Guide

For a project that already has code — whether it’s a mature product with docs and tickets, or something you inherited and barely understand yet.

If your code works but was never really planned, read Taking a prototype to production instead. Same command, different emphasis.


Run it

cd your-project
claude
/archflow:onboard

Nothing is written until you approve it. Onboarding reads, analyzes, proposes — then waits.


The three phases

Phase A — Interactive collection (you’re present, ~5 questions)

Archflow detects your stack and project type (fullstack, frontend_only, backend_only, mobile), then asks where your existing context lives — Jira, Notion, Linear, GitHub, Confluence, Slack, Google Drive, or nothing at all.

If a tool isn’t connected yet, it runs /archflow:setup-mcp inline rather than making you stop and come back.

Phase B — Autonomous analysis (you can walk away)

Up to nine agents run, layered by dependency:

Layer 1   codebase audit · doc deep-dive · design extraction · route/API extraction
Layer 2   product-strategist · ux-designer            (need Layer 1 output)
Layer 3   api-contract-architect · dsl-generator · feature-planner

The audit walks a checklist filtered by your project type, recording what exists and what doesn’t. The extraction agents reverse-engineer your design tokens, component patterns, and endpoints from the code as it actually is — not as documentation claims it is.

Progress is saved to .onboard-progress.yaml, so an interrupted run resumes instead of restarting.

Phase C — Synthesis (you’re back)

Roadmap reconciliation. Stories whose code the audit found already shipped are moved into a completed release. Work in progress lands in the active release. Your roadmap opens describing reality.

Phase determination. Archflow works out which phase you’re genuinely in, so you don’t redo finished work.

Gap report. A plain list of what’s missing — no API contract, no test coverage, no design system. This is the most useful artifact onboarding produces, and the one worth reading twice.


What you end up with

File Reverse-engineered from
.archflow/project-context.md Code, README, imported docs
.archflow/roadmap.yaml Epic labels + release pipeline
.archflow/backlog.yaml Remaining scope, as stubs
docs/api-contract.md Actual routes and handlers
design-artifacts/theme.yaml Colors, spacing, typography in your code
design-artifacts/styled-dsl.yaml Component patterns
design-artifacts/user-flows.md Navigation and routing

Your CLAUDE.md also gains an architecture section derived from the analysis.


Already onboarded?

Running /archflow:onboard again is safe. It detects existing state and shows a status summary instead of redoing the work — plus it validates your roadmap.yaml against the current schema, backfills any missing template files, and offers to auto-fix format violations.

On a v1.0 project it will point you at /archflow:migrate rather than half-upgrading you.


After onboarding

/archflow:status           # confirm the phase it picked
/archflow:feature          # add work
/archflow:groom S2-11      # detail a stub before building it

Read the gap report first, though. It usually reorders your priorities.