Scaffold-to-Production Promotion Checklist (Delivery)
A checklist with 7 steps: Label the build stage before the first prompt is written.
By InnovaAI ResearchPublished
What are the steps?
Scaffold-to-Production Promotion Checklist (Delivery)
- 01
Label the build stage before the first prompt is written
Decide whether the client engagement is a proof of concept, prototype, pilot, or MVP, and write that label into the statement of work. A PoC answers feasibility, a pilot answers operational fit, and an MVP answers whether it can ship; conflating them is the most common cause of scope overrun on AI-assisted builds.
- 02
Freeze the generated architecture in a one-page diagram owned by a named human architect
Agentic platforms such as Verdent can produce authentication, payments, and admin surfaces from a plain-English brief, but the data model and service boundaries still need a human sign-off before any client code is committed.
- 03
Run a dependency and license inventory on every generated package
Scaffolding pulls in libraries the client never chose. Flag anything with copyleft terms or no maintenance activity in the last 12 months, and replace it before the retainer clock starts.
- 04
Diff agent output against the spec, not against the agent's own summary
Use a review surface such as Agent Monitor to inspect per-file change graphs and risk levels, because an agent's self-report of what it changed is not evidence of what shipped.
- 05
Gate promotion on a test suite the agent did not author
Write or commission the acceptance tests independently. Terminal agents like DSCode will plan, edit, and run tests, but a suite generated by the same model that wrote the code tests the model's assumptions rather than the client's requirements.
- 06
Pin model versions and record the prompt set in the repo
Store the prompts, model identifiers, and tool versions alongside the code so a rebuild six months later reproduces the same output. Unpinned model swaps are the quiet source of delivery regressions.
- 07
Schedule the maintenance handover before launch, not after
Assign ownership of API contract drift to a named person or automated job. Tools like Ripple that detect breaking OpenAPI, Protobuf, or GraphQL changes and open fix pull requests only help if someone is accountable for merging them.
More for AI Code Tools
- Operating ProceduresLeadcode Client Workspace Isolation Setup (Onboarding)
- Operating ProceduresGenerativeIDE Client Onboarding with Write-Gate Audit (Delivery)
- Operating ProceduresEphemerals Branch Environment Lifecycle Management (Delivery)
- Operating Proceduresyume Client Session Recovery and Rollback (Delivery)