The Agent Skills Directory

[DATA_EXPOSURE]: The skill manages authentication using environment variables (EXA_API_KEY and TAVILY_API_KEY). The implementation in scripts/read-credential.py and scripts/request ensures that these credentials are never hardcoded, written to project files, or exposed in command-line arguments or diagnostic logs. The API keys are only transmitted to the legitimate service endpoints: api.exa.ai and api.tavily.com.
[COMMAND_EXECUTION]: The skill utilizes a shell script wrapper (scripts/request) to perform network operations via curl. The script uses strict case matching for providers and operations, ensuring that only predefined API endpoints are called. Input to the APIs is passed via standard input and handled using @/dev/fd/3, which prevents shell injection vulnerabilities in the request body.
[INDIRECT_PROMPT_INJECTION]: By design, the skill ingests content from the open web through search and extraction tools. This constitutes an indirect prompt injection surface where external web content could attempt to influence the agent's behavior. However, the skill instructions include specific guidance for the agent to synthesize evidence, preserve uncertainty, and investigate source disagreement, which serves as a natural mitigation against blindly following instructions found in external data.

web-research