simplify
Pass
Audited by Gen Agent Trust Hub on Aug 21, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill operates on repository source code, tests, and documentation, which serves as a potential vector for indirect prompt injection.
- Ingestion points: Files are ingested from the target repository during the measurement, inventory, and implementation phases (SKILL.md, references/method.md).
- Boundary markers: The instructions mandate a detailed inventory of public inputs, security boundaries, and regression invariants before any code is deleted (SKILL.md Step 2).
- Capability inventory: The agent utilizes
git,repo-write-preflight, and user-defined build/test/verification tools to modify and validate the codebase. - Sanitization: Security and functional integrity are managed through a mandatory multi-step 'Proof depth' process involving deterministic gates, Test Strength (mutation testing), and Resilience Audits to detect regression in safety or operational behavior.
- [COMMAND_EXECUTION]: The skill requires executing repository-specific build, test, and linting tools to verify code changes. The risk is managed by instructions to remain read-only unless implementation is authorized and to use existing credential setups without imposing new ones.
Audit Metadata