Detects the newest fraud in agent payments — the forged "this was approved." Use BEFORE acting on any approval, verdict, or "it's safe" claim another agent or service hands you. Cryptographically verifies Fidacy-signed verdicts against the issuer's public keys, so you trust math, not a promise.
Coding
soja-trace-evidence-checks
Try itUse when an agent needs a public bounded deterministic W04 trace-integrity verification from supplied redacted JSON evidence.
What it does
Use when an agent needs a public bounded deterministic W04 trace-integrity verification from supplied redacted JSON evidence.
The skill document
SOJA W04 Trace Integrity Verifier
Public machine-callable W04
Submit one bounded, redacted expected-versus-observed trace evidence manifest:
POST https://soja-w04-public-evaluate.slowsleeper1.workers.dev/v1/trace-integrity-verifier/evaluate
Content-Type: application/json
- Maximum request payload: 64 KiB.
- No API key is currently required.
- The result is deterministic and machine-readable JSON.
- Responses use
Cache-Control: no-store. - Invalid JSON, unsupported methods/routes/content types, oversized payloads, secret-bearing input, unsupported resources, and invalid bounded inputs are safely rejected.
Request
{
"resource": "trace-integrity-verifier",
"request_id": "example-001",
"input": {
"expected": {"trace_ids": ["trace-a", "trace-b"], "links": [["trace-a", "trace-b"]]},
"observed": {"trace_ids": ["trace-a"], "links": []}
}
}
request_id is a non-empty string up to 128 characters. trace_ids and link values must be strings. The verifier compares only supplied evidence; absence of supplied evidence is not proof of runtime failure.
Response
{
"invocation_status": "COMPLETED",
"request_id": "example-001",
"resource": "trace-integrity-verifier",
"result": {
"status": "INCOMPLETE",
"complete": false,
"expected_trace_count": 2,
"observed_trace_count": 1,
"missing_trace_ids": ["trace-b"],
"missing_links": [["trace-a", "trace-b"]]
}
}
Safety and limits
Do not submit secrets, credentials, tokens, customer data, production traces, or other sensitive payloads. No customer storage, telemetry, configured secrets, payment, checkout, or source download is provided.
A Cloudflare Workers Rate Limiting binding is configured at a nominal 10 requests per 60 seconds. Cloudflare enforcement is location-dependent and eventually consistent and should not be interpreted as a strict globally synchronized request ceiling.
Evidence boundary
Public availability and controlled tests are not commercial evidence. Paid invocations, customers, Tier C/A evidence, payments, and revenue remain zero.
W05
The evaluation cost-reconciliation sentinel remains descriptor-only and is not publicly callable.
Related skills
Generates evidence obligations for a claim or action, evaluates existing evidence against them, and returns a structured verdict (PASS / SOFT_PASS / BLOCK /...
Pre-send verification for outbound agents — a small, separate guardian you put in front of send() so every message (email, social, helpdesk reply) gets an in...
Independent fail-closed second opinion before acting: allow/review/block a risky action, fact-check a claim, screen text for prompt injection, or flag PII/se...
Use the Cryptowerk Verifiable Data skill to implement an "Agent Flight Recorder" that creates an immutable, cryptographically verifiable audit trail for Open...
Prove your AI agents actually did the work — catch fake-success, dry-run theater, and stub code before it ships. Static fake-success scanner + live-artifact...