master-debate
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied topics and integrates them into prompts for automated subagents. While this creates a potential surface for indirect prompt injection, the orchestrator implements robust boundary markers and hard constraints to minimize risk. Specifically, it mandates a safety clause that overrides all other rules and requires subagents to use strict formatting for citations.
- Ingestion points: The
topicparameter inSKILL.mdused to define the debate subject. - Boundary markers: The prompt template uses distinct headers such as
【本轮你的任务】and【弹药库】to separate instructional context from user-provided data. - Capability inventory: Invokes the platform's
Task toolto delegate logic to subagents. - Sanitization: Instructions enforce verbatim quoting from internal sources and specific citation formats (e.g.,
《经名》【标识符】). - [EXTERNAL_DOWNLOADS]: The
LICENSE.mdfile references the author's official GitHub repository (xr843/Master-skill) for ethics policies and detailed licensing information. These are vendor-owned resources used exclusively for documentation and compliance purposes.
Audit Metadata