round-table
Pass
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests 'soul' data from external persona files (resonance files, character sheets) and interpolates this content into prompts for sub-agents to simulate a persona-driven debate.
- Ingestion points: Reads persona data from user-specified paths or specific workspace locations like
ψ/memory/resonance/(SKILL.md). - Boundary markers: Prompt templates in the
prompts/directory use clear section headers and instructions to separate the persona 'soul' from the debate context. - Capability inventory: Executes recursive LLM calls for sub-agents, writes deliberative minutes and state files to the local workspace, and runs a local TypeScript script for vote counting (SKILL.md, scripts/table.ts).
- Sanitization: Responses from sub-agents are parsed as JSON and validated by the
table.tsscript to ensure they adhere to defined confidence scales and voting options.
Audit Metadata