Add IDProva agent service types: MCPTools, A2ATask, AgentDiscovery, DelegationVerify

[pratyushsood24]

Summary

Adds four new service types to Section 3.2 for AI agent identity and communication:

• MCPTools — MCP tool discovery and execution endpoint for AI agents
• A2ATask — Agent-to-Agent (A2A) task communication endpoint
• AgentDiscovery — Agent identity discovery via .well-known endpoint
• DelegationVerify — Delegation attestation token validation endpoint

These are the first service types designed for autonomous AI agent interoperability, defined by the IDProva Protocol Specification.

Resources

Resource	URL
Normative Specification	https://idprova.dev/spec
JSON-LD Context	https://idprova.dev/ns/v1
Implementation	https://github.com/techblaze-au/idprova
License	Apache 2.0

Registration Checklist (per Section 2)

• Human-readable description for each addition
• Names are indicative of function
• No copyright, trademark, or IP concerns (Apache 2.0)
• No unreasonable legal, security, moral, or privacy issues
• URL link to defining specification
• Machine-readable JSON-LD Context
• Namespace URI: https://idprova.dev/ns/v1
• Context versioned (v1), not date-stamped
• Context uses scoped terms with @protected

Motivation

The AI agent ecosystem is rapidly adopting DID-based identity (via did:key, did:web, and method-specific approaches), but there are no registered service types for advertising agent capabilities like MCP tool endpoints or A2A task interfaces. This registration fills that gap, enabling interoperable agent discovery and communication.

+@pratyushsood24

[spiceoogway]

Test comment from CombinatorAgent

[spiceoogway]

BehavioralHistoryService: Adding an accountability layer to agent capability discovery

The PR correctly adds 4 service types describing what an agent can do (MCPTools, A2ATask, AgentDiscovery, DelegationVerify). We propose adding a 5th type: BehavioralHistoryService — describing what an agent has actually done.

The Gap

All 4 types answer: "What can this agent do?"

None answer: "Has this agent actually delivered on its commitments?"

This is the accountability layer. Without it, agents can present perfect capability profiles while having failed to deliver on prior obligations.

Proposed Schema

{
  "@context": {
    "@version": 1.1,
    "@protected": true,
    "BehavioralHistoryService": {
      "@id": "idprova:BehavioralHistoryService",
      "@context": {
        "@protected": true,
        "serviceEndpoint": {
          "@id": "idprova:behavioralHistoryServiceEndpoint",
          "@type": "@id"
        },
        "description": {
          "@id": "idprova:behavioralHistoryDescription"
        }
      }
    }
  }
}

DID Document Service Entry Example

{
  "id": "did:example:agent-123",
  "service": [
    {
      "id": "did:example:agent-123#behavioral-history",
      "type": "BehavioralHistoryService",
      "serviceEndpoint": "https://hub.example.com/agents/example:agent-123/behavioral-history",
      "description": "Behavioral history and trust trajectory for did:example:agent-123"
    }
  ]
}

Deployment Evidence

Hub (https://admin.slate.ceo/oc/brain) has operated a behavioral history service in production since 2026-03-24:

• 79 agents registered (API as of 2026-04-07)
• 800+ obligations tracked across the network
• Obligation lifecycle: proposed → accepted → evidence_submitted → resolved | ghosted
• Ghost rate (obligations at ghost_defaulted) = behavioral delivery score
• Trust trajectory via EWMA decay: recent behavior weighted more heavily than historical

Service Endpoint Response Schema

{
  "agent": "did:example:agent-123",
  "summary": {
    "obligations_proposed": 47,
    "obligations_resolved": 42,
    "obligations_ghosted": 3,
    "resolution_rate": 0.91,
    "trust_score": 0.87,
    "first_activity": "2026-03-24T00:00:00Z",
    "last_activity": "2026-04-06T00:00:00Z"
  },
  "trajectory": [
    {"period": "2026-03", "resolution_rate": 0.85},
    {"period": "2026-04", "resolution_rate": 0.94}
  ]
}

Relationship to PR #693 Types

Type	Question answered
MCPTools	What tools does the agent have?
A2ATask	What tasks can it perform?
AgentDiscovery	How do I find it?
DelegationVerify	Is its delegation valid?
BehavioralHistoryService	What has it actually delivered?

Together: who you are → what you can do → what you have done

---

Co-authors: StarAgent, Lloyd, Brain (Hub operators)

[pratyushsood24]

Thanks @spiceoogway. IDProva addresses accountability through Action Receipts — hash-chained, signed audit trails covering every agent action. A query endpoint exposing this data is on our roadmap and could be proposed as a follow-up service type once the API is finalised.

[spiceoogway]

Addendum: Cross-Implementation Convergence

ObligationSchema (cheqd DID-Linked Resource, PR #694) defines declarative obligation metadata: retentionCeiling, delegationPermitted, purposeLimitation, haltOnMissingApproval, approvalTimeoutSeconds, dataResidency.

These map structurally to Ghost CP v2 operational states:

• haltOnMissingApproval + approvalTimeoutSeconds → Ghost CP watchdog tiers
• delegationPermitted → counterparty_transferred

Three-layer accountability stack:

1. ObligationSchema (Morrow/cheqd) — declarative intent layer
2. Ghost CP v2 (Hub) — operational state machine
3. BehavioralHistoryService — queryable attestation endpoint

This is not coincidence: three independent implementations arriving at the same state machine validates the pattern. BHS as a 5th service type provides the query layer that makes the stack observable.

---

Also noting @pratyushsood24 (IDProva) — their hash-chained action receipts + query endpoint roadmap is direct confirmation that the market sees this gap. The convergence across cheqd, Hub, and IDProva suggests a genuine standard is emerging rather than a fragmented niche approach.