The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill requires and accesses a sensitive API key from the execution environment to authenticate with a third-party service.
Evidence: exa.js (line 5) reads the EXA_API_KEY from process.env.
Evidence: SKILL.md instructs the user to set this environment variable for the skill to function.
[DATA_EXFILTRATION]: User-provided search queries are transmitted to an external domain, which may lead to the exposure of sensitive information if queries contain private data.
Evidence: exa.js (line 12) transmits the query data to the exa.ai API endpoint using the exa-js client.
[PROMPT_INJECTION]: The skill retrieves untrusted content from the web and injects it into the agent's context, creating a surface for indirect prompt injection.
Ingestion points: exa.js (line 12) receives search result content from the Exa API response.
Boundary markers: Absent. The results are printed directly as JSON without delimiters or instructions to ignore embedded commands.
Capability inventory: The skill outputs data to standard output (stdout), which is then consumed by the agent's primary reasoning loop.
Sanitization: Absent. The skill does not filter or escape the content returned by the search engine.

exa