Score Reels hooks and remove AI traces to manage Instagram content
This toolkit warns you if a hook is weak before you film, checks if captions get cut off in the feed, and removes signs of AI generation. It helps Instagram account operators automate quality control for content creation. The repository consists of 13 Claude skills and 5 Python scripts, and all operations run locally.
Pre-filming hook validation and timing calculation
The core feature is scoring Reels hooks against 26 formulas. hookscore.py evaluates hooks based on five criteria, including specificity, stake, and target designation, and points out weak areas specifically. For example, a hook that starts with a greeting is judged as 'begging for attention' and receives a low score. beats.py calculates the time it takes to speak each sentence in a script and warns that retention drops if a sentence exceeds three seconds. This allows you to catch long beats or parts lacking specific details before they become problems in the editing stage.
Feed visibility area and AI style correction
The Instagram feed displays only the first 125 characters of a caption and folds the rest, and caption.py visually displays this folded area. You can immediately check if the sentence the user actually sees is cut off in the middle or if key keywords are visible. Additionally, humanize.py and detect.py remove em dashes, special symbols, and cliché expressions like 'stop scrolling' that are common in AI-generated text. It replaces 154 standard AI terms with plain English and cleans up invisible Unicode characters to create a natural style.
Limitations and measurement results based on real data
The repository author tested the accuracy of the hook score using 74 real YouTube Shorts hooks. The results show that the tool is effective at filtering out clearly weak hooks (AUC 0.83), but it has limitations in distinguishing between hits and misses from good creators (AUC 0.56). This is because text analysis alone cannot predict viewer reactions or editing quality. Therefore, it is appropriate to use this tool as a 'filter to remove clearly bad hooks,' and final judgment should be left to your retention graph.
Manual posting principle and local execution environment
These skills do not post directly to Instagram. This is because Meta's content posting API requires complex procedures such as developer app approval, and browser automation risks account suspension due to terms of service violations. All skills end with copyable text blocks, and users must review and post them manually. Also, all Python scripts run locally without external network connections and have no dependencies. By creating a voice.md file to define your voice, you can configure all skills to follow this tone.
Pre-use checks and suitable audience
This repository is released under the MIT license and can be freely modified and distributed. It works best in the Claude Code environment and requires Python 3. It is suitable for creators who directly operate Instagram accounts and want to reduce the mechanical feel of AI-generated content. However, it does not claim to 'guarantee virality' or 'perfectly bypass AI detection.' Since platform policies change frequently, you should regularly check the latest regulations, such as the hashtag limit (changed to 5 on December 5, 2025). Since there is no automatic posting feature, it is recommended only for those who can accept the manual step from content creation to posting.