AI Agent Execution Receipt Template: Prove What Actually Ran

Five states, not one

“I worked on it” is not a completion state.

Agent summaries often compress several different facts into one confident sentence. Keep these states explicit so an operator can see where evidence ends.

AttemptedAction started. No claim yet that it ran or changed anything.

ExecutedTool or command returned. Result may still be wrong or incomplete.

VerifiedIndependent readback or acceptance check matched the expected state.

FailedAction returned an error, bad result, or failed acceptance check.

BlockedWork stopped at an approval, credential, safety, dependency, or scope gate.

Copy the template

Machine-readable execution receipt.

Save one receipt per bounded increment. Download the exact JSON starter or copy it below. Keep secret values, personal data, raw auth material, and unnecessary payloads out of it.

EXECUTION-RECEIPT.jsonDownload .json

{
  "version": 1,
  "receipt_id": "[unique non-secret ID]",
  "objective": "[one bounded result]",
  "status": "attempted | executed | verified | failed | blocked",
  "started_at": "[ISO-8601 timestamp]",
  "completed_at": "[ISO-8601 timestamp or null]",
  "scope": {
    "allowed": ["[approved path or action]"],
    "excluded": ["[out-of-scope action]"]
  },
  "actions": [
    {
      "action": "[what was attempted]",
      "command_or_tool": "[exact non-secret command or tool]",
      "exit_code": "[integer or null]",
      "result": "[short factual output]"
    }
  ],
  "verification": [
    {
      "check": "[observable pass condition]",
      "expected": "[expected state]",
      "observed": "[actual state]",
      "passed": true
    }
  ],
  "artifacts": [
    {
      "path_or_url": "[safe artifact reference]",
      "sha256": "[optional checksum]"
    }
  ],
  "blocked": ["[approval or dependency still required]"],
  "not_verified": ["[claims the receipt does not support]"],
  "external_changes": ["[public, production, account, or financial changes]"],
  "next_best_step": "[one bounded next action or null]"
}

Browser validator

Check receipt structure before accepting “verified.”

This client-side checker tests required fields, timestamps, action evidence, verification results, and state-specific boundaries. It is a review aid—not cryptographic proof that the receipt is truthful.

Paste an execution receipt

{
  "version": 1,
  "receipt_id": "fictional-local-example-001",
  "objective": "Validate one local static guide build.",
  "status": "verified",
  "started_at": "2026-07-23T00:00:00Z",
  "completed_at": "2026-07-23T00:04:00Z",
  "scope": {
    "allowed": ["local guide files and tests"],
    "excluded": ["public deployment and account changes"]
  },
  "actions": [
    {
      "action": "Run the local guide regression.",
      "command_or_tool": "node tests/guide-regression.cjs",
      "exit_code": 0,
      "result": "Fictional example output: regression passed."
    }
  ],
  "verification": [
    {
      "check": "Local guide route responds.",
      "expected": "HTTP 200",
      "observed": "Fictional example: HTTP 200",
      "passed": true
    }
  ],
  "artifacts": [
    {
      "path_or_url": "guides/example/index.html",
      "sha256": ""
    }
  ],
  "blocked": [],
  "not_verified": ["Public deployment", "Search traffic", "Sales"],
  "external_changes": [],
  "next_best_step": null
}

Runs locally in this browser.

Do not paste secrets or personal data. Receipt text is not uploaded or stored by this page. The run event records only that the checker was used—not receipt contents.

Acceptance check

Do not mark verified until evidence survives these questions.

Was a real action executed?
A draft command, plan, or narrative is not execution.
Did the action return a result?
Record exact exit status or tool receipt without secrets.
Was final state read back?
Check the file, route, API, deployment, database, or public surface that matters.
Were failure and blocked states preserved?
Do not rewrite a failed attempt into a clean success story.
Are unsupported claims listed?
Views are not sales; a click is not delivery; a saved file is not a deployed change.

Filled example

Static-site change with honest boundaries.

ObjectiveAdd one indexable guide and preserve existing product price.

ExecutedCreated page, added sitemap entry, ran static and browser tests.

VerifiedLocal tests passed; live route returned HTTP 200; canonical URL and CTA matched expected values.

Not verifiedSearch ranking, qualified traffic, conversion lift, sale, revenue, and business outcome.

StatusVerified for publication state only.

Boundary

Receipts improve review. They do not grant trust.

Use least privilege and independent checks.

Receipt content can still be incomplete, misleading, or fabricated by a compromised process. Protect credentials, use OS and platform controls, constrain scope, preserve logs, test recovery, and require human approval for high-risk actions.

Need the controls around the receipt?

Generate permissions, risk gates, scope stops, and receipts together.

Safe Walk-Away Agent Kit is a Claude Code-first local toolkit. Launch product price is US$5 with code WALKAWAY5 for the first 10 redemptions; regular product price is US$19. Applicable taxes are calculated by Gumroad.

Inspect the complete kit Get launch price — US$5