The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data, including JSON bundles and Markdown PR bodies, representing a standard injection surface.
Ingestion points: The validate, reduce, and validate-pr commands in scripts/verification_contract.py read user-provided file paths.
Boundary markers: The skill enforces a rigorous JSON schema (verification-contract-v2.json) and specific Markdown structure requirements, effectively delimiting untrusted content.
Capability inventory: Operations are limited to data validation and deterministic reduction. The skill does not possess network capabilities or the ability to write to arbitrary files in its operational scripts.
Sanitization: Inputs are strictly validated against types, enums, and required fields defined in the contract. PR bodies are parsed using regex to ensure expected headings and blocks are present without executing the prose content.
[COMMAND_EXECUTION]: The test suite (tests/test_verification_contract.py) uses subprocess.run to verify the skill's own command-line interface.
Evidence: Found in the CliTests class within the test file.
Assessment: The commands are explicitly bound to the skill's internal Python script using sys.executable and do not utilize a shell for execution, adhering to secure coding practices for automated testing.

verification-audit