Cryptographic trust verification for multi-agent Dify workflows using verification (Ed25519) identities.
- Peer Verification: Verify other agents' identities and capabilities before trusting their data
- Workflow Step Authorization: Ensure agents have required permissions for each step
- Trust Score Tracking: Dynamic trust scores based on interaction history
- Audit Logging: Complete audit trail of all trust decisions
- Clone this repository or download the plugin folder
- Package the plugin:
cd dify-plugin zip -r agentmesh-plugin.zip .
- Upload to Dify via Settings → Plugins → Install from Package
When configuring the plugin in Dify, you can set:
| Parameter | Description | Default |
|---|---|---|
| Minimum Trust Score | Score (0-1) required for verification | 0.5 |
| Cache TTL | Seconds to cache verification results | 900 |
| Identity Name | Name for this agent's identity | dify-agent |
| Capabilities | Comma-separated list of capabilities | (empty) |
Verify another agent's identity before trusting their data or delegating tasks.
Parameters:
peer_did(required): The DID of the peer agent (e.g.,did:verification:abc123)peer_public_key(required): Base64-encoded Ed25519 public keyrequired_capabilities: Capabilities the peer must havepeer_capabilities: Capabilities the peer claims to have
Check if the current agent has permission to execute a workflow step.
Parameters:
workflow_id(required): Workflow identifierstep_id(required): Step identifier within workflowstep_type(required): Type of step (llm, tool, code, etc.)required_capability: Specific capability to check
Get this agent's cryptographic identity to share with other agents.
Parameters:
include_capabilities: Whether to include capabilities list
Record successful or failed interactions to update trust scores.
Parameters:
peer_did(required): DID of the peer agentsuccess(required): Whether the interaction succeededseverity: For failures, how much to decrease trust (0-1)
1. Agent A calls "Get Agent Identity" to get its DID and public key
2. Agent A shares identity with Agent B
3. Agent B calls "Verify Peer Agent" with A's DID and public key
4. If verified, Agent B proceeds with collaboration
5. After interaction, Agent B calls "Record Interaction" to update trust score
MIT License - See AgentMesh for details.