art-asset-registry
Pass
Audited by Gen Agent Trust Hub on Nov 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes
subprocess.runto executegitcommands and internal Python scripts for file tracking and self-testing. These executions are performed securely without the use of a shell. Input arguments, such as file paths, are normalized and validated before being passed to the subprocess to prevent command injection. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from project manifests and producer JSON reports. While it lacks explicit natural language boundary markers, the risk is mitigated by the skill's limited capability inventory (local file hashing, Git status checks). All ingested paths undergo strict sanitization and normalization in
registry_ledger.py, ensuring the agent cannot be tricked into accessing files outside the project root. - [DATA_EXPOSURE]: The skill follows a strict data handling policy to avoid secret exposure. The ledger validation logic explicitly rejects the storage of raw prompt text and forbids URLs in the metadata parameters. This prevents the accidental logging of sensitive environment variables or credentials.
- [SAFE]: All operations are performed locally within the project directory. The skill does not perform any network requests or remote code execution. Path traversal is prevented by a thorough validation process that checks both absolute and real paths against the project root.
Audit Metadata