The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to analyze external codebases to identify test targets and implement testing strategies, which creates a vulnerability surface for indirect prompt injection.\n- Ingestion points: The skill workflow requires the agent to identify components, business logic, and API contracts from the existing codebase (SKILL.md).\n- Boundary markers: No instructions, delimiters, or warnings are provided to ensure the agent ignores or sanitizes instructions that might be embedded in the codebase being analyzed.\n- Capability inventory: The skill facilitates the creation of new files and the execution of shell commands such as npm test and npm test -- --coverage (SKILL.md).\n- Sanitization: No validation, filtering, or sanitization mechanisms are defined for the content processed from the codebase before it is used in test generation or execution.

designing-tests