The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/search script is vulnerable to shell command injection. The user-provided $QUERY variable is interpolated into a curl command using double quotes (inside the -d data payload), which allows for the execution of arbitrary commands via subshell syntax like $(...) or backticks if the input contains these patterns.
[EXTERNAL_DOWNLOADS]: The skill connects to https://api.tavily.com/search, a well-known AI-optimized search service, to fulfill search requests.
[PROMPT_INJECTION]: The skill processes untrusted data from the web through search results, presenting an indirect prompt injection surface.
Ingestion points: External content fetched from the Tavily API and processed in scripts/search.
Boundary markers: No delimiters or warnings are used when printing the external content to the agent context.
Capability inventory: The content is printed to standard output and is not used in dangerous operations such as file writing or code execution sinks within the skill's logic.
Sanitization: The script uses JSON parsing to process the API response before printing.

tavily-search