Developer guide / intelligence.1 & agent-review.1

A private review.
A defined boundary.

Use an issued key bound to your approved firm pilot. Submit evidence, inspect findings, keep execution permissions in your own systems, and verify signed receipts independently.

01 / Access and discovery

Inspect the actual capability contract.

Snippet
GET /v1/intelligence/capabilities
GET /v1/intelligence/scenarios
Authorization: Bearer <issued-key>

The capability response includes the JSON input schemas. Read, submit and review operations require intelligence:read, intelligence:write and intelligence:review respectively. An ordinary Watch key does not inherit these permissions.

02 / Save a review

One request reference, one saved run.

Snippet
{
  "module": "recall",
  "title": "Supplier review",
  "sample": false,
  "data_rights_confirmed": true,
  "input": {
    "nodes": [
      {
        "id": "source-a",
        "kind": "source",
        "label": "Changed supplier statement",
        "revision": 2,
        "state": "corrected"
      },
      {
        "id": "decision-a",
        "kind": "decision",
        "label": "Purchase approval",
        "depends_on": [
          "source-a"
        ]
      }
    ]
  }
}

POST to /v1/intelligence/runs with an Idempotency-Key UUID. Reuse it only for an identical retry. A changed request returns 409. The body contains the evidence; keep it out of URLs, telemetry and logs.

Input hygiene: keep titles, labels and reviewer reasons brief and factual (under 12 words). Long multi-word prose and credential-like strings are conservatively rejected with HTTP 422 by secret screening.

03 / Review and retention

Keep conclusions and dispositions separate.

Snippet
GET /v1/intelligence/runs
GET /v1/intelligence/runs/{id}
POST /v1/intelligence/runs/{id}/followups
{ action: 'acknowledged', expected_revision: 0, evidence_ref: 'your-private-review-reference' }

Acknowledge before resolving; a resolved run can be reopened. Stale revisions return 409. Follow-ups retain a keyed fingerprint of the supporting reference, so keep the original in your own record system. Content expires after 30 days and reads return 410; minimal integrity and audit records remain.

Results describe submitted evidence only. No overlap does not prove independence. Recorded expiry does not prove enforcement. A verification plan does not mean checks have run. Synthetic failure replays must carry sample: true.

04 / Agent Action Review (agent-review.1)

Compare proposed action, mandate and evidence.

Snippet
GET  /v1/intelligence/agent-reviews/capabilities
GET  /v1/intelligence/agent-reviews/example
POST /v1/intelligence/agent-reviews/observations
POST /v1/intelligence/agent-reviews
GET  /v1/intelligence/agent-reviews
GET  /v1/intelligence/agent-reviews/{review_id}

Read-only network observations.

Supports 14 configurations across 11 families (XDC Mainnet/Apothem, Ethereum, Base, Arbitrum, OP, Polygon, Avalanche C-Chain, Celo, BNB Smart Chain, Solana Mainnet/Devnet, Stellar Mainnet/Testnet). EVM internal traces, rollup settlement proofs, token identity and Stellar path/Soroban effects remain outside scope.

Authenticated human decisions only.

POST /v1/intelligence/agent-reviews/{review_id}/decisions requires a signed-in firm web reviewer (owner, admin, or analyst) with recent MFA. API keys can submit reviews and inspect records but cannot impersonate a human reviewer.

Advisory boundary.

A submitted mandate is a claim of authority until independently verified. Recording approval never executes, signs, holds or reverses funds.

05 / Agent Evidence Receipts & issuer keys

Verify integrity, publisher trust and lifecycle separately.

HTTP endpoints · issuer keys & receipts
# Official public Ed25519 issuer keys (pin independently over HTTPS)
GET https://8thguard.com/.well-known/8thguard-receipt-keys.json

# Receipt issuance, live status & lifecycle (authenticated pilot)
POST /v1/intelligence/agent-reviews/{review_id}/receipts
GET  /v1/intelligence/agent-reviews/{review_id}/receipts/{receipt_id}
POST /v1/intelligence/agent-reviews/{review_id}/receipts/{receipt_id}/lifecycle
POST /v1/intelligence/agent-reviews/{review_id}/receipts/{receipt_id}/anchors
POST /v1/intelligence/agent-reviews/verify-envelope
Node 20+ · Zero-dependency offline Ed25519 verifier
import { readFileSync } from "node:fs";

const canonical = (v) =>
  Array.isArray(v) ? "[" + v.map(canonical).join(",") + "]"
  : v && typeof v === "object" ? "{" + Object.keys(v).sort().map(k => JSON.stringify(k) + ":" + canonical(v[k])).join(",") + "}"
  : JSON.stringify(v);

const receipt = JSON.parse(readFileSync("./8thGuard-agent-receipt.json", "utf8"));
const keys = await fetch("https://8thguard.com/.well-known/8thguard-receipt-keys.json").then(r => r.json());
const rawPub = Buffer.from(keys[receipt.protected.key_id], "base64url");
const pubKey = await crypto.subtle.importKey("raw", rawPub, { name: "Ed25519" }, false, ["verify"]);
const payload = Buffer.concat([
  Buffer.from("8thGuard Agent Evidence Receipt v1\0", "utf8"),
  Buffer.from(canonical({ protected: receipt.protected, body: receipt.body }), "utf8"),
]);
const ok = await crypto.subtle.verify({ name: "Ed25519" }, pubKey, Buffer.from(receipt.signature, "base64url"), payload);
console.log({ integrity: ok, key_id: receipt.protected.key_id, lifecycle_status: "unknown" });

Independent offline verification.

Fetch and pin the Ed25519 public key map from /.well-known/8thguard-receipt-keys.json before verifying an exported agent-evidence-receipt.1 envelope. Offline verification checks content integrity and issuer trust only — current lifecycle status remains unknown until checked against the live service.

XDC Apothem testnet registry.

Opaque commitments (receipt hash, reference and expiry — never private documents) can be reconciled via /anchors on XDC Apothem testnet. Withdrawing a receipt in the service does not automatically update its on-chain entry. No mainnet registry is deployed.

Planned workflows remain separate.

Payment Preflight, RWA Preflight and Communications Evidence are not live APIs and carry no self-service SKU.