Purpose and fit

Use this when an AI-assisted workflow is becoming recurring work, moving to another team or depending too heavily on its creator. Build a handover pack that explains how to operate the current version, recognise a problem and find help.

The immediate outcome is a draft handover pack and a plan to demonstrate readiness with a named receiving person or team. Readiness is established only by the subsequent authorised rehearsal and acceptance. A folder of prompts alone leaves important knowledge about source updates, review and failure recovery unstated.

Before you start

Use the current approved workflow notes, prompts, source list and known checks you already have. An anonymised past work item is useful but not necessary to draft the pack. Include version and owner information where known; do not share credentials or create test data just to start. Plan any later receiver rehearsal using existing authorised examples in a safe setting.

Expect a draft operating pack and receiver-rehearsal plan, not demonstrated readiness or an accepted handover. The pack should separate documented instructions, unresolved dependencies and steps requiring approval; acceptance and rehearsal results remain blank until observed.

Required inputs

  • The current workflow definition, approved task boundaries and named operational owner.
  • Current prompts or configurations, source references and tool dependencies, without embedded credentials.
  • Acceptance checks, known failure cases, exception routes and the existing fallback.
  • The proposed receiving operator and any existing sanitised examples; an authorised operator and safe rehearsal setting are needed before a later operational rehearsal, not to draft the pack.

Practical workflow

  1. State the workflow’s purpose, eligible tasks and exclusions. Identify its current version and the person responsible for deciding when instructions or tools must change.
  2. Document the start-to-finish operating routine, including input preparation, review, output acceptance and recordkeeping. Mark steps that need another person’s approval rather than assuming the receiver inherits it.
  3. Package reusable components with clear names and locations. List source and tool dependencies, update responsibilities and access requirements. Refer to the approved access process; never place secrets in the pack.
  4. Include known limits and recovery guidance. Show what a failed check looks like, where incomplete work is held and who can authorise a restart or manual alternative.
  5. Ask the receiver to complete a representative rehearsal using only the pack while the creator observes. Include a missing-input or failed-check case. Record questions and gaps rather than coaching through them invisibly.
  6. Revise the pack and agree acceptance with the receiver. Set a review trigger for source, tool or ownership changes, and record where obsolete versions will be marked so they are not reused accidentally.

Expected output

A versioned operating pack and a handover acceptance record:

Workflow name/version: [identifier] Purpose and eligible work: [scope] Workflow owner/receiving operator: [roles] Required access: [approved process references] Inputs and dependencies: [sources/tools/owners] Operating steps: [run sequence] Acceptance checks: [criteria and reviewer] Known limits and exceptions: [triggers/routes] Fallback/restart authority: [process and owner] Rehearsal evidence: [case, result and gaps] Handover decision: [accepted/conditional/not ready] Maintenance/review trigger: [owner and condition] Previous version status: [location and marking]

Hypothetical example: a monthly internal briefing moves to a colleague. A rehearsal reveals that the source list omits one manually checked update. The team adds that dependency and its owner before accepting the handover.

Human checks

The receiver confirms they can understand the instructions and access the approved materials. The workflow owner confirms ongoing support and review capacity. System owners verify actual access separately. Do not treat successful observation, document receipt or silence as evidence that the receiver has accepted responsibility.

Limitations

A handover rehearsal covers only the cases attempted. Tool changes, unusual inputs and staff availability can expose new gaps. The pack documents operations but does not transfer authority, enforce permissions or establish that the workflow is suitable for a different team’s context.

Starter prompt

Create a handover pack for this AI-assisted workflow using the approved, sanitised material provided: [workflow, components, sources, boundaries and checks]. The current owner is [owner] and the proposed receiver is [role]. Preserve existing permissions and mark missing ownership or access information unresolved.

Include these fields: workflow name and version; purpose and eligible work; owner and receiving operator; required access references; inputs and dependencies; operating steps; acceptance checks and reviewer; known limits and exceptions; fallback and restart authority; rehearsal evidence and gaps; handover decision; maintenance owner and review trigger; previous-version status. Add a concise run sequence, a normal-task and failed-check rehearsal plan, and acceptance questions for the receiver. Flag undocumented knowledge; leave future evidence and acceptance unresolved. Do not grant access, share externally, transfer responsibility or claim acceptance.

Adapt the requested fields to the authorised inputs actually supplied. Flag inconsistent versions, undocumented manual steps and unknown access as questions for their owners. Do not infer that receipt of the pack transfers authority or that a planned rehearsal has passed. Keep a safe fallback for unresolved operating steps. State the expected output and acceptance criteria before drafting. Ask for essential missing information; where a useful partial draft is possible, label its limits and assumptions. Do not invent facts, evidence, results or approvals.

Data, model variability and your responsibility

Use relevant data that exists or that you provide, with accuracy and completeness sufficient for the task. Different AI models and versions may produce different results, and responses may vary between runs. You are responsible for checking and validating the output against your own requirements, source information and acceptance criteria before relying on it.