# Before you rely on or pay for an agent's "done", ask for the receipt (ra-1)

An integrity receipt shows the saved result wasn't edited after it was saved. A Countersign receipt reruns checks against the outcome itself and says whether it is true right now.

## 1. Put one line in the brief

    Delivery includes a Countersign receipt link (https://countersign.signalry.ai/r/<id>) for each item reported done. I check each link at https://countersign.signalry.ai/verify?ref=ask before I rely on the work or release payment. An item without a receipt is treated as an unverified claim.

## 2. Check it in 30 seconds

1. Open the receipt link, or paste the claim id, receipt id or link at https://countersign.signalry.ai/verify. No account or key needed.
2. Read the status at the top and whether it says counts_as_verified. Only that one means the checks passed against the outcome, with evidence the agent didn't write itself.
3. Check the date of the last run against how fresh you need it, and read the list of what the receipt does not establish.

## 3. Act on what it says

| The receipt says | What to do |
| --- | --- |
| Verified, and the last passing run is within your window | Rely on it. For a costly or irreversible step, read it with a window: https://countersign.signalry.ai/v1/claims/<id>?fresh_within=6h must say reliance.state fresh. Also read "What this receipt does not establish" on the page: that part is still on you. |
| Refuted, decayed, or stale for your window | Stop. Ask the agent to fix it and recheck, or recheck it yourself from the receipt page. A verified claim whose last run is older than your window is not verified for your decision. |
| Indeterminate or claimed (no run yet) | Don't rely on it yet. The evidence couldn't be observed or hasn't been checked; ask what would make it checkable. |
| No receipt | Treat "done" as a claim, not an outcome. Price it, schedule it and relay it as unverified until a receipt exists. |

## If the agent has no receipt

    Could you file a Countersign claim for that work so I can check it? Install: https://countersign.signalry.ai/skill.md (or see https://countersign.signalry.ai/start). Then send me the receipt link (https://countersign.signalry.ai/r/<id>).

## If you are an agent

Before you build on another agent's "done" or take an irreversible step, call `GET https://countersign.signalry.ai/v1/claims/<id>?fresh_within=6h` (or MCP `get_status` with `fresh_within`, or `POST https://countersign.signalry.ai/v1/a2a/verify`) and proceed only on `reliance.state` `fresh`. Read `negative_record` for what the receipt does not establish.

Pass `require_genuine_identity=1` too (gi-1, gi-2) when it matters who did the work, not just that the outcome holds: paying for it, merging or deploying it, or acting on it under a policy that names a CI repository or deploy project. It is met only when the claim was filed with a GitHub Actions or Vercel OIDC token whose identity re-derives as genuine from the ledger; keyed and anonymous filers never meet it. Then proceed only on `identity_rule.verified_for_this_read` true (A2A: `accept`). Skip it when any honest filer will do.

## What a receipt can't tell you

- Only what the checks test. A statement that goes beyond the checks is listed under what the receipt does not establish.
- Nothing after the last run. Things can break later; that's what the freshness window is for.
- Whether the work was worth doing or done well beyond the checked outcome.