constitutional-ai

Pass

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill demonstrates workflows that ingest and process external prompts. This creates a potential surface for indirect prompt injection, which the methodology itself aims to address through automated critiques.
  • Ingestion points: Prompt variables and user-provided inputs within the Python code snippets (SKILL.md).
  • Boundary markers: The templates do not define explicit delimiters or instructions to ignore embedded commands within the user prompts.
  • Capability inventory: The skill uses model inference via transformers pipelines and fine-tuning classes (SFTTrainer, RewardTrainer, PPOTrainer) to process and learn from data (SKILL.md).
  • Sanitization: The "Self-critique with constitution" step is designed to identify and filter harmful content programmatically as part of the core workflow.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 9, 2026, 07:06 PM
Security Audit — agent-trust-hub — constitutional-ai