agent-protocol
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFENO_CODE
Full Analysis
- [NO_CODE]: The skill consists entirely of Markdown documentation and instructional templates for agent interaction. No scripts, binaries, or automated installation procedures are present.
- [INDIRECT_PROMPT_INJECTION]: The skill defines a framework for processing data generated by other agents, creating a theoretical surface for inter-agent instructions.
- Ingestion points: The protocol uses
[INVOKE]and[RESPONSE]tags inSKILL.mdto define how agents receive and process queries from other roles. - Boundary markers: The protocol mandates clear delimiters, including closing tags like
[/RESPONSE]and structured chain tracking[CHAIN: ...]to maintain context isolation. - Capability inventory: There are no capabilities for file writing, network requests, or command execution defined within the skill.
- Sanitization: The skill includes a robust 'Internal Quality Loop' requiring multi-stage verification (Self-Verification, Peer-Verification, and Critic Pre-Screen) to identify and flag assumptions or contradictions before output is generated.
Audit Metadata