RestartCrew

How We Work

This document explains the development workflow for RestartCrew — a project built collaboratively by humans (vibecoders) and AI coding agents.

---

Development Model

RestartCrew uses a milestone-driven, checkpoint-gated development process:

1. Each milestone has a set of atomic, checkable tasks
2. Each milestone ends with a verification checkpoint
3. Nothing proceeds until the checkpoint passes
4. All work is documented in DEVLOG and DECISIONS

---

Agent Workflow

When an AI agent starts a task:

1. Read MAIN_DEVPLAN_RESTARTCREW.md → Understand WHAT to build and find the specific task
2. Read AGENTS.md            → Follow coding standards and architecture rules
3. Implement                 → Write code following the guidelines
4. Verify                    → Run the checkpoint criteria
5. Document                  → Update DEVLOG, DECISIONS, ROADMAP

Before you code, always check:

• Which milestone and task number am I working on?
• What are the acceptance criteria from the dev plan?
• Are there relevant decisions in DECISIONS.md I should know about?
• Does the DEVLOG show any blockers or context from previous work?

After you finish, always update:

• docs/DEVLOG.md — What you did, files changed, blockers, next step
• docs/DECISIONS.md — Any non-trivial decisions you made
• docs/ROADMAP.md — Move items between sections if milestone status changed
• Strategy status table — Check off completed tasks

---

Document Update Cadence

Document	When to update
DEVLOG	After every work session
DECISIONS	When making a non-trivial technical or product decision
ROADMAP	When a milestone’s status changes (not started → in progress → done)
Strategy task table	When individual tasks are completed
MAIN_DEVPLAN_RESTARTCREW	When product requirements or execution plan change

---

Quality Gates

Every milestone has a checkpoint. Common verification patterns:

• Foundation: Build succeeds, tests pass, lint clean, CI green
• Design System: All primitives render at all breakpoints, tokens applied correctly
• Database: Migrations run clean, RLS policies tested, seed data loads
• Features: Feature works end-to-end, loading/error states handled, accessible
• Launch Prep: All checkpoints pass, security audit clean, performance targets met

---

Key Principles

1. Small, verifiable steps — Complete one task fully before starting the next
2. Document as you go — Don’t batch documentation; update after each session
3. Follow existing patterns — Look at how similar things are done in the codebase
4. Security by default — RLS on every table, auth checks on every protected route
5. Mobile-first — Build for mobile, enhance for tablet/desktop