> For the complete documentation index, see [llms.txt](https://docs.autommate.com/userguide/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.autommate.com/userguide/mcp-and-ai-reference/use-cases-and-examples/configurable-employee-offboarding.md).

# Recipe: configurable employee offboarding

Prepare one Service blueprint that an MSP can configure for several customers. Select each customer's supported applications and released Work Items during deployment; keep customer credentials and tenant mappings out of the shared definition.

**This is a design and acceptance recipe, not a claim of an executed reference implementation.** Start with one application and one controlled test account. A useful first result is a correct, versioned blueprint with a clear manual deployment handover. Automatic deployment is not required to achieve that result.

## Agree the service boundary

The outcome is to restrict an identified employee's agreed access, independently verify the resulting state, and retain the evidence. Decide which systems and access types are in scope. Disabling an account, revoking sessions, removing group memberships and deleting data are different actions; do not substitute one for another.

Use installed native applications and actions where they cover the operation. Entra ID, Google Workspace and ServiceNow are possible targets only when the connected environment exposes the required actions. Inspect the actual catalog rather than assuming those actions exist.

## Define inputs by source

| Data                             | Source to establish                                                                   | Avoid                                                                     |
| -------------------------------- | ------------------------------------------------------------------------------------- | ------------------------------------------------------------------------- |
| Employee and requester identity  | Validated Engage payload and an authorized native identity lookup                     | Using a display name as an immutable account ID                           |
| Effective time                   | Request field with a documented time zone and policy                                  | A fixed deployment timestamp reused for every employee                    |
| Approval record                  | Request reference or earlier Work Item output, then a current read from the authority | Treating a caller-supplied true/false value as approval evidence          |
| Customer connection and instance | Tenant-specific App Connector and approved configuration                              | Asking for API keys in the prompt or hardcoding another tenant's base URL |
| Target account ID                | Verified lookup output                                                                | Guessing an ID or accepting the first ambiguous match                     |
| Execution result                 | Actual native action output                                                           | A sample response or constant success message                             |
| Proof                            | Separate authoritative read after execution                                           | Reusing the Execute step's success flag as proof                          |

Keep these values typed. Record missing/ambiguous fields and resolve them before the affected action. See [input resolution](https://docs.autommate.com/userguide/mcp-and-ai-reference/deployment-input-resolution).

## Compose the smallest useful service

| Phase   | Example Work Item                  | Contract and behavior                                                                                                                                                      |
| ------- | ---------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Engage  | Receive offboarding request        | Document the accepted payload. Choose one supported entry route first.                                                                                                     |
| Govern  | Validate request and authorization | Resolve identity, timing, eligibility and the actual approval record. On refusal or missing authority, prevent the restricted action through the supported phase behavior. |
| Execute | Restrict account access            | Use the native action and tenant App Connector. Return the real target ID and result.                                                                                      |
| Prove   | Verify restricted account state    | Read the target again and compare its actual state with the required outcome. Report mismatch or unavailable evidence.                                                     |

These are recommended labels, not built-in tool names. Repeated Work Item instances need distinct labels and stable lane IDs. Add another Govern item only for a separate policy decision; empty checks and arbitrary Boolean inputs do not establish governance.

Segregation of duties requires an actual identity/policy check and appropriate execution permissions. A Work Item named “SoD” does not implement that control. Keep the requester, approver and verifier roles explicit where required by the customer's policy.

```mermaid
flowchart LR
    A[Validated request] --> B[Verify authority]
    B --> C[Native access restriction]
    C --> D[Independent state read]
    D --> E{Matches outcome?}
    E -->|Yes| F[Record verified result]
    E -->|No| G[Record exception and owner]
```

## Choose an Engage route that actually works

Teams, Slack, WhatsApp, email and AI clients can represent separate intake requirements. A webhook labelled “Teams” is not evidence of a Teams integration. Discover the supported native event, callback and payload before claiming that channel works.

Manager approval through Teams can require a card, response callback, waiting state and correlation to the original request. An available Send Card action alone does not prove that whole interaction is supported. If the installed action cannot meet the requirement, record the gap and use a supported, documented alternative or system-of-record approval. See [native discovery](https://docs.autommate.com/userguide/mcp-and-ai-reference/native-app-trigger-and-data-discovery).

## Hand over for manual deployment

1. Read back every released Work Item and retain the real published version ID. If a release response omits it, recover the unambiguous matching version through a supported read; never manufacture an ID.
2. Read back the Service blueprint and its pinned versions. The reviewed `services_create` apply publishes initial blueprint version 1.0.0.
3. Give the operator the target tenant/environment, selected components, credentials to confirm, and a field-by-field input/source mapping.
4. In the deployment UI, select only the agreed components and configure their bindings. Preserve existing configuration when updating.
5. Check each selected Execute item has its required verification and governance coverage. Do not assume the product automatically forces a matching Prove selection unless that behavior is verified for the installed build.
6. Have the authorized operator review phase behavior, missing inputs and the resulting package before activation.

Use the same blueprint for the next tenant with that tenant's own connections and mappings. A custom integration change creates a new reviewed service version; configuration-only differences should remain deployment configuration.

## Acceptance tests

| Scenario                                                   | Evidence required                                                                                |
| ---------------------------------------------------------- | ------------------------------------------------------------------------------------------------ |
| Valid request                                              | The correct test account is restricted; a separate read proves the target state.                 |
| Missing employee, effective time or required request field | An actionable validation result names the missing field; no restricted action occurs.            |
| Missing, rejected or stale approval                        | Govern prevents the action according to the implemented policy.                                  |
| Same requester and approver when policy forbids it         | The explicit segregation-of-duties rule is enforced.                                             |
| Native action or system step has incomplete metadata       | Authoring identifies the exact missing contract instead of inventing fields.                     |
| Provider timeout or failed action                          | Errors remain visible; no “verified success” is inferred. Inspect prior effects before retrying. |
| Prove cannot read the state                                | The outcome remains unverified or an exception according to the agreed policy.                   |
| Tenant A and Tenant B                                      | Both use their own approved connections; evidence is kept separate.                              |
| Interrupted authoring                                      | Resume from saved object/version IDs and remaining work without creating duplicates.             |

Record actual test evidence before describing this recipe as working in a customer environment. Report blueprint prepared, manual deployment configured, controlled test passed and business outcome verified as separate results.

Related: [Structured service authoring](https://docs.autommate.com/userguide/mcp-and-ai-reference/structured-service-authoring), [reusable tenant service](https://docs.autommate.com/userguide/mcp-and-ai-reference/use-cases-and-examples/reusable-tenant-service), [run evidence](https://docs.autommate.com/userguide/mcp-and-ai-reference/track-run-status-and-verify-results).
