Getting Started

Getting Started

This walkthrough shows the smallest useful way to use 8-habit-ai-dev: define the work, make the architecture decision explicit, implement with context, review before commit, and plan deployment before production.

Tip

Use a real small feature. The workflow is clearer when the output matters.

1. Define The Work

/requirements I need to add a rate limiter to our /api/search endpoint

Expected output: a short PRD with purpose, users, scope, out-of-scope items, success criteria, and Definition of Done.

Human checkpoint: correct the intent before any code is written.

2. Decide Architecture

/design

Expected output: at least two viable options with trade-offs for important decisions. The AI proposes; the human chooses.

Use an ADR when the decision affects architecture, public API, persistent data, security, or more than a small local change.

3. Break The Work Down

/breakdown

Expected output: atomic tasks with dependencies, priorities, and scope boundaries. Tasks should be small enough to review and test independently.

4. Build With Context

/build-brief

Expected output: a task brief based on existing code, local patterns, edge cases, and files likely to change.

Only implement after the agent has read the relevant code.

5. Review Before Commit

/review-ai

Expected output: findings ordered by severity, with concrete fixes. This is the default review step for AI-generated work.

Add focused checks when needed:

• /security-check for auth, secrets, user input, dependencies, config, or infrastructure changes.
• /cross-verify for larger plans or pre-PR confidence.
• /scrutinize when the question is whether the change should exist at all.

6. Plan Deployment

/deploy-guide production

Expected output: staging steps, production steps, rollback commands, verification, and post-deploy checks.

For provider-managed canaries or capacity changes, the guide also asks you to reconcile the planned target with the provider-selected target before calling the change complete.

7. Observe And Reflect

/monitor-setup
/reflect

Expected output: monitoring gaps closed, runbook updates identified, and a short lesson captured for future work.

When To Use Less

Small, obvious fixes usually do not need every step. They still need /review-ai, and production-impacting changes still need /deploy-guide.

See Workflow Overview for skip guidance.