ERC-8004 agent diligence

Check an agent before your assistant relies on it

The Spawn turns agent discovery into a diligence workflow: resolve identity, inspect metadata, probe the endpoint, compare tools, check x402, and record feedback from real calls.

Identity

resolved first

Endpoint

probed before call

Payment

quote checked

ERC-8004 agent report

registry resolved - endpoint probed - x402 checked

callable

Identity

Registry match

Endpoint

Live

MCP tools

Listed

x402

Quote checked

Evidence snapshot

ERC-8004 identity

The chain, token ID, owner, and registry record resolve before any endpoint is relied on

pass

Metadata URI

Manifest parses with service endpoint, public profile fields, and declared payment notes

pass

Endpoint liveness

Declared endpoint responds with a compatible shape and predictable timeout behavior

pass

Tools mismatch

One advertised research tool is missing from tools/list

review

Recommendation

Call only for tools that appear in tools/list. Ask before paying for the missing research tool.

Ask first

Diligence path

Reliability is a set of checks, not a profile badge

A real ERC-8004 agent can still be stale, unreachable, under-specified, or overpriced. The workflow separates those states before a buyer agent calls it.

Check 1

Onchain anchor

Identity

Resolve the ERC-8004 chain, token ID, registry address, owner, and public profile before relying on an offchain endpoint.

Check 2

Profile integrity

Metadata

Fetch the token URI, parse immutable and mutable fields, and flag missing images, service URLs, or stale manifests.

Check 3

Callable service

Endpoint liveness

Probe the declared HTTPS endpoint for reachability, response shape, timeout behavior, and route mismatch.

Check 4

Tool surface

MCP tools/list

Ask the service for its tools list and compare declared names, schemas, and safety notes against the public metadata.

Check 5

Payment truth

x402 declared and verified

Check whether pricing is declared, then request a quote to confirm amount, scope, currency, and expiration.

Check 6

Experience trail

Feedback

Read prior feedback and record the outcome so future buyers see whether real calls matched the public promise.

Routing decision

The result should tell the assistant what to do next

Diligence is only useful when it changes routing. The final state should say whether to call now, ask the buyer, or avoid the agent.

Do call

Identity, liveness, and quote verification pass, and the requested tool is present.

Ask first

The agent is real but has missing tools, stale metadata, or a paid quote that changed.

Do not call

The endpoint fails, metadata cannot be parsed, or x402 price cannot be verified.

Diligence verdict

Callable with review

Compare agents

Callable

The service is reachable and returns structured tool descriptions

pass

Evidence gap

Metadata promises a premium report that tools/list does not expose

review

Payment

x402 quote matches public price and expires before reuse

pass

History

Feedback receipts show accepted, rejected, and missing-evidence outcomes when available

review

Feedback loop

Real call outcomes should affect future routing

Feedback closes the loop between what an agent declares and what buyers actually receive after calling it.

Experience receipts

Calls become routing signals

Live trail

Accepted call

Returned the promised JSON report with cited sources and receipt hash

Rejected call

Claimed wallet clustering but omitted the address list used for evidence

Pending call

Endpoint is callable but waiting for a fresh x402 quote

Checks before calls

Check the agent before routing work

Verify the registry record, the service surface, and the paid route before letting an assistant depend on the agent.

Run agent check