The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The script scripts/saved.py utilizes the yt-dlp library to extract session cookies from the user's local Google Chrome profile. This sensitive operation allows the skill to authenticate with Instagram's private web API to retrieve 'Saved' content without manual login.
[COMMAND_EXECUTION]: The skill performs its primary functions by executing a local Python script (scripts/saved.py) with various arguments (fetch, last, search).
[EXTERNAL_DOWNLOADS]: The skill depends on the installation of the third-party yt-dlp package to handle the extraction of browser cookies.
[PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by ingesting untrusted content from Instagram captions and account names. These are stored in a local JSON file which the agent is instructed to read directly.
Ingestion points: Data is retrieved from the Instagram API and saved to ~/.instagram-saves/saved.json via scripts/saved.py.
Boundary markers: The instructions lack specific delimiters or warnings to ignore instructions embedded within the fetched captions.
Capability inventory: The skill provides capabilities for local file reading and script execution, which could be abused if the agent obeys instructions hidden in captions.
Sanitization: Captions are truncated to 800 characters in the media_to_entry function, but no validation or sanitization of the content is performed to prevent prompt injection.

insta-saves