Audit and correct VeOmni documentation links, Markdown, language, and technical claims against repository sources. Use for documentation scans, broken links, typos, or stale instructions in README, docs, examples, and agent guidance.
72
91%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
Passed
No findings from the security scan
Audit documentation and apply only corrections supported by evidence. This skill is self-contained: it requires no bundled script, extra Python package, GPU, or network connection. Use the repository's available search, build, and validation tools; do not imply that a manual search is an exhaustive parser-based scan.
AGENTS.md. A review-only request calls for findings, not edits. A correction request permits verified documentation fixes, not unrelated code changes or a push.git status --short before editing. Preserve unrelated changes. Use requested paths; otherwise inspect tracked Markdown/MDX in README.md, docs/, .agents/, and component documentation. Include relevant new unignored files.Enumerate candidates with git ls-files --cached --others --exclude-standard -- '*.md' '*.mdx'. Do not use rg --files, which skips hidden paths such as .agents/, or plain git ls-files, which omits untracked files. Search for Markdown links/images (including reference definitions), HTML a[href] and img[src], headings, code fences, and repeated words. Search output is a candidate list, not proof that all links were found: multiline Markdown, footnotes, comments, code spans, MyST directives, and HTML can defeat simple patterns. Use an available Markdown parser or renderer when comprehensive coverage matters, without introducing a permanent dependency just for this skill.
For each candidate link or image:
.., URL decoding, and exact filename case. Do not resolve every link from the repository root. Check whether an absolute-site path is handled by the documentation renderer.If a target is missing or ambiguous, search the repository for its intended replacement. Fix only when the destination is clear; otherwise report the uncertainty with file and line.
Review prose outside code blocks for objective spelling, grammar, punctuation, incomplete sentences, contradictory guidance, and outdated terminology. Verify commands, working directories, CLI flags, configuration paths and keys, defaults, versions, hardware claims, model registration examples, and API names against implementation, tests, current configs, CI, or Dockerfiles. rg definitions and call sites rather than relying on nearby prose. For procedures, check prerequisites and ordering; for examples, preserve executable quoting, indentation, and placeholders.
Apply a technical correction when an authoritative repository source proves it. Apply a language correction when there is only one reasonable reading. If evidence is insufficient, retain the original and report the candidate with the source checked and what remains uncertain. Do not turn a completed link check into a claim of semantic correctness.
python scripts/ci/check_doc_task_paths.py for task paths and python scripts/ci/check_agent_doc_paths.py for agent paths. When documentation dependencies are available, build Sphinx HTML and inspect warnings; for changed anchors, compare rendered links with actual generated IDs. A successful build alone does not prove every anchor is correct.git diff --check, and inspect the final diff. Do not install heavy training dependencies solely for documentation review.86c5ec1
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.