The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/generate_carousel.py uses subprocess.run() to execute a sibling skill script (asset-generator/scripts/generate_asset.py) to render each slide. The command is constructed as a list, which is a secure practice that prevents shell injection. This is a functional component of the skill's slide-generation pipeline.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted text and images from Instagram competitor posts to derive blueprints and hashtags. Malicious instructions hidden in these external captions could potentially influence the AI's generation behavior.
Ingestion points: External captions and media metadata are scraped from Instagram via Apify in scripts/research_carousels.py and scripts/research_hashtags.py.
Boundary markers: The prompt builder in scripts/generate_carousel.py uses [[headline]] delimiters and explicit instructions to the model to ignore these markers, providing some protection against content confusion.
Capability inventory: The skill has the ability to execute local scripts via subprocess.run, perform network requests via urllib.request, and write files to the local directory.
Sanitization: While the skill uses text normalization for hashtags and slugs, there is no advanced sanitization performed on the scraped caption text before it is used in analysis prompts.
[EXTERNAL_DOWNLOADS]: The skill fetches data and images from the Apify API and Instagram. These interactions target well-known services and are essential for the skill's primary research and content analysis features.
[CREDENTIALS_UNSAFE]: The skill manages sensitive API keys (Gemini and Apify) through environment variables or a local config.json file. It includes a utility script scripts/setup_key.py to help users manage these credentials without hardcoding them into the main instruction files.

carousel-generator