agentleFS
priyanshu-aryapriyanshu-arya/LinkedIn_Automation/.claude/skills/extract-hook/SKILL.md

extract-hook

Skill0 starsChanged 59 days agoMITView on GitHub

Use when the user wants to reverse-engineer the hook formula behind an existing viral post, or explicitly invokes /extract-hook. Takes pasted post text (URL accepted as best-effort only, per REQUIREMENTS.md §27 — never treated as reliable on its own) and classifies its opening against the shared taxonomy in Content-Learnings/hook-formulas.md, returning a blank fill-in-the-blank template. If the formula isn't in the taxonomy yet, proposes a new entry citing the source post rather than forcing a bad match or refusing to classify. Never claims a match guarantees virality — pattern-matching only, not a scoring or drafting tool.

The file

We split this skill into the rules it gives an agent

12 rules. About 0 are common, 8 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈8

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If pasted text is provided: use it directly as ground truth.

    line 36
  2. Distinctive:

    If only a URL is given: attempt WebFetch on it.

    line 37
  3. Distinctive:

    Never silently proceed to classify a partial or gated fetch as if it were the whole post.

    line 42
  4. Distinctive:

    If neither is given, ask the user to paste the post.

    line 44
  5. Very specific:

    Read Content-Learnings/hook-formulas.md in full — every row, every run.

    line 47
  6. Distinctive:

    Never rely on a cached memory of the taxonomy, since it grows over time as this skill (and human promotions) add to it.

    line 48
  7. Very specific:

    If nothing fits well: never force a bad match, and never refuse to classify.

    line 71
  8. Very specific:

    If a new formula was proposed: say so plainly, note it was appended to Content-Learnings/hook-formulas.md as status: proposed, and that it …

    line 105
  9. Distinctive:

    Never fabricate what a post says beyond text actually provided — pasted text, or a WebFetch result the user has explicitly confirmed …

    line 113
  10. Distinctive:

    Never claim a WebFetch succeeded when it returned a login wall, a paywall, or obviously partial/stripped content — say so plainly instead.

    line 116
  11. Very specific:

    Never assert that matching a formula "guarantees" engagement or virality.

    line 118
  12. Distinctive:

    Never drop a hook that doesn't fit the taxonomy well, and never force it into a canonical formula just to avoid proposing a new one.

    line 128

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Hook Extractor (Phase 16)

Reverse-engineers the hook formula behind an existing post — pasted text, ideally someone else's viral post but any post works — and returns a blank, fill-in-the-blank template of that formula's structure. Classifies against the shared taxonomy in Content-Learnings/hook-formulas.md, which Post Writer (a later, separate build) also reads from to draft new posts. This skill never drafts a finished post itself — output is always a blank template, never ghostwritten content.

Taxonomy adapted (not copied verbatim) from the MIT-licensed sergebulaev/linkedin-skills project's linkedin-post-writer skill's formula reference — see Content-Learnings/hook-formulas.md's own header for the full attribution.

Reading third-party post content: see REQUIREMENTS.md §27 — pasted text is the primary, reliable input; a URL is optional metadata only, and a WebFetch attempt on it is best-effort convenience, never ground truth.

Arguments

  • post text (primary) — the user pastes the post they want classified, directly in the conversation.
  • URL (optional, best-effort only) — if the user gives a URL instead of or alongside pasted text, per §27: attempt a WebFetch, but never treat its result as reliable on its own.

Process

1. Get the input, per §27's convention

  • If pasted text is provided: use it directly as ground truth.
  • If only a URL is given: attempt WebFetch on it. Then look hard at what came back — a login wall, a truncated/partial extract, or missing the actual post body are all common outcomes, not edge cases. If the result looks truncated, gated, or otherwise unreliable, say so explicitly to the user and ask them to paste the full post text before proceeding. Never silently proceed to classify a partial or gated fetch as if it were the whole post.
  • If neither is given, ask the user to paste the post.

2. Load the taxonomy fresh

Read Content-Learnings/hook-formulas.md in full — every row, every run. Never rely on a cached memory of the taxonomy, since it grows over time as this skill (and human promotions) add to it.

3. Isolate the hook

The hook is the first 1-3 lines / the first visible fold — what a reader sees before "…see more". Quote it back to the user so there's no ambiguity about what's being classified.

4. Classify along four dimensions

  • Opening mechanic — how the first line grabs attention (claim, confession, number, contrarian statement, etc.).
  • Tension/curiosity device — what makes a reader want line 2 (a gap, a contradiction, unresolved stakes).
  • Pacing — line-break rhythm and sentence-length curve (short punchy fragments vs. a longer developing sentence).
  • Payoff device — how (or whether, within the hook itself) the tension starts to resolve.

5. Match against the taxonomy, or propose a new formula

Compare the four-dimension read against every canonical row in hook-formulas.md (never match against a proposed row — those aren't trusted yet). Pick the best real match.

If nothing fits well: never force a bad match, and never refuse to classify. Instead, append a new row to hook-formulas.md's table per that file's own "Adding a new formula" rule — next unused F<n>, status: proposed, source: extracted from <short post reference/date>, and a short illustrative excerpt of the hook line only (never the full post). This is a live edit to a shared file — make it for real, don't just describe it.

6. Produce the blank template

Strip the matched (or newly proposed) formula down to a fill-in-the-blank skeleton the user can reuse for a different post — never the source post's actual words, always blanked placeholders describing what goes in each slot.

Output shape

FORMULA: F10 — Contrarian + Historical Receipts (matched, canonical)
Line 1 (claim reversal): "[Everyone believes ___. Here's why that's wrong.]"
Line 2 (stakes): "[What it costs you to keep believing it]"
Body: [2-4 concrete evidence beats, one per short paragraph]
Payoff: [the reframe/insight that resolves the tension]
CTA: [question tied directly to the payoff, not generic]

(The exact slots vary by formula — this is illustrative of the shape, not a fixed universal schema. Build the slots from what the actual hook is doing, per step 4's four dimensions.)

Report back

  • Which formula matched, and whether it was canonical or newly proposed this run.
  • The blank fill-in-the-blank template (full output shape above).
  • If a new formula was proposed: say so plainly, note it was appended to Content-Learnings/hook-formulas.md as status: proposed, and that it needs explicit human promotion to canonical (e.g. during /review-drafts, or a manual edit) before Post Writer will draft from it — this skill does not promote its own proposals.

Hard rules

  • Never fabricate what a post says beyond text actually provided — pasted text, or a WebFetch result the user has explicitly confirmed matches what they see. No inferring the "rest of the post" from a login wall.
  • Never claim a WebFetch succeeded when it returned a login wall, a paywall, or obviously partial/stripped content — say so plainly instead.
  • Never assert that matching a formula "guarantees" engagement or virality. This is descriptive pattern-matching against past structure, not a scoring tool (that's /critique-draft's Viral Potential Score, a different mechanism) and not a prediction.
  • Store only a short illustrative excerpt of a source post in hook-formulas.md when proposing a new formula — the hook line only, never a full reproduction of someone else's post.
  • Output is always a blank template, never a finished ghostwritten post — turning a template into an actual new post is Post Writer's job, a separate, not-yet-built skill.
  • Never drop a hook that doesn't fit the taxonomy well, and never force it into a canonical formula just to avoid proposing a new one.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.