HREVNAgentProof

A practical guide to AI evidence verification

What makes AI agent evidence verifiable?

A claim becomes checkable when it is narrow enough, the artifact is available and the verification can be repeated. AgentProof applies that discipline to a local AI agent session receipt.

It can validate a receipt chain and compare checked repository commitments. It does not establish truth, complete capture, exclusive authorship or work quality.

See the current product See the evidence layers

Current marketplace receipts are local, hash-only, unsigned and unanchored.

AGENTPROOF / CHECKABLE CLAIMSLOCAL
  1. 01
    OBSERVECollector-visible eventshost-specific surface
  2. 02
    COMMITCanonical receipt chainSHA-256 · selected metadata
  3. 03
    COMPARERepository commitmentsrecorded state ⇄ current state
  4. 04
    DECLARECapabilities and limitsobserved ≠ complete
CHECKABLEReceipt + commitmentsRepeatable local validation
NOT ESTABLISHEDTruth + authorityNo independent witness
Local receiptSHA-256 chainHost-specific scopeMATCH ≠ correctExplicit limits

Two repeatable checks

First validate the record. Then compare what it committed to.

“Verifiable” is not a general badge. It applies to a specific claim, against a specific artifact, using a specific procedure.

CHECK 01

RECEIPT / VALIDATION

Is the receipt internally consistent?

Local Verify checks canonical form, the event chain and the collector-specific receipt schema.

  • Repeatable from the local receipt
  • Rejects malformed or non-canonical input
  • Preserves the order of recorded commitments
  • Carries capabilities, warnings and limits
Narrow conclusionA valid chain shows internal consistency of the produced receipt. It does not make the producer independent or the capture complete.
CHECK 02

REPOSITORY / COMPARISON

Do checked commitments still match?

Verify compares selected final repository commitments in the receipt with the repository as it exists now.

  • MATCH means checked commitments agree
  • MISMATCH means at least one checked path differs
  • A later legitimate edit can produce MISMATCH
  • Authorship and causation remain unknown
Narrow conclusionThe comparison reports agreement or difference for checked commitments. It does not judge whether either repository state is correct.

Recorded then. Compared now.

A repeatable check is only as strong as the claim attached to it.

receipt_finalRecorded commitment

The collector committed to repository state it checked during the run. The marketplace receipt carries no independent timestamp or signature.

MATCH | MISMATCHCurrent comparison

The result reports whether checked commitments agree now. It does not explain who changed anything, when or why.

Continue from the question you actually have

Two paths, one carefully bounded product.

OPERATIONAL QUESTION

Does this count as an AI agent audit trail?

Learn what a session receipt can support in later review and why it is not an independent or exhaustive audit.

Read the audit-trail guide → CURRENT PRODUCT

Which AgentProof collector can I use?

Compare Codex, Claude Code and OpenClaw behavior, current availability, installation paths and host-specific limits.

Open the product page →

Honest boundary

Checkable evidence is not automatic proof of the larger story.

AgentProof makes a small set of technical claims repeatable. It deliberately keeps broader conclusions outside the result.

It can establish

  • Whether the canonical receipt and internal chain validate
  • Which collector profile and declared limits accompanied it
  • Which paths changed within supported snapshot scope
  • Whether checked final commitments currently match

It does not establish

  • Complete capture or independent observation
  • Exclusive authorship, causation or civil identity
  • That the work is correct, safe or compliant
  • Signing, anchoring, legal admissibility or payment authority

Questions about verifiable AI evidence

Define the claim before trusting the check.

What is verifiable AI agent evidence?

It is evidence paired with a precisely stated claim and a repeatable check. AgentProof supports narrow checks of a local receipt's canonical form, internal chain and selected repository commitments; it does not establish truth, complete capture, exclusive authorship or work quality.

Does a valid AgentProof receipt prove that the AI agent did good work?

No. Receipt validity establishes only that the canonical receipt and its internal chain satisfy the format's checks. It does not establish that the work was correct, safe, useful or compliant.

What does MATCH mean in AgentProof?

MATCH means the repository commitments checked now match the final commitments recorded in the receipt. It does not establish who produced that state, why it exists or whether it is correct.

Are AgentProof marketplace receipts independently verifiable evidence?

Their canonical form, internal chain and selected repository commitments can be checked locally and repeatedly. The receipts are unsigned and unanchored operator-side records, so they are not independent evidence of authorship, completeness or time.

When should I use AgentProof?

Use it before an important AI coding run when a small local record of collector-observed events and checked repository commitments may be useful later. Do not use it as a substitute for sandboxing, backups, review or legal evidence.

From definition to use

See what each collector records—and whether you can install it today.

See AgentProof Session Receipts Read the audit-trail guide