agent-identity-resolver

name

agent-identity-resolver

description

Resolve, verify, and audit an agent's identity using the AID (Agent Identity & Discovery) v2 protocol. Performs DNS-based discovery, signature verification, capability negotiation, and produces a structured identity report with trust signals. Use whenever a user wants to verify which agent is calling their service, look up an agent's capabilities and provenance from its handle, audit agent-to-agent traffic, asks 'who is this agent', 'is this AID valid', 'verify this agent signature', 'what can agent X do', or is building a service that needs to authenticate non-human callers. Trigger for agentic commerce, x402 micropayments, agent-to-agent (A2A) flows, and any system gating actions on caller identity.

Agent Identity Resolver

Resolve and verify agent identities under AID v2 — DNS discovery, signature verification, capability negotiation, and a structured trust report.

Workflow

1. Take an input identifier — an AID handle (agent.example.com), an AID URL (aid://agent.example.com/v2), or a raw signature + claimed identity pair to verify.
2. Perform DNS discovery:
   • Query _aid.<domain> TXT records for the agent's discovery record
   • Parse the v2 record fields: v=2, pk=<public-key>, caps=<capability-set>, well-known=<https://...>, iss=<issuer> (optional)
   • Reject records with v other than 2 or missing required fields
3. Fetch the well-known document at https://<domain>/.well-known/aid/v2.json. Verify:
   • JSON Schema validity
   • Public key in the doc matches the TXT record pk
   • iss (issuer) chain validates if present
   • Document is signed by the agent's private key (detached signature header or embedded sig field)
4. Verify any provided signature against the resolved public key. Use ed25519 by default; reject RSA-1024 and unsupported curves.
5. Negotiate capabilities: intersect the agent's declared caps with the verifier's allowed-caps list. Return the negotiated set.
6. Compute trust signals: domain age, DNSSEC status, issuer reputation if iss chain present, capability scope (narrow scopes are higher-trust by default), key rotation history if available.
7. Emit a structured identity report. Cache the resolution per (handle, key) tuple for at most the TTL of the DNS record; never longer than 1 hour.

Output Format

Produce aid-report-<handle>.json plus a human summary aid-report-<handle>.md:

{
  "handle": "agent.example.com",
  "resolved_at": "<ISO 8601>",
  "version": 2,
  "discovery": {
    "txt_record": "v=2; pk=...; caps=...; well-known=...",
    "well_known_url": "https://.../.well-known/aid/v2.json",
    "dnssec": true
  },
  "identity": {
    "public_key": "<ed25519 pk>",
    "algorithm": "ed25519",
    "signature_verified": true,
    "issuer": "<issuer handle or null>",
    "issuer_chain_valid": true
  },
  "capabilities": {
    "declared": ["..."],
    "negotiated": ["..."]
  },
  "trust": {
    "domain_age_days": 412,
    "dnssec": true,
    "key_rotation_history_known": false,
    "score": 0.78,
    "score_basis": ["dnssec=true (+0.2)", "issuer_chain_valid=true (+0.3)", "..."]
  },
  "verdict": "verified | partial | rejected",
  "warnings": ["..."]
}

Verdict Rules

• verified — all required fields present, signature verifies, capabilities negotiable, trust score ≥ 0.7
• partial — identity resolves but at least one of: trust score 0.4–0.7, DNSSEC absent, issuer chain unverified, key fresh (< 24h)
• rejected — signature fails, public-key mismatch, version != 2, well-known fetch fails, or trust score < 0.4

Quality Bar

• Never declare verified without a signature check. Resolution alone is not verification.
• Cache TTLs are honored. Stale identity decisions are a security bug.
• Reject deprecated crypto: no RSA-1024, no MD5, no SHA-1.
• Capabilities are always intersected with the verifier's allowed set. Never trust the agent's declared caps verbatim.
• Trust scoring is transparent: every contribution to the score is listed in score_basis.
• Errors are explicit: never silently downgrade verified → partial without a warning in the output.
• This skill is for verification only. It does not issue identities — that belongs to an AID registry.