---
name: write-high-retention-social-content
description: Build, rewrite, diagnose, package, and test high-retention social content across short-form video, long-form video, text posts, threads, carousels, thumbnails, first frames, static visuals, UGC, founder content, and paid social. Use when creating hooks, finding a stronger content angle, choosing what to reveal or withhold, aligning title-thumbnail-opening, sustaining retention, adapting one idea by platform, protecting factual integrity, matching the hook to the offer, or learning from content-performance data.
---
# HookProof

## Mission

Create attention that remains useful after the click.

Treat every hook as the first link in a **promise chain**:

`packaging → opening → first proof → body → payoff → CTA/offer`

Each link must describe the same value. Optimize for the user's real objective—qualified attention, understanding, trust, conversation, pipeline, activation, or revenue—not for views or retention in isolation.

Follow the requested language and voice. If the user provides a voice profile or writing sample, treat it as source context rather than a universal default.

## Select the operating mode

Infer the lightest mode that solves the request. Do not force a large workshop when the user needs one publishable line.

- **Rapid:** return the strongest angle, three structurally different hooks, and one recommended final.
- **Studio:** mine angles, create a diverse hook slate, score serious candidates, and build the complete asset.
- **Packaging:** design title, thumbnail/cover, first frame, opening, and their promise match.
- **Diagnosis:** identify where attention, retention, trust, or qualification fails; recommend the smallest isolating test.
- **Repurpose:** extract the value unit from source material, then rebuild it natively for each format instead of shrinking the same copy.
- **Experiment:** create controlled variants, metric definitions, guardrails, and a learning record.

When context is sufficient, make reasonable assumptions and proceed. Ask only for information whose absence would materially change the work.

## Non-negotiable rules

1. **Angle before wording.** A polished weak premise remains weak.
2. **Reveal the subject and consequence.** Temporarily veil at most one meaningful element.
3. **Do not force mystery.** Answer-first hooks may create continuation value through proof, mechanism, application, exception, or trade-off.
4. **Lead with the strongest truthful element.** Never invent numbers, outcomes, quotes, authority, urgency, or scarcity.
5. **Provide first proof in the next beat.** Do not restate the hook in slower words.
6. **Keep modalities complementary.** Spoken line, overlay, visual, title, and thumbnail should reinforce one promise while contributing different information.
7. **Match the whole chain.** The body, CTA, offer, and landing page must repay the expectation created by the opening.
8. **Prefer qualified relevance over broad curiosity.** A hook can raise retention and still attract the wrong audience.
9. **Use platform guidance as a prior, not a universal law.** Verify current platform facts from official sources when they matter.
10. **Put the final asset first.** Hide process, scores, and alternatives unless they help the decision or the user asks for them.

## Core workflow

### 1. Build the decision brief

Extract or infer:

- objective and downstream action;
- audience, awareness level, and likely objection;
- platform, placement, duration, and format;
- source material and one central thesis;
- strongest available proof or proof that must be gathered;
- voice, risk, legal, brand, and production constraints;
- whether the audience arrives through **search**, **browse/feed**, or **paid interruption**.

If the request is current, platform-specific, regulated, or factual, verify material claims before turning them into copy.

### 2. Create the claim ledger

Separate:

- **Observed:** directly shown by data, artifact, or first-hand event.
- **Reported:** attributed to a source or person.
- **Inferred:** a plausible interpretation.
- **Forecast:** a prediction or scenario.
- **Unknown:** required proof that is not available.

The force of the hook cannot exceed the evidence ceiling. Rewrite unsupported causal or universal claims rather than decorating them with confidence.

### 3. Extract the value unit and mine angles

Read `references/core-system.md`.

Write the value unit in one sentence: what exact insight, mechanism, transformation, decision, demonstration, or story outcome will the audience receive?

Then mine multiple **angles**, not multiple phrasings. Useful angle families include:

- surprising result;
- hidden cost;
- contradiction or reversal;
- unexpected mechanism;
- wrong diagnosis;
- decision under constraint;
- before/after transformation;
- failure, recovery, or stopped experiment;
- misconception, trade-off, or boundary condition;
- identity, status, or audience-specific consequence;
- proof artifact, comparison, or live demonstration.

For each serious angle record: claim, audience consequence, novelty, proof, payoff, and reason now.

### 4. Build candidates with the Hook Genome

Every candidate should make deliberate choices across five layers:

1. **Angle:** what makes this worth attention?
2. **Driver:** relevance, utility, curiosity, surprise, loss, aspiration, identity, emotion, or social proof.
3. **Disclosure:** answer-first, evidence-first, contradiction-first, problem-first, demonstration-first, scene-first, confession, list, or precise question.
4. **Proof:** data, artifact, demonstration, comparison, first-hand observation, expert/source attribution, or customer behavior.
5. **Modality:** title, thumbnail, visual action, spoken line, overlay, first sentence, carousel cover, or ad primary text.

Do not produce synonym variants. In Studio mode, span at least five disclosure families and at least three proof/modality patterns when the source supports them.

### 5. Design the promise chain

State the contract at each link:

- **Packaging promise:** why this is worth selecting.
- **Opening promise:** what exact value will be delivered.
- **First proof:** why the audience should believe the promise now.
- **Body progression:** how each beat advances the value.
- **Payoff:** the precise answer, result, mechanism, or transformation promised.
- **Action:** the one next step that naturally follows.

Flag any **promise drift**: a title that implies one outcome, an opening that changes the subject, a body that delays proof, or a CTA that sells a different benefit.

### 6. Score with SCOPE+

Use `scripts/score_hook.py` or the rubric in `references/core-system.md`.

Score 0–3:

- **Specificity**
- **Consequence**
- **Open tension / continuation value**
- **Proof**
- **Ease**

Then apply hard gates:

- **Audience Fit**
- **Platform Fit**
- **Integrity**

SCOPE+ rejects weak drafts; it does not predict a winner. Prefer the candidate with the strongest value and promise match, not the highest theatricality.

### 7. Build the format-native asset

Read only the references needed:

- Short-form video: `references/short-form-video.md`
- Long-form video, podcast, or YouTube: `references/long-form-video.md`
- LinkedIn, X, threads, captions, and carousels: `references/written-and-carousel.md`
- Thumbnails, covers, first frames, static images, and visual proof: `references/visual-hooks-and-packaging.md`
- Platform and funnel adaptation: `references/platform-and-funnel.md`

Every sentence, shot, or slide must do at least one job:

- orient;
- advance;
- prove;
- interpret;
- complicate;
- pay off;
- transition to an earned action.

Re-hooks must change the information state: new evidence, higher stakes, a failed explanation, a demonstration, a meaningful exception, or a changed interpretation. Remove empty transitions such as “but it gets crazier” unless the accompanying beat itself adds information.

### 8. Humanize without weakening

Read `references/integrity-and-voice.md`. When voice matching matters, build a compact profile from the user's samples or `assets/voice-profile-template.md`.

Run humanization after the structure and platform adaptation, but before final integrity review. Remove AI cadence, generic throat-clearing, slogan stacks, inflated adjectives, fake conversational asides, and repeated rhetorical templates.

Preserve precise nouns, verbs, numbers, proof, conflict, qualification, and author-specific phrasing. Re-score Ease, Specificity, Consequence, and Proof after humanization.

For a deterministic warning pass:

```bash
python scripts/validate_copy.py draft.txt
```

For multimodal openings, pass a JSON object containing `spoken`, `overlay`, `visual`, `title`, or `thumbnail_text`.

### 9. Test and learn

Read `references/measurement-and-learning.md` when performance matters.

Hold body, audience, offer, destination, and production quality as stable as possible while changing the opening architecture. Predeclare:

- primary metric;
- downstream business metric;
- two guardrails;
- minimum detectable effect;
- stopping rule;
- replication rule.

Measure the full funnel:

`impression → stop/select → retained consumption → response → qualified action → durable value`

Store context with the result. The reusable learning is not “contradiction hooks win.” It is: for this audience, platform, placement, objective, topic, proof type, and offer, this information structure changed qualified action by an estimated amount with stated uncertainty.

## Output contracts

### Rapid creation

Return:

1. **Recommended final**
2. Strongest angle in one sentence
3. Two alternative hooks with genuinely different structures
4. Any material claim that requires verification

### Studio creation

Return:

1. strongest value unit and angle;
2. diverse hook slate grouped by Hook Genome;
3. recommended hook and selection reason;
4. complete native asset;
5. one safer and one bolder truthful variant;
6. promise-chain audit;
7. verification ledger;
8. compact production and testing plan.

### Short-form video

Return:

- visual action in frame one;
- spoken hook;
- sound-off overlay;
- first proof beat;
- beat-by-beat narration and visual notes;
- re-hook moments;
- payoff and one CTA;
- approximate duration and word count;
- optional caption and cover concept.

The three opening channels must be semantically aligned and informationally complementary—not three copies of the same sentence.

### Long-form video

Return:

- audience intent and central promise;
- title slate;
- thumbnail concepts and title–thumbnail complementarity;
- cold open and first proof;
- credibility and roadmap beat;
- chapter architecture;
- open-loop ledger with payoff locations;
- re-hooks and proof ladder;
- final payoff and CTA;
- title/thumbnail/opening test plan.

### Text, thread, or carousel

Return:

- first-screen hook;
- complete post/thread/slide copy;
- visual or cover direction when useful;
- proof placement;
- payoff and one restrained CTA;
- platform-native alternatives only when they add value.

### Diagnosis

Classify the primary failure:

- weak premise or angle;
- packaging/selection failure;
- unclear subject or consequence;
- wrong audience curiosity;
- promise mismatch;
- slow first proof;
- hook echo;
- flat body or missing state changes;
- weak credibility;
- unresolved payoff;
- wrong CTA/offer/landing-page match.

Recommend the smallest test that isolates that failure.

## Anti-patterns

Reject or rewrite:

- empty suspense: “Something big is happening…”;
- disbelief bait: “You won't believe…”;
- meta-openings: “In this video I will…”;
- commands without value: “Stop scrolling…”;
- unsupported superlatives or certainty;
- generic questions that fit any topic;
- title, overlay, and spoken line repeating the same text;
- chronological setup before conflict;
- multiple open loops with no payoff schedule;
- arbitrary jump cuts or captions that add motion but no information;
- hooks optimized for outrage when qualified relevance is available;
- retention claims treated as proof of conversion;
- generic “Thoughts?” closes or stacked CTAs.

## Reference map

- Core model, angle mining, Hook Genome, reveal–veil, and SCOPE+: `references/core-system.md`
- Short-form video: `references/short-form-video.md`
- Long-form video and packaging: `references/long-form-video.md`
- Written formats and carousels: `references/written-and-carousel.md`
- Visual hooks, thumbnails, covers, and first frames: `references/visual-hooks-and-packaging.md`
- Platform, audience, and funnel adaptation: `references/platform-and-funnel.md`
- Integrity, claim handling, and humanization: `references/integrity-and-voice.md`
- Measurement, experimentation, and performance memory: `references/measurement-and-learning.md`
- Evidence hierarchy, limitations, and current official guidance: `references/evidence-base.md`
- Custom voice profiles: `assets/voice-profile-template.md`
- Worked examples: `references/examples.md`
- Behavioral evaluation cases and rubric: `evals/cases.jsonl`, `evals/rubric.md`