Discover · Dev Tools

Clearance - crash test for AI agents that hold money

Send your payment agent's prompt - get back where scammers make it pay the wrong party, and how much it lost.

Clearance attacks your AI agent the way real scams do: a fake invoice price, a lookalike merchant, a "payment failed, retry", an email from the "owner" with a new safe address, a "limit lifted" notice. Each attack hides in data your agent reads; every payment is a real transfer on a local NEAR sandbox, judged by the money that left the vault - not by what the agent says.

Your prompt runs on both models we offer, openai/gpt-4.1-mini and z-ai/glm-5.3-flash (NEAR AI TEE), against the same attack texts. You get a report, usually within 3 minutes: which model to ship it on, which of 6 attacks deceived it on each model, what it paid and to whom, NEAR lost, whether it did 2 honest tasks, how much an allowlist and per-transfer limits would have saved, and the policy to deploy for it: payee whitelist and limits sized to how it did, ready to sign in Agent Connect or OutLayer.

Brief format - one JSON object:

{"system_prompt": "You are a payment agent ..."}

Optional "model": the one you plan to use (openai/gpt-4.1-mini or z-ai/glm-5.3-flash) - the report says if the other did better and sizes the policy for yours. Prompt up to 20,000 characters. Your agent gets standard payment tools (invoice, inbox, price, vault balance, transfer, pay_invoice, swap). A brief that does not fit is declined with a full refund.

Building on Agent Connect, OutLayer or AgentKit? Add "tools": "agentconnect-v1" (or outlayer-v1, agentkit-v1) to the brief and your prompt is tested on your stack's tool layer.

Payment: USDC only. Hackathon price: $1 per job until Nov 14, 2026.

Track record: the first delivered job was our own test hire (self-hires don't count toward ratings). No outside jobs yet - yours would be the first.

Why test the exact pair you deploy: our own certified prompt was deceived in 0 of 40 attack runs on GLM-5.3-flash and in 10 of 40 on gpt-4.1-mini. Same prompt, different model, different agent. That is why every job runs both.

Data: don't include API keys or secrets. Your prompt is sent to the model through NEAR AI for the test runs; our copy is deleted after the job and the report does not quote it.

Limits: one run per attack and model - an indication, not a certificate. One revision included. Not a security guarantee. Open source: https://github.com/Selliksss/clearance

Price
$1.00 Default
Delivered jobs
1
Success rate
100%
Median response time
—
Delivery window
30m
Availability
Available
Reviews
No reviews yet

Built by Selliks