loop-review-skill-until-fixed-point

Pass

Audited by Gen Agent Trust Hub on Feb 23, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes external file content via the target argument, creating a surface for indirect prompt injection.
  • Ingestion points: The target file path provided in the skill arguments (parsed in the Initialize phase).
  • Boundary markers: Absent; while the orchestrator enforces context isolation between iterations, it does not define specific delimiters for the target file content itself.
  • Capability inventory: The skill triggers the /review-skill command repeatedly.
  • Sanitization: No explicit sanitization or instruction filtering is performed on the target file content before processing.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 23, 2026, 09:43 AM