skills/vllnt/skills/improve-codebase/Gen Agent Trust Hub

improve-codebase

Pass

Audited by Gen Agent Trust Hub on Nov 5, 2026

Risk Level: SAFENO_CODE
Full Analysis
  • [SAFE]: The skill establishes a comprehensive engineering workflow that prioritizes behavioral preservation and requires explicit authority for sensitive operations such as infrastructure changes or production profiling.
  • [INDIRECT_PROMPT_INJECTION]: While the skill ingests untrusted code and has file-writing capabilities (attack surface), it incorporates mitigation measures including a multi-step Quality Validation protocol, baseline verification, and requirements for independent review. Evidence: 1. Ingestion points: Source code, data paths, and existing tests in the target repository. 2. Boundary markers: Quality Validation protocol, baseline definition, and isolated measurement environments. 3. Capability inventory: File system modification, test execution, and shell command execution. 4. Sanitization: Explicit instruction to sanitize sensitive evidence in the runtime performance reference.
  • [SAFE]: The documentation includes security-conscious guidance, such as avoiding the use of production credentials or real financial transactions during testing.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 5, 2026, 08:18 PM