Agent Commerce Rails

If agents will route the majority of commerce, which station breaks first when the rails don't hold?

The Thesis​

Agents are about to move the majority of commerce. They need five things in order: a signed identity, a discoverable service manifest, an executable intent, a settled payment, a verified receipt. Without any one station, the chain breaks.

The AP2 gap is real. It is not the first gap. Identity is. Fix identity first, then discovery, then intent, then settlement, then receipts. Reputation and disputes come after there's history to reason over.

Dreamineering's current state — phase=1-discoverable | stripe=wired | ucp=partial | ap2=gap — skipped the first two stations. This PRD locks the order.

The Seven Stations​

Identity → Discovery → Intent → Settlement → Receipts → Reputation → Disputes.

Five in scope (COMRC-001 through COMRC-005). Two deferred to Phase 2 (reputation + disputes) — those depend on history that does not yet exist.

Problem​

Situation: WM-NAV @COMMERCE records phase=1-discoverable | stripe=wired | ucp=partial | ap2=gap | next=first-external-mcp-tool-call. Dreamineering wants to sell tools to external agents and buy tools from external agents. Stripe is wired but no external agent has ever signed an intent, called a tool, and settled through Dreamineering's rails.

Intention: Agents transact across the rails in the correct order — identity first, then discovery, then intent, then settlement, then receipts — with verifiable proof at every hop. Dreamineering pays its own first MCP tool via x402 before selling the rails to anyone else.

Obstacle: Engineering was about to build AP2 (settlement) before identity and manifest exist. Settlement without identity has no party to credit. Manifest without identity has no origin trust. Receipts without mandates have nothing to anchor. The order matters more than the velocity.

Hardest thing: The seven stations are ordered — identity first — but the business pressure is to ship revenue, which reads like "ship settlement." Holding the order under pressure is the hard part. The dogfood rule (Dreamineering pays its own MCP tool) is the corrective loop on the pressure.

Why Now​

Relationship to Other PRDs​

Priority Score​

PRIORITY = Pain × Demand × Edge × Trend × Conversion

Scorecard​

Priority Score: 400 — Pain 4 × Demand 2 × Edge 5 × Trend 5 × Conversion 2

Demand-Side Jobs​

Job 1: Merchant with an MCP Tool to Sell​

Situation: A developer built an MCP tool that other agents would pay to call. They do not want to build identity, discovery, or payment rails from scratch.

Job 2: Agent Operator (Consumer or B2B)​

Situation: An agent operator wants their agent to buy data, trigger a service, or pay for a tool call — with a spend cap and a verifiable receipt.

Job 3: Integrator / Reseller​

Situation: A systems integrator serving a vertical (construction, solar, DePIN) wants a reference stack they can install for their own customers.

Build Sequence — The Seven Stations​

STATION 1: IDENTITY (COMRC-001)
  ──────────────────────────────────────
  did:web:dreamineering.com issued.
  VC schema for merchants and agents defined.
  Dreamineering signs its own origin VC.
        │
        │ proves: "this call came from who they say they are"
        ▼
STATION 2: DISCOVERY (COMRC-002)
  ──────────────────────────────────────
  /.well-known/agent.json hosted on dreamineering.com.
  Tool registry entry for Dreamineering's first sellable MCP tool.
  Manifest signed with identity from Station 1.
        │
        │ proves: "this service exists and this is what it offers"
        ▼
STATION 3: INTENT — first external MCP tool call (COMRC-003)
  ──────────────────────────────────────
  Dreamineering's agent discovers a Dreamineering tool via manifest.
  Makes a signed call. Tool responds 402 Payment Required (x402).
  Payment settles on Base in sub-second.
  (Dogfood: Dreamineering pays itself first.)
        │
        │ proves: "an agent can execute an intent and pay for it"
        ▼
STATION 4: SETTLEMENT — AP2 mandate handler (COMRC-004)
  ──────────────────────────────────────
  Agent signs an AP2 mandate (intent + spend cap + merchant).
  Handler validates, routes to x402 (or Stripe for fiat),
  returns settlement proof to the calling agent.
        │
        │ proves: "intent → mandate → money movement is one atomic flow"
        ▼
STATION 5: RECEIPTS — IntentTrace v1.2 (COMRC-005)
  ──────────────────────────────────────
  Schema upgrade: add intent_ref (hash), action_hash, outcome, verified.
  Every settled mandate emits a receipt.
  Receipts anchor to identity from Station 1.
  Pre-chain anchor now, chain anchor when justified.
        │
        │ proves: "every transaction leaves verifiable history"
        ▼
[PHASE 2 — DEFERRED]

STATION 6: REPUTATION (COMRC-006, deferred)
  ──────────────────────────────────────
  Read-only oracle over IntentTrace history.
  Score counterparties by completion rate, dispute rate, freshness.
  Cannot build until Station 5 has produced real history.
        │
        ▼
STATION 7: DISPUTES (COMRC-007, deferred)
  ──────────────────────────────────────
  Held settlement + release state machine.
  Mediator role defined.
  Cannot build until settlements exist to dispute.

What Exists vs What's Missing​

Features (Frozen Scope: 5)​

Station 1: Identity — COMRC-001​

• Publish did:web:dreamineering.com document with verification methods
• Define VC schema for MerchantCredential and AgentCredential
• Dreamineering signs its own MerchantCredential as first issuer
• Verification helper: any caller can resolve did:web → signed VC → origin proof
• First agent on Agent Platform receives an AgentCredential signed by Dreamineering

Acceptance: A third party resolves did:web:dreamineering.com, reads a signed MerchantCredential, and can verify the signature without Dreamineering's help.

Station 2: Discovery — COMRC-002​

• /.well-known/agent.json hosted on dreamineering.com with manifest schema
• First sellable MCP tool registered in the manifest with pricing hint
• Manifest is signed with the identity from COMRC-001
• Public tool registry (read-only) lists the tool with resolvable endpoint
• Manifest refreshes on tool publish (no manual re-deploy)

Acceptance: An external agent fetches /.well-known/agent.json, verifies the signature against the did:web document, and reads a callable tool endpoint.

Station 3: Intent — COMRC-003 (First External MCP Tool Call)​

• Dreamineering hosts one billable MCP tool with a 402 Payment Required response
• Dreamineering's own agent discovers the tool via manifest and calls it
• Call triggers x402 on Base, settles in sub-second
• Tool returns successful response after settlement confirms
• End-to-end dogfood proof: Dreamineering pays Dreamineering for a real tool call

Acceptance: WM-NAV @COMMERCE state moves from phase=1-discoverable to phase=2-transacting. One real settled call in the audit log.

Station 4: Settlement — COMRC-004 (AP2 Mandate Handler)​

• AP2 mandate schema accepted on a Dreamineering endpoint
• Signed intent validated against agent identity from COMRC-001
• Spend cap enforced at the mandate layer, not the tool layer
• Handler routes mandates to x402 (crypto) or Stripe (fiat) based on currency
• Settlement proof returned to the calling agent in the response

Acceptance: An external agent (not Dreamineering's own) signs an AP2 mandate, calls a Dreamineering tool, and receives a settlement proof. Spend cap is enforced — a second call over cap is rejected with a clear reason.

Station 5: Receipts — COMRC-005 (IntentTrace v1.2)​

• IntentTrace schema upgraded to v1.2 with intent_ref hash field
• Every settled mandate emits a receipt to .invisible/context/receipts/
• Receipt references the AP2 mandate hash + settlement proof + identity VC
• Read API returns a transaction's full trace from intent to settlement
• Pre-chain anchor (signed file + hash). Chain anchor deferred until volume justifies it.

Acceptance: Given any completed transaction, a commissioner can replay it end-to-end: identity → manifest → intent → mandate → settlement → receipt, with cryptographic proof at each hop.

Phase 2 (Out of Scope — Justification for the Cut)​

The 5-feature cap is respected. Phase 2 will be a new PRD when Phase 1 produces signal.

Kill Signals — Per Station​

PRD-level kill signal: By 2026-07-15, if fewer than 1 external agent has completed a settled mandate through Dreamineering's rails, deprioritise the entire PRD and loop back to merchant demand validation. Dogfood passing is necessary but not sufficient.

The VVFL Loop​

This PRD is a virtuous loop. Every completed intent improves the template for the next one.

IDENTITY issued
   │
   ▼
MANIFEST signed → DISCOVERED by agent
   │
   ▼
INTENT signed → MANDATE validated → SETTLEMENT executed
   │
   ▼
RECEIPT anchored → HISTORY grows → REPUTATION scorable
   │
   ▼
TEMPLATE improves (VVFL Legacy Rule)

The feedback: every receipt is a data point for reputation. Every reputation score tightens the next mandate. Every tighter mandate makes the next settlement cheaper. The loop compounds.

The setpoint: verifiable intent at every hop. The gauge: IntentTrace receipts. The controller: the mandate handler.

Commissioning​

Protocol Coverage Link​

Every COMRC feature must appear in the Protocol Coverage Commerce Auth Chain gauge. The gauge currently reports 3 broken links on that chain — this PRD is what fills them.

Players​

The three unrecruited players are the Demand score 2. They are also the kill signal at 2026-07-15.

Issues​

Context​

• Business Factory — Factory activation is the intent behind this PRD
• IntentTrace Standard — The receipt schema this upgrades
• Blockchain Verifiable Truth — The proof layer the rails inherit
• Protocol Coverage — The gauge this PRD is measured against
• Trust Commerce — The downstream product on top of these rails
• Identity & Access — The peer human identity plane

Questions​

What is the single most important thing that must be true for Agent Commerce Rails to succeed — and how would you know if it weren't?

• If identity is Station 1, who is the first external agent that holds a Dreamineering-signed credential — and what makes them trust it?
• The dogfood proof is "Dreamineering pays itself." What is the first external transaction that proves the rails carry weight others trust?
• If Demand scores 2 and Conversion scores 2, which of the three audiences (merchant / operator / integrator) moves first — and what conversation produces that signal?
• At station 4, AP2 is settlement — but the real test is a mandate with a spend cap that rejects the second call. What does "graceful rejection" look like to the caller's agent, and who defines it?
• Phase 2 defers reputation and disputes until history exists. How much history is enough history — 10 transactions, 100, 1000? What measurement graduates Phase 2 from speculation to spec?