A human-readable audit of a smart contract costs somewhere between $5,000 and $15,000, and a complex multi-chain protocol can run past $250,000. CertiK, the largest Web3 security auditor, does not publish a rate card at all — the price is a custom quote.
Now price the other side of the same question: someone hands you an artifact produced by an AI agent. What does verifying it cost?
Today: nothing. You read it and believe it, or you paste it into another chat window and hope. There is no check, no signature, no price.
That gap is not an oversight in the crypto world — it is the norm. But the infrastructure around agent output has already started behaving the way infrastructure does. Cloudflare's own documentation for error 1010 says:
"The owner of this website has banned your access based on your browser's signature."
— Cloudflare Support docs, Error 1010, last updated Apr 23, 2026
A browser signature is a fingerprint, not a promise. When a pipeline can be fingerprinted, it can also be checked. The same argument applies to agent output: a signed, hashable artifact can be verified by someone who was not in the room.
The market, for calibration
| What | Typical price | Source |
|---|---|---|
| Smart contract audit (standard) | $5,000–$15,000 | cryptojobslist.com, Jul 28, 2026 |
| Complex multi-chain audit | from $5,000 up to $250,000+ | spark.money |
| CertiK | custom quote, no public rate card | kannaudits.com, Aug 29, 2026 |
| Agent artifact / skill verification | $0.005 per call | this post |
The last row is the interesting one. Verifying an agent's output is structurally cheaper than verifying a contract, because the input is text, not a state machine with value at rest. The hard part was never compute. It was having a check that anyone can run.
What "cheap" actually means here
A verifier is only useful if it is:
- Runnable by the skeptic. No account, no upload, no sales call. Ours runs locally, offline.
- Reproducible. The verdict carries the SHA-256 of the exact bytes checked. Two people lint the same file and get the same answer.
- Tamper-evident. The result is signed with Ed25519; verification needs the public key that travels with the signature, not a secret held by us.
- Cheap enough to be unconditional. At $0.005 a call you check every artifact, not the one you are nervous about. Unconditional checking is the only kind that catches anything.
The stack, briefly
- skilllint — private, offline linting of agent skills: structure, triggers, dead references, embedded secrets, injection patterns.
- attest — verification of an artifact before you act on it: schema conformance, emptiness checks, provenance, entitlement. Signed receipts, verifiable offline.
- drift — mandate checks before an action is allowed to happen: per-action and per-day limits, counterparty allowlists, drawdown stops, kill switch.
- agentpay — accounting by actual spend, not by plan.
- mitracore — one entry point that wires them together: $49 per environment per month, prepaid, $0.005 per call, 20 free checks a day without a card.
Enterprise buyers pay four to five figures for a human-readable audit of a contract. We think the equivalent for agent output should cost cents and be runnable by anyone.
If your agents produce artifacts other agents act on, try the linter on one of your own skills before you trust the next output. It runs on your machine; nothing leaves it.
Sources: Cloudflare Support docs (Error 1010, Apr 23, 2026); cryptojobslist.com (Jul 28, 2026); spark.money; kannaudits.com (Aug 29, 2026).
Top comments (0)