criticism-self-criticism
Pass
Audited by Gen Agent Trust Hub on Mar 27, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is entirely instructional and does not include executable code or shell commands.
- [PROMPT_INJECTION]: Instructions focus on internal quality control and logical review processes; no attempts to bypass safety filters or extract system prompts were detected.
- [DATA_EXFILTRATION]: No network tools or sensitive file paths are used in the skill.
- [REMOTE_CODE_EXECUTION]: The skill does not download or execute external scripts or packages.
- [INDIRECT_PROMPT_INJECTION]: The skill processes user feedback, creating a surface for instruction injection. 1. Ingestion points: User-provided feedback as defined in SKILL.md. 2. Boundary markers: Uses a structured 'Work Review Report' template to isolate review findings. 3. Capability inventory: No tools or dangerous capabilities are invoked by the skill itself. 4. Sanitization: Includes specific logic to judge if criticism is based on facts and investigation-first principles.
Audit Metadata