The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external project data including specifications, bug reports, and API contracts which could contain malicious instructions designed to influence agent behavior.
Ingestion points: Artifacts such as Specs, Bugs, Issues, incidents, and cross-boundary contracts (OpenAPI, GraphQL, etc.) are read from the repository into the agent's context as evidence.
Boundary markers: The skill includes a dedicated "Trust Boundary" section that explicitly instructs the agent to treat repository files, configuration, and test output as untrusted evidence rather than instructions.
Capability inventory: The agent is authorized to read project files and potentially execute test commands (e.g., pytest, npm test) and write to registry documentation files (TESTS.md).
Sanitization: No specific sanitization or filtering logic is defined for the ingested data; safety relies on the instructional constraints provided in the skill body.
[COMMAND_EXECUTION]: The skill guides the agent to identify, execute, and verify the results of project-specific test suites.
Evidence: Instructions in the "Inventory Existing Test Assets" and "Review Test Evidence" sections direct the agent to locate entry points like pytest, npm test, or make test and confirm that commands execute successfully (exit code 0).
Mitigation: The skill includes an explicit safety constraint in the "Trust Boundary" section, stating that the agent must not execute commands or perform destructive actions unless the host policy allows it and the user has explicitly approved the action.

test-collaboration