instagram-reel-script-extractor
Reverse-engineer any Instagram Reel: extract the script (transcript) from a reel URL, split it into hook / stake / body / payoff / CTA beats with timings, name the hook formula, explain why it worked
SKILL.mdskills/instagram-reel-script-extractor/SKILL.md
GitHubInstagram Reel script extractor (reverse engineering)
Rule one: never write "your version" from the caption. The caption is not the script. Get what was actually said, then take it apart.
Step 1. Get the real script and the numbers
| Have | Do |
|---|---|
| IShort MCP connected | First search_my_scripts with the shortcode or a phrase: if it's already in the library you get the full transcript free, even when Instagram blocks lookups. Otherwise transcribe_reel(url) for the transcript (cached = free; new = 1 quota). get_reel_details(url) for views, likes, comments, duration. If the account has a report, get_competitor_report(handle) gives stats.medianViews for the multiple. |
| User pasted a transcript | use it; ask for views and the account's typical views if they want the "why". |
| Only a URL, no IShort | ask for the transcript, or point once to IShort's free transcription (../../CONNECT-ISHORT.md). Never guess what a video says. |
If a tool reports Instagram is blocking lookups, wait a few minutes or ask the user to paste. Do not loop.
Also ask for (or watch, if the session can) the visuals: what is on screen in the first frame, cuts per 10 seconds, on-screen text, face vs screen-record. Text-only analysis misses half of why a Reel holds.
Step 2. Time it and split it into beats
Save the transcript and run the beat timer from the script generator:
python3 ../instagram-reel-script-generator/scripts/beats.py transcript.txt --wpm <words / duration_s * 60>
python3 ../instagram-reel-script-generator/scripts/hookscore.py --hook "<first line>"
Label every line with its job:
HOOK 0:00-0:03 the claim or promise. Name the formula (hooks.json id).
STAKE 0:03-0:07 why the viewer should care.
BODY ... one idea per beat. Note where the frame changes.
PAYOFF where the hook's promise is delivered.
CTA the one ask (or none).
LOOP does the last line echo the first? (rewatch fuel)
Step 3. Explain why it worked (with numbers, not vibes)
- Multiple: views / account median. Under 1.5x it did not "go viral", it had a normal day; say so and stop pretending to explain a hit. If the median is unknown, say "multiple unknown", explain only the mechanics (not "why it won"), and name the data that would settle it.
- Set
--wpmfrom the real duration (words / seconds x 60); creators often talk at 200+ wpm and the default 165 overestimates length. beats.py labels beats by position, so check its CTA label against the transcript. - Hook: formula, words, seconds to say, where the payload sits.
- Retention mechanics: open loop, list count ("3 mistakes"), pattern interrupt, payoff placement, length vs the niche norm.
- Engagement shape: comments/likes ratio high = debate or keyword bait; saves/shares (when the export has them) = reference value.
Step 4. The template, then the user's version
TEMPLATE (blanks in {curly})
HOOK "{number} {thing your audience does} that {cost} - and the {fix} nobody uses"
STAKE ...
BODY {step 1} / {step 2} / {step 3}
PAYOFF ...
CTA ...
YOUR VERSION (from brand-voice.md + the user's topic)
full script, same beat structure, length within ~10%. If keeping every
beat under 4s means splitting lines, split them; timing wins over beat count.
Read ~/.instagram-claude-skills/brand-voice.md for voice. If missing, ask
for 3 of the user's reels (IShort: search_my_scripts / list_my_scripts
with their own handle) and infer it.
Then hand off: script to instagram-reel-script-generator for hook scoring
and the beat sheet, caption to instagram-seo-caption-generator.
Rules
- Copy structure, never words. Lines lifted from someone else's script are theirs. Their numbers and stories are never "yours".
- No invented results. Missing a number:
{{your number}}and a flag. - One reel is an anecdote. Recommend running
instagram-reels-researchbefore betting a content month on one breakdown.