# n8n tool-approval acceptance worksheet

Published November 12, 2026. Companion to the documentation-based n8n tutorial: https://permissionprotocol.com/blog/n8n-human-in-the-loop-tool-approval . Execution status for every case below is **Not run**.

Use a synthetic tool that records a mock refund in a test ledger. Choose a test environment with durable execution storage and a ledger that survives the restart case. Record the actual n8n version and configuration before drawing conclusions.

## Record the environment

| Item | Value to fill |
| --- | --- |
| n8n version and deployment mode | |
| Workflow revision or export checksum | |
| AI Agent, Chat Trigger and Chat node versions | |
| Model and settings used to propose tool calls | |
| Review channel and its authentication configuration | |
| Execution storage and restart method | |
| Mock tool and ledger location | |
| Reviewer identity and assigned authority | |
| Test date and operator | |

For the documented Chat path, use a top-level workflow with Chat Trigger response mode `Using Response Nodes`. The Chat node documentation lists restrictions for Embedded mode, sub-agents and sub-workflows. Verify this topology before adapting it. [Chat node documentation](https://docs.n8n.io/integrations/builtin/core-nodes/n8n-nodes-langchain.chat)

## Define a proposed action

These are application test fields. Map each field to your actual workflow and tool input before testing.

| Field | Synthetic value | Where enforced in your workflow |
| --- | --- | --- |
| operation_id | refund-demo-001 | |
| order_id | order-demo-041 | |
| amount_cents | 4800 | |
| currency | USD | |
| destination | customer-demo-017 | |
| reason | Duplicate order | |
| designated_reviewer | reviewer-demo-a | |
| deadline | Test start plus 60 seconds | |

Record the resolved input at the tool boundary as well as the displayed review message. Include fixed inputs and defaults that affect the action. A reviewer needs the complete action context to make a decision. The documented `$tool.name` and `$tool.parameters` expressions can supply dynamic tool context. Compare their rendered values with the resolved input in your version. [Human review for tools](https://docs.n8n.io/build/integrate-ai/ai-examples/human-in-the-loop-for-tools.md)

## Capture evidence for each attempt

Use a fresh operation identifier for each independent case. Preserve the same identifier across retries of that case. Keep the initial proposal, decision and invocation records so a later attempt cannot overwrite earlier evidence.

| Evidence | Value to record |
| --- | --- |
| Case and operation identifier | |
| Execution and review-request identifiers | |
| Proposed and displayed parameters | |
| Authenticated reviewer and authority check result | |
| Decision value and timestamp | |
| Actual emitted node payload | |
| Branch selected after review | |
| Tool invocation identifiers and resolved inputs | |
| Mock effect count before and after | |
| Error or uncertain execution state | |
| Evidence file references | |

Count tool invocations separately from ledger effects. An invocation can fail before writing an effect. A repeated invocation can also produce no additional effect if the mock ledger deduplicates. Preserve both observations.

## Run the acceptance cases

The expectations below are application acceptance criteria to investigate. Record the behavior observed in your installed version before assigning a result.

| Case | Procedure | Evidence and acceptance question | Result |
| --- | --- | --- | --- |
| Approval | Propose the fixture, inspect its review request and approve as the assigned reviewer. | Does one accepted decision lead to one mock effect with the exact reviewed inputs? | Not run |
| Rejection | Propose a fresh fixture and select denial. | Does the denied call produce zero effects? Capture what the agent does next and any subsequent proposal as a separate call. | Not run |
| Changed action | Pause for review, then attempt to change the amount or destination using the application path you intend to support. | Does the change invalidate the earlier decision and require fresh review? If the path cannot change a pending action, record that limitation and test a new proposal after denial. | Not run |
| Expired review | Wait beyond the configured deadline, inspect the emitted payload and attempt a late approval. | Which branch runs on timeout? Does a late decision produce an effect? Capture the actual decision semantics. | Not run |
| Repeated decision | Submit approval, then repeat the same supported decision action or resume attempt. | Does the accepted result remain stable and the effect count stay at one? Distinguish redelivering a decision from starting a new workflow execution. | Not run |
| Restart | Pause for review, restart the test deployment with its storage intact, then attempt approval. | Is the pending review recoverable and still bound to the original action? Count effects across the restart. | Not run |
| Concurrent decisions | From two separate test sessions, attempt opposing decisions on the same pending request as close together as practical. | Which decision wins? Record both responses and whether execution happens more than once. Repeat with an agreed test count and retain every attempt. | Not run |
| Wrong reviewer | Try the request using a separate synthetic identity without assigned authority. | Is the decision rejected before a tool effect? Identify which component authenticates the person and checks their authority. | Not run |

For Chat, add a typed-response check when custom replies are enabled. The documentation describes typed replies as disapproval with a message. Capture both the decision value and message. The wait limit can resume a workflow, so inspect the timeout payload and selected branch before relying on that path. [Chat approval options](https://docs.n8n.io/integrations/builtin/core-nodes/n8n-nodes-langchain.chat#approval-parameters-and-options)

If the test environment cannot exercise a case, retain **Not run** and record the missing condition. A failed setup is an execution limitation to describe separately from an observed tool-approval result.

## Decide what the observation establishes

For each result, name the layer responsible.

| Layer | Evidence to retain |
| --- | --- |
| Native n8n review behavior | Version, node configuration, decision response and execution trace |
| Application controls | Exact code or workflow condition implementing identity, action binding, expiry or deduplication |
| External action provider | Provider request identifier, idempotency behavior and reconciliation result, when separately tested |
| Remaining production work | Recovery ownership, credential management, retention and unresolved execution states |

An execution log records observations. Record any independently issued signature and its verification procedure separately if your system supplies one.

## Source and publication record

Primary documentation was checked November 12, 2026. No n8n workflow was executed for this worksheet. The top-level Chat topology follows the Chat node restrictions. The general human-review guide describes sub-agent review more broadly, so establish channel-specific support before teaching a nested setup.

This worksheet is an ungated planning and acceptance artifact. Record actual test results only when the corresponding workflow has been run. Keep credentials and personal information out of shared evidence.
