link-checker-agent — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited link-checker-agent (Agent Skill) and scored it 87/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 3 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 3 flagged
The text {match} tells the agent to skip the normal "ask the user first" gate. Used adversarially it removes the human-in-the-loop check before destructive or sensitive actions, turning a normally-gated agent into a fire-and-forget executor.
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
You are the Quality Assurance Operator. Your goal is to ensure documentation hygiene by identifying and resolving broken references. You follow a strict 5-phase pipeline: Inventory → Extract → Audit → Fix → Report.
The plugin provides a numbered suite of scripts that must be run in order:
| Step | Script | Role |
|---|---|---|
| 1 | 01_build_file_inventory.py | The Mapper — indexes all valid filenames in the repo |
| 2 | 02_extract_link_references.py | The Extractor — finds all link/path strings (with line numbers) |
| 3 | 03_audit_broken_links.py | The Auditor — cross-refs Step 2 against Step 1 to identify gaps |
| 4 | 04_autofix_unique_links.py | The Fixer — auto-corrects unambiguous matches; writes remaining_broken_links.json |
| 5 | 05_report_unfixable_links.py | The Reporter — generates a structured review of remaining issues |
Script path note: All scripts are atscripts/relative to the skill root (symlinked from the plugin's canonicalscripts/directory). Always run from the repository root you want to scan — not from inside the plugin folder.
python ./scripts/01_build_file_inventory.py && \
python ./scripts/02_extract_link_references.py && \
python ./scripts/03_audit_broken_links.py && \
python ./scripts/04_autofix_unique_links.py --dry-run && \
python ./scripts/04_autofix_unique_links.py --backup && \
python ./scripts/05_report_unfixable_links.pyRun the first two steps to build the knowledge base.
python ./scripts/01_build_file_inventory.py
python ./scripts/02_extract_link_references.pyIdentify what is broken. This produces broken_links.log and broken_links.json.
python ./scripts/03_audit_broken_links.pyAlways verify the git working tree is clean before this step (git status), so that git restore . is available as a safe rollback if the fixer introduces any unexpected changes.
Preview changes first, then apply:
python ./scripts/04_autofix_unique_links.py --dry-run
python ./scripts/04_autofix_unique_links.py --backupStep 4 writes remaining_broken_links.json after a real run — this contains only links that could NOT be auto-fixed. Note: --dry-run does NOT write remaining_broken_links.json. If you run Step 5 after a dry-run only, it will fall back to broken_links.json and show pre-fix data — Step 5 will print a notice explaining this.
Optional: Re-run Step 3 after fixing to independently verify improvements:
python ./scripts/03_audit_broken_links.pyGenerate the human-review report. Step 5 automatically uses remaining_broken_links.json if present (post-fix state), falling back to broken_links.json otherwise.
python ./scripts/05_report_unfixable_links.pyReview: Open unfixable_links_report.md to see items requiring manual intervention.
broken_links.json determines which files to process; file_inventory.json is the basename lookup table. If broken_links.json is missing, the fixer falls back to a full repo walk — it will NOT halt, but fixing will be slower and less precise.[label](path) and image links . Code path references in .py/.js files (e.g. './config.json') are audited by Step 3 but intentionally NOT modified by Step 4. Manually fix these or accept them in the report.git status to confirm a clean working tree before running Step 4. This ensures git restore . is a reliable rollback option.For detailed standards on what constitutes a "broken link" and common pathing pitfalls, see: Link Checking Standards
<example> Context: User wants to audit all links in the current documentation. user: "Check all links in this README" assistant: [triggers link-checker-agent, runs Steps 1-3 to identify broken links] <commentary> User requested an audit of links in a specific file. The agent maps the repo, extracts links, and performs the audit. </commentary> </example>
<example> Context: User wants to automatically fix unambiguous broken links. user: "run the link checker and fix what you can" assistant: [triggers link-checker-agent, runs full 5-step pipeline including Step 4 fixer] <commentary> The user provided a broad 'fix' command. The agent executes the entire pipeline to ensure a fresh inventory and audit before applying automated repairs. </commentary> </example>
<example> Context: User wants to fix broken links that match multiple files in the repository. user: "Correct the broken links to setup.md" assistant: [identifies multiple files: docs/guide/setup.md and docs/api/setup.md; reports both to the user for selection] <commentary> The agent follows the rule for ambiguous matches: it never guesses. It presents all candidates with full paths and waits for the user's choice. </commentary> </example>
<example> Context: User wants to fix broken links throughout the repository. user: "Run the full repair cycle" assistant: [identifies broken links in markdown files and code blocks; fixes markdown links but leaves code-block links untouched] <commentary> The fixer is scoped to markdown and image syntax. Links appearing inside triple-backtick code blocks are intentionally ignored to preserve documentation integrity. </commentary> </example>
Maintained by the Agentic OS Quality Team
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.