Skip to content
OperatorNest

AI agent receipt template

Draft a reviewable record for one AI agent task.

Example: Sam’s overnight vendor research: Compare five vendors overnight. Public web pages and review sites were consulted; one price remains unverified. No approval was needed because nothing was sent or booked.

Describe the task

Example: Sam’s overnight vendor research. Replace the sample details with records from your own task.

One timed action per line. Include the outcome.

One system per line. Include only systems the task actually used.

One proposed action, decision, and decision time per line. Leave empty if none applied.

One source per line. Write a label, then its exact URL. Linked sources appear as Markdown links in exports.

This is a local template. It does not check the sources or run the task. Do not paste passwords or payment details.

Task receipt

Illustrative example. Checked field guidance: 28 September 2026.

Example: Sam’s overnight vendor research

Research · draft receipt

Purpose: identify this task and distinguish the example from a verified run.

Trigger and time

Purpose: show what started the task and when.

Sam’s Sunday night request; exact time not recorded in this illustrative example.

Original request

Purpose: preserve the words the result must answer.

Compare pricing and support quality for five project management vendors. Have a summary ready Monday morning.

Actions and timing

Purpose: reconstruct the order of work and check each step.

  1. Sunday night, exact time not recorded in this illustrative example: Read public pricing pages and support reviews for five vendors.
  2. Monday morning, exact time not recorded in this illustrative example: Prepared a comparison; Vendor E’s price remained unverified.

Systems and sources

Purpose: show what the task touched and where each claim came from.

Public vendor websites; third-party review sites. Exact source URLs are not recorded in this illustrative example and are needed before relying on the comparison.

Approval decisions and times

Purpose: record each proposal, decision, and decision time.

None in this example; no approval decision or time. Confirm no consequential action occurred.

What was sent or changed

Purpose: name the exact external action or change, including when there was none.

Nothing sent, booked, paid, published, or changed outside the draft comparison.

Result and exceptions

Purpose: separate the delivered work from gaps and failed steps.

Comparison prepared for Sam. Vendor E’s price came from a cached review page and needs checking against a current vendor quote.

Errors and retries

Purpose: keep failed attempts and retries visible.

Vendor E’s current price was unavailable. Retry with a direct vendor quote.

Model used

Purpose: identify the model and measured usage for this task.

Not recorded in this illustrative example; check the run log.

Whose key or access

Purpose: show whose key or access was used.

Not recorded in this illustrative example; check the run log.

Cost, if known

Purpose: tie measured charges to this task.

Not recorded in this illustrative example; check measured usage and paid tool charges.

What is still open

Purpose: show unresolved work after the result was delivered.

Verify Vendor E’s current price and obtain exact source URLs before relying on the comparison.

Verification and next step

Purpose: name what a person should check before relying on the result.

Open each source and verify Vendor E’s current price before a purchase decision.

How it works

The template groups one task into its request, timed actions, systems, sources, approvals, result, exceptions, model usage, cost, and checks still needed.

OpenTelemetry describes named events for meaningful points in time and GenAI attributes for operations, tools, models, and token usage. These references support the logging fields, not the illustrative vendor findings. Checked 28 September 2026.

The form stays in this browser. Copy, download, and print use the current draft; none of these actions runs the task.

Sources and checked dates

Limits

  • This creates a draft record. It does not observe an agent or verify that an action occurred.
  • Replace example entries and fill in exact timestamps, source links, approval decisions, model usage, and cost from the system that ran the task.
  • A receipt can miss actions if the underlying system did not record them. Check consequential claims against the linked sources and system logs.

Common questions

What belongs in an AI agent receipt?
Record the original request, dated actions and systems touched, sources, approval decisions, result, failed or uncertain steps, and any measured model usage or cost.
Does this template capture a real agent run?
No. It prepares a record for you to complete from task logs and source material. The Sam example is illustrative.
How should I record a missing price or failed step?
Keep it in the timeline and exceptions. Name the source used instead, mark the figure unverified, and state what needs checking before a decision.
Can I use the JSON as an audit log?
The JSON is an export of the text shown here. It has no independent proof of when an action happened or who approved it.

Hand recurring vendor research to your operator, with approval before consequential steps and a receipt after each run.

Hand off your first task tonight.

Tell us your email and what you'd hand off first. We'll send your access details and help you set up your operator.