Skip to content
Back to skills

Link Checker Agent

ASecurity

Specialized QA operator for documentation link integrity, auditing, and auto-repair across repository markdown files.

  • 7 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added September 2, 2026
ai-agentspythonbashgitdocumentation

Security analysis

A100/100

Pro scans all 15 files and shows the line behind each finding

Scanned October 3, 2026

npx -y skills add richfrem/agent-plugins-skills --skill link-checker-agent --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Link Checker Agent?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Link Checker Agent
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/richfrem-link-checker-agent-agent-plugins-skills/badge)](https://www.skillsdirectory.com/skills/richfrem-link-checker-agent-agent-plugins-skills)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: link-checker-agent
plugin: dev-utils
description: Specialized QA operator for documentation link integrity, auditing, and auto-repair across repository markdown files.
allowed-tools: Bash, Read, Write
---

# Link Checker Agent (`link-checker-agent`)

Audits and repairs broken documentation and image references across repositories using a strict 5-step pipeline.

## Contents

- [Critical Constraints](#critical-constraints)
- [Quick start](#quick-start)
- [Workflow](#workflow)
- [Verification](#verification)
- [References](#references)

## Critical Constraints

1. **Pipeline Order**: Run steps strictly in sequence: 1 -> 2 -> 3 -> 4 -> 5. Never skip inventory or audit.
2. **Clean Git State**: Verify git working tree is clean before running Step 4 (`--dry-run` first).
3. **Scope Restriction**: Step 4 only auto-corrects unambiguous markdown/image links; code paths in `.py`/`.js` are never modified.
4. **No Guessing**: Ambiguous link targets matching multiple files are left for manual user review in Step 5.

## Quick start

Execute the complete 5-step audit and autofix pipeline:

```bash
python3 scripts/01_build_file_inventory.py && \
python3 scripts/02_extract_link_references.py && \
python3 scripts/03_audit_broken_links.py && \
python3 scripts/04_autofix_unique_links.py --dry-run
```

## Workflow

1. **Step 1 (Inventory)**: Run `scripts/01_build_file_inventory.py` to index all valid filenames in repository.
2. **Step 2 (Extraction)**: Run `scripts/02_extract_link_references.py` to extract all link/path strings with line numbers.
3. **Step 3 (Audit)**: Run `scripts/03_audit_broken_links.py` to audit references against the file inventory.
4. **Step 4 (Auto-Fix)**: Run `scripts/04_autofix_unique_links.py --dry-run` to preview fixes, followed by `--backup` to apply.
5. **Step 5 (Report)**: Run `scripts/05_report_unfixable_links.py` to produce `unfixable_links_report.md`.

## Verification

Confirm zero remaining broken links or review unfixable report:

```bash
test -f broken_links.json && python3 -c "import json; d=json.load(open('broken_links.json')); print(f'Broken links: {len(d)}')"
```

## References

- [acceptance-criteria.md](references/acceptance-criteria.md) — QA verification criteria for the link checker pipeline.
- [fallback-tree.md](references/fallback-tree.md) — Failure triage and recovery tree for missing artifacts or broken scripts.

Files in this skill

  • SKILL.md6.3 KB
  • acceptance-criteria.md2.3 KB
  • assets/diagrams/link-checker-workflow.mmd53 B
  • assets/diagrams/logic.mmd37 B
  • assets/diagrams/workflow.mmd40 B
  • evals/evals.json707 B
  • evals/results.tsv301 B
  • fallback-tree.md2 KB
  • references/acceptance-criteria.md42 B
  • references/link-checker-workflow.mmd45 B
  • references/link_checker_workflow.mmd45 B
  • references/link_checker_workflow.png45 B
  • references/logic.mmd29 B
  • references/unpacking.mmd33 B
  • references/workflow.mmd32 B

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.