Step-by-step AI workflow guides for product and engineering teams.

Brownfield Flow

Use this when AI touches existing product behavior, legacy code, or active customer workflows.

1. Map current behavior, dependencies, and known defects.
2. Define the smallest safe change and explicit non-goals.
3. Create a context packet with relevant files, tests, APIs, and constraints.
4. Ask the assistant for a narrow candidate patch, not a broad rewrite.
5. Run regression checks for changed and unchanged behavior.
6. Review the patch until a human can explain it without model authority.
7. Accept with a before/after product demo.
8. Release with monitoring, rollback criteria, and a learning log.

Greenfield Flow

Use this for new features, services, internal tools, and MVP surfaces.

1. Frame the user problem, MVP boundary, success metric, and non-goals.
2. Write an architecture note with data contracts, ownership, and risk controls.
3. Prepare a model-neutral prompt contract for one vertical slice.
4. Generate a scaffold or candidate implementation inside that boundary.
5. Validate tests, performance, usability, and architecture fit.
6. Run product acceptance against the original customer scenario.
7. Pilot with a controlled audience, owner, monitoring, and support path.
8. Scale, stop, or iterate from pilot evidence.

Config & Infrastructure

Use this when AI touches deployment, policy, configuration, environments, or infrastructure as code.

1. Declare the operational intent, affected customers, and non-goals.
2. Map blast radius, dependencies, permissions, cost, security, and rollback.
3. Inventory current state, runbooks, and validation commands.
4. Ask the assistant for a minimal config diff and rollback notes.
5. Review plan output, security posture, and operational readiness.
6. Approve the change window, owner, communication, and rollback trigger.
7. Roll out under observation with incident path ready.
8. Complete post-checks and update runbooks.

Claude Code

1. Open the Artifacts page and copy the CLAUDE.md starter into your repository.
2. Copy the workflow runbook that matches the change type into your repo or planning doc.
3. Update CLAUDE.md with project build commands, review rules, and the AIDLC guide location.
4. Start Claude Code from the repository root.
5. Ask Claude to use the matching flow and produce a candidate artifact plus evidence notes.
6. Run the required checks and update the evidence ledger before merge or release.

Use Brownfield Flow from docs/aidlc-workflow.md.
Change only the approved scope.
Return: candidate patch, tests to run, risks, and evidence ledger updates.

Codex

1. Open the Artifacts page and copy the AGENTS.md starter into your repository.
2. Copy the chosen workflow guide, prompt contract, and review gate into your team workspace.
3. Add durable repo instructions in AGENTS.md so Codex sees the workflow expectations.
4. Open Codex in the app, CLI, IDE extension, or web surface from the repository context.
5. Prompt Codex with the workflow type, risk level, allowed scope, and required evidence.
6. Review all edits, commands, and test results before accepting the output.
7. Use the evidence ledger for acceptance, release, and learning decisions.

Use Config and Infrastructure Changes.
Risk: high.
Do not apply changes directly.
Return: minimal diff, plan validation, rollback notes, and post-checks.

Reusable Prompt Contract

1. Name the workflow: Brownfield, Greenfield, or Config/Infra.
2. State product intent, risk level, and non-goals.
3. List relevant files, APIs, tests, and constraints.
4. Define allowed and prohibited actions.
5. Ask for output, verification steps, risks, and evidence updates.
6. Keep human review and release decisions outside the assistant.

Workflow:
Risk:
Intent:
Context:
Allowed:
Not allowed:
Evidence required: