Report:
- Fix script:
- Raw JSON:

For small knowledge bases, include concrete file examples. For large ones, include top 10 examples per category and write full details to a report file.

---

Safe Fix Policy#

Classify fixes by risk:

RiskExamplesPermissionLowgenerate report, list broken links, suggest linksno extra confirmationMediumcreate fix script, add missing backlinks in draft outputask before writing filesHighdelete notes, rename files, rewrite links globally, split filesexplicit confirmation required

Default behavior: report and propose, do not mutate.

---

Bundled Scripts#

Use these when available:

• scripts/health_check.py — core scanner for hollow files, broken links, density, and graph stats.
• scripts/report_generator.py — HTML report generation.
• scripts/auto_fix.py — fix-plan or repair-script generation.

Run scripts from the skill directory or pass absolute paths. If a script lacks CLI ergonomics, inspect it and adapt safely rather than guessing destructive behavior.

---

Example Commands#

Basic scan:

bash

python3 scripts/health_check.py /path/to/knowledge-base

Generate a report from scan results if supported:

bash

python3 scripts/report_generator.py results.json --output health-report.html

Generate a fix plan, not auto-apply:

bash

python3 scripts/auto_fix.py results.json --dry-run

If the bundled script does not support these exact flags, read the script first and use its actual interface.

---

Test Prompts#

Use test-prompts.json for Darwin-style regression evaluation. Good test coverage should include:

• small Markdown folder with broken links
• Obsidian-style wiki links and missing headings
• placeholder-heavy exported notes
• a large graph with orphan clusters
• request for safe fix plan without auto-apply

---

Anti-Patterns#

Avoid:

• equating more notes with better knowledge
• deleting or rewriting files without confirmation
• checking external URLs by default on large vaults
• treating all orphan notes as bad; some are intentionally private/draft
• creating huge reports with no prioritized next action
• producing a repair script without explaining risk
• ignoring non-English filenames and encodings

---

Quality Bar#

A good knowledge health check must be:

• safe: no destructive changes without confirmation
• specific: names files and link targets
• prioritized: P0/P1/P2, not a flat dump
• actionable: includes exact repair suggestions
• scalable: summarizes large vaults without flooding context
• portable: works for Obsidian, Logseq, Notion exports, and plain Markdown

If the output only says “you have broken links” without showing where, why it matters, and what to do next, it failed.