The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill provides instructions for the agent to generate and execute Python scripts on the fly using heredocs (python3 << 'PYEOF') or command-line one-liners (python3 -c). These scripts are used to orchestrate tool calls and filter data, which involves runtime code generation and execution.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data retrieved from the open web via search results. This creates a surface for indirect prompt injection attacks where malicious content on a web page could influence the agent's downstream reasoning or actions.
Ingestion points: Web content is retrieved through the tvly search and tvly extract commands as documented in SKILL.md.
Boundary markers: The skill does not mandate specific boundary markers or 'ignore' instructions for the content printed to the context, relying on the agent's manual formatting.
Capability inventory: The generated scripts have access to the subprocess module for command execution and the ability to read and write to the local filesystem (e.g., /tmp/tavily_results.json).
Sanitization: There is no automated sanitization or escaping of the retrieved web content before it is printed into the agent's context window.
[EXTERNAL_DOWNLOADS]: The skill refers to the installation of the tavily-cli package via pip or uv. This tool is provided by Tavily, which is a well-known service for AI-powered search and data extraction.

tavily-dynamic-search