| name | flavor |
| description | Scan a blog post for unlinked pop culture refs, proper nouns, named theories, and historical figures. Research and apply hyperlinks automatically. |
| argument-hint | <file_path> |
| allowed-tools | Read, Edit, Grep, Glob, WebSearch, WebFetch, AskUserQuestion |
Flavor: Link Pop Culture and Proper Nouns
Scan a blog post for pop culture references, proper nouns, named theories, historical figures, and film/book/music references that lack hyperlinks. Research and apply links that give the reader instant context.
Process
- Read the file.
- Scan for every unlinked reference that a reader might not recognize.
- Research each reference. Use WebSearch to find the best link: YouTube trailers/clips for films, DOIs for papers, Wikipedia for historical figures. Verify the link exists and points to the right thing.
- Report each as:
L{line}: {type} — "{quoted text}" → [link] {why this link}
- Apply all fixes directly. Link inline, don't add footnotes.
- Report what was linked.
What to Link
Pop culture. Films, TV shows, books, albums, characters. Link to a YouTube trailer, clip, or scene when available. Prefer the most recognizable moment. If no YouTube link exists, use Wikipedia or IMDB.
Historical figures and events. People, battles, legislation, court cases. Link to Wikipedia or the most authoritative short summary.
Economic/scientific theories. Named theorems, models, papers. Link to the original paper (DOI preferred), or Wikipedia if the paper is paywalled.
Proper nouns the reader might not know. Companies, products, organizations, places that aren't household names. Link to the company site or Wikipedia.
What NOT to Link
- Household names that need no explanation (Google, Apple, Microsoft, etc.)
- Terms already linked elsewhere in the same post
- The author's own posts (those are internal links, not flavor)
- Generic concepts that don't have a single canonical reference
- Terms that appear as examples, not as arguments. If a reference is illustrating a point made elsewhere (e.g., "prions" in a post about courage, not about biology), linking it adds context the reader doesn't need for this post's argument. Only link references that substantiate the argument being made.
Link Waterfall
For each reference, try link types in this order. Use the first one that exists and fits:
- Video / interactive — YouTube clip, trailer, demo, or interactive explainer. Most engaging; the reader gets context in 60 seconds.
- High-quality article — A well-written news piece, blog post, or profile (TechCrunch, Stratechery, Ars Technica, etc.) that gives narrative context, not just facts.
- Official page — Company site, product page, project README, or conference landing page. Good when the thing is the product.
- Wikipedia — Reliable and fast-loading. Best for historical figures, named theories, established companies.
- Academic paper — arXiv, DOI, or SSRN. Use when the post is citing research and the reader might want to verify the claim. Prefer open-access links.
- Hugging Face / GitHub — For models, libraries, and tools. The reader can see the thing itself.
Move down the waterfall when higher options don't exist or feel forced. A Vickrey auction deserves Wikipedia, not a YouTube video. A startup deserves its launch post, not a dry Crunchbase page.
Judgment Calls
- For references that work as jokes or Easter eggs, the link IS the joke. A reader who clicks should smile.
- When multiple link options exist, pick the one that loads fastest and requires no login.
- One link per reference. Don't double-link the same thing.
- Match the link's tone to the post's tone. Technical posts can link to papers. Narrative posts should prefer articles and videos.
Dampener
Default to "probably leave it alone." The skill should barely run — a typical pass finds one or two genuinely useful links, not five borderline ones. If the candidate fails any of these:
- Would the reader's understanding actually change after clicking?
- Is the reference doing argumentative work in this paragraph, or just appearing in it?
- Does the link substantiate a claim the reader might otherwise doubt?
… leave it unlinked. Per-pass yield should taper toward zero by pass two. If a second pass finds another link to add, that's a sign of drift, not of thoroughness — flavor without damping is a slow walk toward link soup. The "a bit" qualifier on /tighten and the narrow-preserving invariant on /sharpen do the same job in their respective skills; flavor needs the same brake to satisfy the monoidal contract flavor(flavor(x)) == flavor(x).