meta-prompt-engineering
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill acts as a documentation resource and template guide for meta-prompt engineering. It defines best practices for establishing AI roles, decomposing complex tasks, and defining explicit constraints to ensure high-quality and consistent outputs.
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied prompts as part of its core transformation task. While this constitutes ingestion of untrusted data, the instructions provide a robust framework for wrapping that data in structured templates and specific safety guardrails (addressing hallucination, bias, and jailbreaking). The agent's associated capability is limited to local markdown file creation, representing a low-risk profile. 1. Ingestion points: User-provided baseline prompts. 2. Boundary markers: Structured markdown templates and task-based steps. 3. Capability inventory: Writing outputs to local markdown files (meta-prompt-engineering.md). 4. Sanitization: Explicit guardrails and quality checks are recommended to the agent to filter problematic input content during the transformation process.
Audit Metadata