The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides explicit shell commands for the agent to execute, such as uv run and uvx, to perform web searches and fetch YouTube transcripts when built-in tools are not available.
[DYNAMIC_EXECUTION]: A fallback search mechanism uses an inline Python script provided via a shell heredoc to execute search queries using the DuckDuckGo library.
[EXTERNAL_DOWNLOADS]: The skill leverages the uv and uvx package managers to dynamically download and execute the ddgs and web-forager packages from the Python Package Index (PyPI) at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves ingesting and processing untrusted data from web searches and third-party websites, which presents a surface for indirect prompt injection.
Ingestion points: Web search results, external website content, and YouTube video captions retrieved via CLI tools or session tools.
Boundary markers: The skill includes explicit instructions for the agent to treat all fetched content as "untrusted evidence" and to follow the skill's workflow rather than instructions embedded in the retrieved content.
Capability inventory: The agent has the capability to execute shell commands, perform network requests, and read local files referenced in the instructions.
Sanitization: The skill relies on natural language instructions to separate user claims from verified facts and to ignore embedded instructions, rather than programmatic sanitization of the input data.

competitive-intel