arrange-act-assert
Pass
Audited by Gen Agent Trust Hub on Aug 31, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute a local Python script (
scripts/inspect-tests.py) to analyze test files within the repository. The script performs static analysis and does not require elevated privileges or network access. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data in the form of test files from the local repository.
- Ingestion points:
scripts/inspect-tests.pyreads content from files matching common test suffixes (e.g., .py, .js, .java) within the repository root. - Boundary markers: The script output is provided to the agent without explicit boundary markers or sanitization to prevent potential instructions embedded in the tests from influencing the agent's review process.
- Capability inventory: The skill uses python3 to execute the inspection script and instructs the agent to review and refactor code based on the findings.
- Sanitization: The script performs regex-based scanning and does not sanitize the source code before presenting analysis to the agent.
- [METADATA_POISONING]: The skill's metadata contains anomalous entries, including future dates (2026) and a reference to a non-existent AI model version (gpt-5.6-sol). While not directly malicious, this information is inaccurate.
Audit Metadata