The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources which could contain malicious instructions designed to influence the agent's output.
Ingestion points: Technical snippets, patent titles, and claims are retrieved from SerpAPI (Google Patents) and Exa.ai during the Prior Art Search phase (SKILL.md).
Boundary markers: None identified; the instructions do not specify delimiters or warnings to ignore instructions found within external patent descriptions.
Capability inventory: The agent can execute curl commands and interact with the lark-cli to create or modify cloud documents.
Sanitization: No explicit sanitization or filtering of the retrieved external content is performed before it is used to draft the application form.
[COMMAND_EXECUTION]: The skill uses shell commands to interact with external APIs, creating a potential surface for command injection.
Evidence: Phase 2 (SKILL.md) provides templates for curl commands that interpolate user-provided technical concepts and search keywords into a shell execution context.
Environment Variables: The skill utilizes ${SERPAPI_KEY} and ${EXA_API_KEY} within these commands for authentication.
[EXTERNAL_DOWNLOADS]: The skill connects to external services to retrieve patent data.
Evidence: Fetches data from serpapi.com and api.exa.ai. These are established services for search and patent retrieval.

patent-architect