fix(examples): parse the canonical EEP gate response in the langgraph agent

[ucekmez]

Summary

Audit finding (agent-ergonomics vertical): the flagship LangGraph example —
the one place an agent-builder copies from — did not parse the canonical
EEP gate response it advertises. handle_gate_challenge() read a flat
{gate_type, amount, currency} body and split credential/agreement handling
onto 403 by an error string. None of those fields exist in the real
response, so the example would never match a conformant publisher.

Canonical shape

Both 402 access_restricted and 403 access_forbidden
(schemas/v0.1/gate.402-response.json / gate.403-response.json) carry:

{ "error": "access_restricted", "resource": "...", "current_tier": "...",
  "required_tier": "...",
  "unmet_requirements": [ { "type": "payment", "resolution_hint": "Pay $0.10 …", ... } ] }

Every gate type (payment, credential, agreement, identity, …) is an entry in
unmet_requirements, each with a machine-readable resolution_hint.

Change

• Rewrite the parser to iterate unmet_requirements and build one proof per
  requirement, routed on each requirement's type, using the canonical proof
  shapes from @eep-dev/gates (PaymentProof.token, CredentialProof.format,
  agreement signature over document_hash, …). The resolution_hint is what
  lets the agent decide without an LLM — the whole point of the vertical.
• test_agent.py (5 tests, all passing): payment, multi-requirement,
  403 access_forbidden, non-gate error, and unsatisfiable custom x-*.
• README + requirements made honest: the example inlines the wire contracts for
  readability and points to eep-gates/eep-signer/eep-validator for
  production, instead of claiming to import packages it never used.

Part of the EEP vertical-audit follow-up (Wave 1). No wire/schema change.

🤖 Generated with Claude Code