Agent Audit Ledger

Creator Fusion Labs — Agent Protection Suite

Give your AI agent a tamper-evident, hash-chained record of everything it did — a forensic trail you (or an auditor) can verify later. Feed it a list of actions and it returns a sha256 chain where every link depends on the one before it, so a single altered, reordered, or inserted event breaks the chain and is pinpointed exactly.

Built for agents: call it over MCP, curl, JS, or Python. Deterministic by design — the same events always produce the same hashes (no wall-clock is ever mixed into a hash), so any party can reproduce and check the ledger.

---

What it does

• append mode — canonicalizes each event (stable key order), sha256-hashes it (eventHash), and hash-chains it: chainHash[i] = sha256(chainHash[i-1] + eventHash[i]), seeded by your prevHash or a built-in genesis constant. Emits one row per event plus a summary (headHash, eventCount, totalCost, verifiable).
• verify mode — takes events that already carry their eventHash/chainHash, recomputes everything, and reports the first broken link (breakIndex) — catching tampered values, reordering, and insertions.

Chain runs together: pass one run's headHash as the next run's prevHash to build a single unbroken ledger across many runs.

---

Input

Field	Type	Required	Description
events	array	yes	Action objects. Each needs a string action; optional ts, actor, detail, cost (number). In verify mode, include the eventHash/chainHash each event was emitted with.
mode	string	no	append (default) or verify.
prevHash	string	no	Head hash of a prior chain to continue from. Omit to seed from genesis.

An event with no ts keeps ts: null — the ledger never injects the current time, so the chain stays reproducible.

{
  "events": [
    { "ts": "2026-08-19T10:00:00Z", "actor": "agent-a", "action": "search", "detail": "query=competitors", "cost": 0.01 },
    { "ts": "2026-08-19T10:00:01Z", "actor": "agent-a", "action": "fetch", "detail": "url=example.com", "cost": 0.02 },
    { "actor": "agent-a", "action": "summarize", "detail": "3 sources", "cost": 0.03 }
  ],
  "mode": "append"
}

Output

Per-event rows:

{ "rowType": "event", "seq": 0, "ts": "2026-08-19T10:00:00Z", "actor": "agent-a",
  "action": "search", "detail": "query=competitors", "cost": 0.01,
  "eventHash": "…", "chainHash": "…" }

One summary row:

{ "rowType": "summary", "headHash": "…", "eventCount": 3, "totalCost": 0.06,
  "verifiable": true, "breakIndex": null }

In verify mode each event row also carries storedEventHash, storedChainHash, and ok; the summary sets verifiable: false and breakIndex to the first failing index when the chain is broken.

---

Verifying a chain (tamper detection)

Take the rows from an append run, change any one field, and re-run in verify mode. The summary comes back verifiable: false with breakIndex at the altered event.

{ "mode": "verify", "prevHash": "<the append run's prevHash>",
  "events": [ /* the emitted event rows, one of them tampered */ ] }

---

Integration

MCP — expose this actor to your agent via the Apify MCP server and call it by name (apricot_blackberry/agent-audit-ledger) with the input above.

curl

curl -X POST "https://api.apify.com/v2/acts/apricot_blackberry~agent-audit-ledger/run-sync-get-dataset-items?token=$APIFY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"events":[{"actor":"agent-a","action":"search","cost":0.01}],"mode":"append"}'

JavaScript

import { ApifyClient } from 'apify-client';
const client = new ApifyClient({ token: process.env.APIFY_TOKEN });
const run = await client.actor('apricot_blackberry/agent-audit-ledger').call({
  events: [{ actor: 'agent-a', action: 'search', cost: 0.01 }],
  mode: 'append',
});
const { items } = await client.dataset(run.defaultDatasetId).listItems();
console.log(items);

Python

from apify_client import ApifyClient
client = ApifyClient(os.environ["APIFY_TOKEN"])
run = client.actor("apricot_blackberry/agent-audit-ledger").call(run_input={
    "events": [{"actor": "agent-a", "action": "search", "cost": 0.01}],
    "mode": "append",
})
for item in client.dataset(run["defaultDatasetId"]).iterate_items():
    print(item)

---

Pricing

Pay-per-event: a small actor-start fee plus one ledger charge per successful run. Failed runs (bad input) are not charged.

Notes

• Crypto is standard node:crypto sha256 over a canonical (sorted-key) JSON encoding of each event.
• Fully deterministic — no randomness, no wall-clock in any hash.
• Run npm test locally for a free self-check of the chain and tamper-detection logic.