backend-refactor
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill executes internal utility scripts such as 'checkpoint.js' and 'security-scan.js' located in a shared directory to manage the refactoring lifecycle and detect security regressions.
- [COMMAND_EXECUTION]: The workflow involves extensive use of shell commands, including Git operations for state management and the execution of project-specific test runners such as Jest, Pytest, or Go test to verify behavior preservation.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted project files and executes code through test suites, establishing a surface for indirect prompt injection.
- Ingestion points: Backend source code, test files, and project configuration metadata are read from the local development environment.
- Boundary markers: There are no explicit prompt delimiters used to isolate code content from the agent's instructional logic.
- Capability inventory: The skill is capable of modifying the codebase, running shell commands, and executing the project's test suite which may run arbitrary code.
- Sanitization: No sanitization is performed on the ingested code as the tool is designed to move and restructure existing logic without altering its functionality.
Audit Metadata