simplify
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill analyzes external code which could contain adversarial instructions embedded in comments or logic intended to trick the agent into performing unsafe refactors.
- Ingestion points: Target files are read and evaluated against a rubric in Phase 1 (Inspect & Clean-Pass Check).
- Boundary markers: The skill instructions demand a 'Changeset' summary and a 'Pre-Mutation Gate' that halts execution until human authorization is received.
- Capability inventory: The skill has the capability to write mutations to disk and execute shell commands (tests/linters) during the verification phase.
- Sanitization: No automated sanitization is described; safety depends on the agent's adherence to invariants and the user's review of the proposed changes.
- [COMMAND_EXECUTION]: The skill executes local developer tools to verify the integrity of refactored code.
- Evidence: Phase 3 (Mutate & Verify) explicitly directs the agent to 'Run existing tests, linters, or syntax checks to verify behavioral invariance.'
Audit Metadata