The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs several shell operations to manage files and retrieve system information.
In workflow.js, the agent is instructed to run security find-generic-password -s imgur-api -w to retrieve stored credentials.
In new_shortcodes.py, the subprocess.run function is used to execute rg (ripgrep) for searching the contents of the local Obsidian vault.
In workflow.js, git add and git commit commands are utilized to manage version control for the vault repository.
[CREDENTIALS_UNSAFE]: The skill programmatically accesses the system's secure storage to retrieve API secrets.
workflow.js explicitly directs the agent to fetch the Imgur Client ID from the macOS Keychain service named "imgur-api".
[DATA_EXFILTRATION]: Local image files are transmitted to an external third-party service.
workflow.js facilitates the upload of local thumbnail images (_thumb.jpg) from the user's disk to the Imgur API (api.imgur.com) or via a browser-based upload session.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content (Instagram captions) which could be used to influence agent behavior.
Ingestion points: Instagram captions and metadata are read from a local SQLite database (urls.db) and subsequently passed to other processing agents.
Boundary markers: None are present to delimit untrusted caption text from agent instructions.
Capability inventory: The skill possesses significant capabilities including system keychain access, file system write permissions, and network access.
Sanitization: There is no evidence of sanitization for the caption content before it is processed by the cleaning agent.

ingest-new-instagram-recipes