The Agent Skills Directory

[PROMPT_INJECTION]: Indirect prompt injection surface identified where the skill processes untrusted external data and reflects it back into responses.
Ingestion points: Request handlers in SKILL.md read data from request.json(), url.searchParams, and cookies.
Boundary markers: No delimiters or warnings are used to prevent the agent from executing instructions that might be embedded in the mocked data.
Capability inventory: The skill is limited to API mocking and does not include capabilities for arbitrary code execution, file writes, or unexpected network requests.
Sanitization: Example code interpolates query parameters directly into strings (e.g., Result for "${query}") without sanitization, which could lead to instruction injection if the mocked response is subsequently processed by the LLM.
[COMMAND_EXECUTION]: Instructions include standard shell commands for environment setup.
Evidence: Recommends executing npm install msw and npx msw init public/ for tool installation and initialization.

msw