The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to read and interpret repository-based artifacts such as instructions, briefs, and specifications to guide its implementation cycles. This creates an attack surface where malicious instructions embedded in the repository files could influence the agent's behavior.\n
Ingestion points: Repository instructions, authoritative tasks, briefs, specs, and review findings as described in the 'Contract gate' and 'Workflow' sections of SKILL.md.\n
Boundary markers: The instructions do not specify the use of clear delimiters or guardrails to distinguish between the agent's core instructions and the potentially untrusted data retrieved from the repository.\n
Capability inventory: The skill has the authority to write to the repository (creating implementation and test code) and execute local commands (detecting and running test suites).\n
Sanitization: There are no explicit instructions for the agent to sanitize, escape, or validate the content it reads from external repository files before processing them.\n\n- [COMMAND_EXECUTION]: The skill directs the agent to identify and execute the project's native test runners and tooling (e.g., Jest, Pytest, PHPUnit). It advocates for a high-autonomy workflow where the agent proceeds through the 'red-green-refactor' cycle without pausing for user approval once the initial task criteria are met, which relies on the agent's correct interpretation of the task boundaries.

tdd