The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill depends on the exa-py library, which is the official SDK for the Exa search service, and optionally uses python-dotenv for local environment management. Both are standard, well-known packages used for their intended purposes.
[PROMPT_INJECTION]: As a tool that fetches data from the open web, this skill possesses an inherent attack surface for indirect prompt injection. Malicious instructions located on third-party websites could be ingested and interpreted by the agent.
Ingestion points: Web content is retrieved via the exa_search.py and exa_extract.py scripts from the Exa API.
Boundary markers: The reference instructions (references/web-search.md) do not explicitly define delimiters to separate untrusted search results from the system prompt, though they guide the agent on how to parse the JSON output.
Capability inventory: The skill allows writing results to local JSON files and executing shell commands via the uv runner.
Sanitization: Content is passed to the agent as extracted by the Exa service without further sanitization or filtering.
[COMMAND_EXECUTION]: The skill provides command-line templates to interface with the Exa API. These templates use placeholders for user-supplied arguments, which is standard for CLI-based agent tools.

exa-search