# AI agent handoff

## Run

- Task:
- Date and time:
- Reviewer:
- Source record:

## 1. Requested

What outcome did the human ask for?

## 2. Authorised

What actions and boundaries were allowed, reserved, denied, or unclear?

## 3. Executed and changed

| Evidence label | Record reference | What happened or changed |
| --- | --- | --- |
| Observed |  |  |
| Governed |  |  |
| Prevented |  |  |

## 4. Claims and gaps

| Evidence label | Claim or gap | Why this label applies |
| --- | --- | --- |
| Reported |  |  |
| Inferred |  |  |
| Unknown |  |  |

## 5. Verified

- Independent check:
- Actual result:
- Missing evidence:

## Next human decision

- Decision: review / approve / approve with conditions / retry / investigate / stop
- Reason:
- Conditions:

## Evidence label reference

- Observed: a concrete event or result is present in the supplied record.
- Governed: an authority or policy decision is recorded.
- Prevented: the record says an action was denied before its intended effect.
- Reported: the agent or runtime makes a claim the record does not independently establish.
- Inferred: a cautious interpretation is drawn from names or context.
- Unknown: evidence is absent, ambiguous, unsupported, or outside the record.

> Trust boundary: this handoff describes evidence in the supplied record. It does not prove that omitted events did not occur. A receipt identifier is a reference until checked against its canonical store. A successful tool call does not by itself prove the intended outcome.

Free template from Traceplain: https://traceplain.zakgov.com/ai-agent-handoff-template