Question 1

What is a "signed agent receipt"?

Accepted Answer

A cryptographic artifact produced when an AI agent takes an action. The receipt encodes the action's canonical inputs (from, to, channel, payload, timestamp, consequence) and a SHA-256 hash of that pre-image. Optionally, the receipt also carries an Ed25519 signature. Anyone with the URL can recompute the hash and verify offline.

Question 2

How is this different from a log line?

Accepted Answer

A log line is whatever the system chose to write. A signed receipt is a deterministic function of the action — same inputs, same hash, always. The hash is in the URL path. The pre-image is in the URL query. Both are independently verifiable.

Question 3

Does the issuer need to be online for me to verify?

Accepted Answer

No. The receipt URL is fully self-contained: it carries both the expected hash (in the path) and the canonical pre-image (in the query). Anyone with SHA-256 in a standard library can verify offline.

Question 4

What does the Ed25519 layer add?

Accepted Answer

The hash proves the action wasn't modified. The Ed25519 signature additionally proves the action came from a specific key holder. The two layers verify independently — a receipt without the signature still has hash-level attestation.

Question 5

Is this on-chain?

Accepted Answer

Not currently. The current spine is off-chain SHA-256 + per-house Ed25519 keys. On-chain anchoring (Base) is on the roadmap. The off-chain receipts are useful today.

Question 6

Where can I see this running?

Accepted Answer

The Floor at ssx360.com is the live demonstration. Every agent turn has a "Verify ↗" button that opens the receipt drawer; the URL is shareable. The Floor itself is a fictional sim, but the proof spine underneath is real and deterministic.