hook-audit

Brand context: !cat .agents/brand-content-context.md 2>/dev/null || echo "No brand context found. Run /brand-content-context first to set up."

Hook Audit

You audit hooks specifically and only. Different from content-audit, which scores six dimensions across whole pieces. This skill drills into the first 3 seconds — the part that controls skip rate, which is the algorithm's number-one ranking signal.

Brand context: Read .agents/brand-content-context.md first. Voice, audience symptoms, performance baselines (which hook type is currently working). If missing, run /brand-content-context.

When to use this skill vs content-audit

• content-audit: broader — scores hook strength as one of six dimensions across whole posts.
• hook-audit: narrower and deeper — only scores hooks. Use when the user has 10+ hooks to compare, or when content-audit flagged hook strength as the weakest dimension.

If the user has fewer than 5 hooks, use content-audit instead. The patterns only become visible across batches.

What you score

For each hook, score on four sub-dimensions (1 to 5 each):

1. Hook type fit

Is the hook clearly one of the five types (question, list, story, negative, contrarian)? Is the chosen type the right fit for the topic?

• 5 = clear hook type, well-executed, fits the topic
• 3 = hook type is OK but a different type would suit the topic better
• 1 = no clear hook type, or the type is forced and doesn't work for this topic

2. Symptom-first

Does it speak to a customer feeling/situation, or is it expert-talk?

• 5 = customer would say "yes, that's me" within 3 seconds
• 3 = mixed
• 1 = expert-talk, terminology-heavy, customer can't see themselves

3. Curiosity gap

Does the hook make someone need to know what comes next?

• 5 = strong open loop, brain commits to finding the answer
• 3 = mild curiosity, viewer might stay
• 1 = no curiosity, or worse, the hook spoils the answer

4. Demographic call-out

Does the hook call out a specific demographic, identity, life stage, or scenario?

• 5 = explicit call-out (athletes, tradies, 30-something moms, etc.) or scenario specificity (a specific time of day, a specific context)
• 3 = general but specific enough
• 1 = totally generic, could apply to anyone

Total: 4 to 20 per hook.

The audit report

For each hook, output:

HOOK: "[the original hook]"
Format: [DTC / carousel / B-roll / unknown]
Hook type: [Q / L / S / N / C]

SCORES:
- Hook type fit:        [1-5]
- Symptom-first:        [1-5]
- Curiosity gap:        [1-5]
- Demographic call-out: [1-5]
TOTAL: [4-20]

WHAT'S WORKING:
[1-2 lines, specific]

WHAT'S BROKEN:
[1-2 lines, specific]

REWRITE:
[Stronger version of the same hook, same type]

The aggregate report

After the per-hook breakdowns, produce:

HOOK AUDIT SUMMARY: [n] hooks

DIMENSION AVERAGES:
- Hook type fit:        [avg]
- Symptom-first:        [avg]
- Curiosity gap:        [avg]
- Demographic call-out: [avg]

HOOK TYPE BREAKDOWN
| Type       | Count | Avg total |
|------------|-------|-----------|
| Question   | [n]   | [avg]     |
| List       | [n]   | [avg]     |
| Story      | [n]   | [avg]     |
| Negative   | [n]   | [avg]     |
| Contrarian | [n]   | [avg]     |

WINNING TYPE FOR THIS BRAND: [type] — avg [score]
LOSING TYPE FOR THIS BRAND: [type] — avg [score]

PATTERNS OBSERVED:
- [pattern 1, with examples]
- [pattern 2, with examples]
- [pattern 3, with examples]

TOP 3 PRIORITY FIXES:
1. [most leverage]
2. [second]
3. [third]

WHAT TO DO NEXT:
[Specific recommendation. Usually: "Lean into [winning type], rewrite [losing type] hooks, and update brand context performance baselines."]

Patterns to look for in aggregate

These are the most common ones. Surface them when you see them:

"Every hook is symptom-first but no demographic call-out"

The brand has internalised the doctor-patient principle but is leaving reach on the table. Adding "if you're a [identity]" or "by [specific time of day]" to existing hooks would lift performance without changing the substance.

"Strong hook types but they're all the same type"

E.g. every hook is a question. Brand has found one type that works and is over-relying on it. The audience is going to fatigue. Diversify: keep the strongest type at 30-40% and add 1 to 2 others.

"Hooks are good but the topic mix is narrow"

Hooks score high but they're all about the same 2-3 things. Run pillars-and-topics to widen the topic surface.

"Expert-talk in product hooks, symptom-first in personal hooks"

Common for DTC brands where founders write the personal-brand content well but ad copy still sounds like a press release. Run symptom-first-messaging on the product-side hooks specifically.

"Strong contrarian hooks but body doesn't deliver"

The hook is a 5/5 but the body doesn't pay it off. Brand is writing checks the post can't cash. Audit body content, not just hooks. Switch to content-audit.

When you don't have enough info to score

If the user gives you a hook with no context (no topic, no format, no link), score it on the 4 dimensions you can see and flag the missing info. Don't refuse to audit — partial info still produces useful diagnoses.

If the user gives you screenshots, transcribe the hooks first, then audit.

Updating the brand context after an audit

After a hook audit, ask the user if they want to update the brand context's performance baselines with the findings:

• Best-performing hook type → put in the baselines section
• Worst-performing hook type → useful for posting-grid weighting

Updating the baselines makes future hook-generator calls more accurate.

Related skills

• content-audit: broader audit if hooks aren't the only issue
• hook-generator: the rewrite step after an audit
• symptom-first-messaging: focused fix for the symptom-first dimension
• brand-content-context: update baselines with the audit findings