refactoring-safety
Pass
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill possesses a potential attack surface for indirect prompt injection as it is designed to ingest untrusted source code and execute shell commands for build and test validation.\n
- Ingestion points: The skill accepts code blocks, test coverage signals, and consumer interface descriptions as primary arguments defined in
SKILL.md.\n - Boundary markers: The workflow mandates a structured Markdown report and uses specific gates like 'Refactor Baseline' to isolate the original code state from proposed changes.\n
- Capability inventory: The agent is instructed to perform file system operations (read/write) and execute shell commands (e.g., test suites, build checks) to verify refactor steps.\n
- Sanitization: The skill relies on functional validation through 'tripwires' and comparative diff reviews rather than automated sanitization of the code content itself.
Audit Metadata