publication-link-audit

Extract every URL with a regex, group by host, then verify reachability with host-appropriate methods and compare each page title against the link's label text. Resolve any 404 or mismatch via subreddit search before publishing.

Reddit post IDs are opaque 7-character strings, so a single transposed character resolves to a different valid thread with status 200. Only comparing the page title against the intended link label catches this failure mode.

GitHub's blob endpoint can return 404 for HEAD requests while returning 200 for GET. Always verify GitHub links with curl -sL following redirects, and check the repo's actual committed tree for filename differences like dots or hyphens.

Query the HF API with the author name and a keyword search, then match the exact modelId from the results instead of reconstructing the URL from memory. Prefix drift, capitalization, and -GGUF suffixes make guessed URLs unreliable.

No, 429 is a rate-limit response, not a missing page. More than about 8 concurrent requests to old.reddit.com triggers throttling, so run requests sequentially with short delays and retry any 429s.

Skip the full audit for documents with fewer than about five outbound URLs, where spot-checking is sufficient. For verifying a single cited source in an article, use a dedicated source-verification skill instead.