iso-42001-ai-governance
Warn
Audited by Gen Agent Trust Hub on Jun 16, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill includes a 'GLOBAL PROTOCOLS' section that uses mandatory language ('MANDATORY', 'Iron Law', 'No cognitive labor occurs outside of a defined mode') to override the agent's default reasoning process and operational modes (BUILD, INCIDENT, EXPERIMENT). It mandates specific tool usage sequences such as 'sequentialthinking' and 'context7' MCP loops before any task execution.
- [COMMAND_EXECUTION]: The instructions mandate that all terminal actions be executed via an 'ExecutionProxy' with a specific prefix (e.g., 'rtk npm test'), which encourages the agent to perform shell operations as part of its development and audit lifecycle.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to evaluate untrusted data such as system descriptions and technical documentation, creating a vulnerability surface.
- Ingestion points: Technical documentation, model cards, and system descriptions provided as inputs to the 'ai_system_description' and 'existing_documentation' fields.
- Boundary markers: No explicit delimiters or instructions are used to separate user-provided documentation from the agent's core instructions.
- Capability inventory: Access to terminal execution (via rtk proxy), persistent memory storage (MemoryStore/Obsidian), and context verification tools.
- Sanitization: The skill lacks instructions to sanitize or validate the content of technical documentation before processing, allowing potential malicious instructions embedded in documents to influence the agent's actions during the audit.
Audit Metadata