Shorts Script Audit: Hook, Dead Zones, Fix List

How it works

Shorts Hook & Retention Audit takes a short-form video script you already have (YouTube Shorts, Reels or TikTok, about 15-60 seconds, with or without timing) and audits it before you film. Four independent reviewers check it blind, cross-examine each other, then give one board verdict. It audits; it does not write the video. (To write a twist story from a seed, use a script-writing Skill such as Twist-Ending Shortform Script Planner. This one audits any existing script.)

Phase 1 — Beats. Splits the script into numbered beats and flags a dead zone when more than 5 seconds pass without a new visual, new information or a turn.

Phase 2 — Four blind reviewers. Hook Scout (first 3 seconds: hook type, promise clarity, first-word weight), Pacing Editor (beats, speech and on-screen text words per second), Payoff Auditor (open loops, clickbait vs delivery), Cold-Viewer Skeptic (sound-off clarity, CTA placement, ending).

Phase 3 — Cross-examination. Each reviewer challenges or corroborates the others, so a finding is CORROBORATED, CONFIRMED or DISPUTED.

Phase 4 — Verdict. A beat table with keep/tighten/cut, 3 alternative hook lines (each labeled by hook type, each promising only what your script delivers), and a prioritized fix list (P1-P3).

All thresholds are labeled heuristics, not platform facts. No views or revenue are promised. One-time download: no hosting, no API key.

Sample output

Real, unedited output from running this Skill on a sample problem:

YouTube Shorts | 40 s (user timestamps) | hook type: None | delivery: PARTIAL (2 of 3) | 2 dead zones

## Intake
- Platform: YouTube Shorts. Title/promise: "The 3 mistakes killing your sourdough starter". Timing from the user's timestamps (not estimated). Spoken words: 129 (avg 3.2 words/s, high; heuristic).
- Visuals and on-screen text are marked, so text load is measured. (assumed) the title is shown as the video title, not on screen.

## First 3 seconds (heuristic)
- Hook type: None (greeting plus topic announcement: "Hey guys, welcome back to the channel").
- Promise in the spoken hook: Missing. "3 mistakes" appears only in the title. First deliverable arrives at 0:12; the problem ("why your starter keeps dying") is named at about 0:07.
- First word: "Hey" = Filler. Hook delay: about 5 s of greeting and channel intro before anything is promised.
- On-screen text "Sourdough Starter Mistakes" matches the topic, but states no payoff.

## Beat table
| Beat | Time | Content | New | Words/s | Text w/s | Flag | Verdict |
|---|---|---|---|---|---|---|---|
| B1 | 0:00-0:05 | Greeting, topic announced | V (opening shot), I-weak | 2.6 | 0.6 | filler opener | TIGHTEN |
| B2 | 0:05-0:12 | "Mine died three times" backstory | I-weak | 3.6 | - | DEAD 0:05-0:12 (7 s, only weak); speech above 3.2 | TIGHTEN |
| B3 | 0:12-0:19 | Mistake 1: chlorinated tap water | V, I, T | 2.9 | 2.6 (18 words on screen) | text above 12 words; duplicates voiceover | TIGHTEN |
| B4 | 0:19-0:21 | "Easy fix." pour water | V | 1.0 | - | none | KEEP |
| B5 | 0:21-0:30 | Mistake 2: fridge, never fed | V, I, T | 3.3 | 0.6 | 9 s, two ideas (problem and fix); speech above 3.2 | TIGHTEN |
| B6 | 0:30-0:35 | CTA: follow plus bio link | none | 4.2 | - | DEAD 0:30-0:40 (10 s, only weak); 2 asks; before promised item 3 | TIGHTEN (move) |
| B7 | 0:35-0:40 | Sign-off, loaf shot | V-weak | 3.6 | - | pure sign-off | CUT |

## Open loops and payoff
- Promise (title): 3 mistakes killing a starter. Spoken promise: none.
- L1 "why your starter keeps dying" opens B2 (0:05), partly closed B3 (0:19), B5 (0:30).
- L2 "3 mistakes" (title) opens at 0:00, **NEVER CLOSED**: mistake 3 never appears.
- Delivery verdict: **PARTIAL** (2 of 3 delivered; "Mistake number one/two" in speech sets up a third that is missing).

## CTA and clarity
- CTA at 0:30 (75% of runtime, est. from timestamps). 2 asks (follow, link). It arrives before the promised third item, so it is premature. It relates to starters, so topic fit is fine.
- Sound-off test: on-screen text B3 carries the fix; B5 text ("Fridge, never fed") has no fix; B1 text has no promise. Sound-off viewers get mistake 1 only.
- Ending: trailing sign-off ("go bake some bread, thanks for watching") after the CTA; no loop back to the hook.

## Board verdict
- CORROBORATED (R1, R3, R4): the hook is a greeting and no promise is spoken; "3 mistakes" lives only in the title (B1, B2).
- CORROBORATED (R3, R4): promise PARTIAL; mistake 3 is missing and the CTA at 0:30 lands before it.
- CORROBORATED (R2, R1): B2 is a backstory dead zone from 0:05 to 0:12.
- CONFIRMED (R2): B6-B7 form a 10 s tail with no non-weak new element.
- CONFIRMED (R2): B3 puts 18 words on screen at once and repeats the voiceover; B2/B5 speech runs 3.3-3.6 words/s (heuristic above 3.2).
- DISPUTED: R2 flags B2's "mine died three times" as weak; R1 argues it is the only personal stake and could become the hook. Creator decides; the hook options below use it.

## 3 alternative hooks
1. [Bold claim] "Your tap water might be killing your starter." Backed by B3 (chlorinated tap water).
2. [Open loop] "I killed my starter three times. These two mistakes are why." Backed by B2, B3, B5 (confirm the creator's own deaths came from these two; the script does not say so).
3. [Pattern interrupt] "Stop leaving your starter in the fridge without feeding it." Backed by B5.

## Fix list
1. P1 (B1, B2): replace the greeting with one of the hooks above; state the count aloud. Why: promise is missing for the first 12 s. effort: small.
2. P1 (promise, B5-B6): either add mistake 3 as a new beat before the CTA, or change the title to "2 mistakes". Why: PARTIAL delivery, loop never closed. effort: medium (add content) or small (retitle).
3. P2 (B2): cut to one short line ("Mine died three times.") or cut entirely, which removes the 0:05-0:12 dead zone. effort: small.
4. P2 (B6, B7): move the CTA after the final mistake, keep one ask (the bio link), and cut the sign-off. effort: small.
5. P3 (B3): shorten on-screen text to "Mistake 1: Tap water" (4 words) and let the voiceover carry the fix. effort: small.
6. P3 (B5): split into "what" (fridge, never fed) and "fix" (feed weekly) with a new visual on the fix; trim speech under about 3.2 words/s. effort: medium.

## Validation
1. Beats B1-B7 cover 0:00-0:40 with no gap: pass.
2. DEAD flags: 0:05-0:12 (7 s) and 0:30-0:40 (10 s), both over 5.0 s: pass.
3. Findings cite beats or quotes: pass.
4. 3 hooks, 3 types (Bold claim, Open loop, Pattern interrupt), all at most 12 words, none the original type (None), first words Heavy (Your, I, Stop): pass. Hook 2 carries a "confirm" note because the script does not link the creator's own deaths to these mistakes.
5. All timings from user timestamps; no estimates needed: pass.
6. No platform-behavior, view or revenue claims: pass.
7. Fix list ordered P1 first, 6 items: pass.

## Notes
- Confirm: the title is not shown on screen; whether mistake 3 exists in your plan.
- Thresholds (3 s hook window, 5 s dead zone, 3.2 and 3.0 words/s, 12 words on screen) are heuristics. This reads the script only, not the finished video; your own analytics outrank these numbers.

Want a re-audit after you revise the script?

Use cases

• Checking a script's first 3 seconds before you spend time filming
• Finding the stretch where nothing changes and viewers may drift
• Catching a title or hook that promises more than the video delivers
• Spotting too much on-screen text, a CTA that comes too early, or a trailing sign-off

FAQ

Does it write my script? No. It audits an existing script and suggests 3 hook lines and fixes.

Will it improve my views? It makes no promise. It finds structural weak spots by common editing rules of thumb, which it labels as heuristics. Your own analytics outrank them.

Does it watch the video? No, it reads the script text. Pacing is judged from your timestamps, or estimated and marked as an estimate.

Which tools? Any SKILL.md-compatible agent (Claude Code, Codex CLI, OpenClaw). No key, no hosting.