ai-engineering-toolkit
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The 'Agent Safety Guard' workflow (Skill 4) is designed to process and analyze untrusted external data, specifically agent configurations and system prompts, to perform security audits.
- Ingestion points: Processes user-provided agent instructions, system prompts, and RAG document samples within the 'Agent Safety Guard' workflow.
- Boundary markers: Implements a 'Mandatory confirmation gate' that requires the user to specify the target, confirm authorization, and approve specific commands before the AI generates or evaluates adversarial content.
- Capability inventory: The skill is instructed to generate adversarial test prompts (prompt injection, tool abuse, information extraction). It functions as an advisory tool and does not demonstrate capabilities to write to the file system or initiate network requests directly in the provided instructions.
- Sanitization: No explicit sanitization or filtering of the ingested agent data is described beyond the advisory analysis process.
- [EXTERNAL_DOWNLOADS]: The installation section provides instructions to clone the skill's source code from the author's GitHub repository at 'github.com/viliawang-pm/ai-engineering-toolkit'.
Audit Metadata