The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the terminal tool to perform shell operations.
The SKILL.md instructions and the accompanying social_listening_sweep.py script both call external binaries including tinyfish and curl.
There is a risk of command injection in the social_listening_sweep.py script. It extracts URLs from external search results and places them directly into a shell command (curl -s "{url}") without sanitization, which could allow a malicious search result to execute arbitrary code.
[EXTERNAL_DOWNLOADS]: The skill is designed to interact with and download content from external web services.
It uses the tinyfish tool to search and fetch content from various platforms including Reddit, Hacker News, and general news sites.
It uses curl to retrieve content directly from YouTube URLs for metadata and date verification.
[PROMPT_INJECTION]: The skill processes untrusted content from the web, creating an indirect prompt injection attack surface.
Ingestion points: Data is fetched from various third-party websites via tinyfish search, tinyfish fetch, and curl.
Boundary markers: The instructions do not define clear delimiters or "ignore embedded instructions" directives for the fetched content before it is passed to the LLM for summarization.
Capability inventory: The skill relies on terminal() and execute_code() to perform its primary logic, which are powerful capabilities to expose to untrusted data.
Sanitization: There is no significant sanitization, escaping, or filtering of the fetched content before it is processed by the agent.

tinyfish-social-listening