validate
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to manipulation through malicious content embedded in the subject code or intent descriptions it evaluates.\n
- Ingestion points: The core logic in tests/validate.py reads file content using os.walk and Path.read_bytes, and SKILL.md indicates the agent reviews diffs and intent documentation.\n
- Boundary markers: SKILL.md explicitly directs the agent to "Read RPI boundaries... before judgment".\n
- Capability inventory: Manifest building, digest calculation, and filesystem access are central to the skill's function.\n
- Sanitization: Employs SHA-256 content addressing and a specialized JSON parser to prevent key-collision confusion.\n- [DYNAMIC_EXECUTION]: Test infrastructure relies on dynamic loading of skill components.\n
- Evidence: tests/check_contract_corpus.py, tests/test_evidence_cli.py, and tests/test_validate.py utilize importlib.util.spec_from_file_location to load the reference Python implementation for validation and parity testing.\n- [COMMAND_EXECUTION]: The skill verifies functionality by executing platform binaries and reference scripts.\n
- Evidence: tests/test_evidence_cli.py and tests/test_validate.py use subprocess.run to call the ao provenance binary and local Python scripts during functional testing.
Audit Metadata