constitutional-ai
Pass
Audited by Gen Agent Trust Hub on Sep 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documents workflows for processing and critiquing potentially harmful user-provided prompts.
- Ingestion points: The 'prompts' variable in the Python snippets within SKILL.md.
- Boundary markers: No explicit delimiters or boundary markers are used in the demonstration code.
- Capability inventory: The skill uses the transformers library for text generation and the trl library for updating model weights via SFTTrainer and PPOTrainer.
- Sanitization: No input sanitization is demonstrated, as the workflows are designed to analyze and correct harmful content. This ingestion surface is a standard requirement for the training methods described.
- [SAFE]: The skill uses reputable dependencies (transformers, torch, trl) and references official research from Anthropic and Arxiv. No malicious commands, exfiltration, or obfuscation were detected.
Audit Metadata