The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external library documentation through the Context7 MCP server. This creates an attack surface where malicious instructions could be embedded in documentation pages (e.g., hidden HTML comments or markdown) to influence the agent's behavior.
Ingestion points: The query-docs tool fetches documentation and code snippets from remote library IDs (e.g., /vercel/next.js).
Boundary markers: None provided in the instructions to help the agent distinguish between documentation text and potential embedded instructions.
Capability inventory: The agent uses the fetched data to generate code examples and answer user questions. If the agent has other active skills with file-write or command execution capabilities, these could be targeted.
Sanitization: The skill lacks explicit sanitization of the fetched documentation content, though it does instruct the agent to redact user secrets before sending queries to the service.
[DATA_EXPOSURE]: The skill includes a 'Best Practices' section that explicitly warns the agent to redact API keys, passwords, and tokens before sending queries to the external service. This is a positive safety control that mitigates the risk of accidental credential exfiltration via the query parameters.

documentation-lookup