Skip to content

Agent-to-agent (A2A) communication scanning #58

@debu-sinha

Description

@debu-sinha

Problem or use case

Multi-agent architectures are becoming standard. Cisco DefenseClaw introduced A2A scanning that validates inter-agent communication channels are authenticated and that no unauthorized agent can inject instructions into a trusted agent's workflow.

As agents use MCP, A2A protocol, and custom tool-based delegation, the communication surface between agents becomes an attack vector for instruction injection and privilege escalation.

Proposed solution

Add an a2a scanner module that checks:

  1. Agent delegation configs for authentication requirements
  2. Inter-agent communication channels for encryption (TLS)
  3. Tool schemas that accept arbitrary text from other agents (injection surface)
  4. Permission boundaries between agents (can agent B access agent A's tools?)
  5. Circular delegation patterns (agent loops)

This is a new scanner module alongside installation, skill, mcp, and credential.

Area

New scanner module

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or requestsecuritySecurity hardening

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions