code-reviewer

Pass

Audited by Gen Agent Trust Hub on Jul 25, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection because it is designed to ingest and analyze untrusted code provided by users.
  • Ingestion points: The 'input' and 'targets' parameters in REVIEW and SCAN modes (identified in skill.md).
  • Boundary markers: The instructions do not define clear delimiters or use 'ignore' constraints to distinguish between the skill's instructions and the untrusted content being reviewed.
  • Capability inventory: The agent is instructed to perform local filesystem writes (logging to .skill-audit/) and modify its own triggers, which could be exploited if an injection succeeds.
  • Sanitization: No content sanitization or instruction-filtering is specified before the code is processed.
  • [SAFE]: The 'Use-to-Evolve' (UTE) framework is a documented feature for performance tracking and autonomous trigger optimization. The associated filesystem operations (appending to .skill-audit/usage.jsonl) and micro-patching of keywords are used for maintenance purposes and do not involve remote exfiltration or unauthorized system modifications.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 25, 2026, 09:44 AM
Security Audit — agent-trust-hub — code-reviewer