Quickstart
Ten minutes: initialise a project, plan one objective, build it, verify it.
This walks the shortest path that still exercises the whole loop. Use a scratch directory — you are going to let agents write code.
1. Initialise
mkdir hello-devflow && cd hello-devflow && git init
claude --dangerously-skip-permissions
In the session:
/devflow:new-project
DevFlow asks what you’re building, researches the domain, scopes requirements into
v1 and v2, and writes a roadmap. You will be asked to confirm at three gates
(project, requirements, roadmap) — that is gates.confirm_* doing its job.
When it finishes, .planning/ exists:
.planning/
PROJECT.md what you're building and why
REQUIREMENTS.md scoped v1/v2 requirements with IDs
ROADMAP.md objectives, ordered, with success criteria
STATE.md decisions, blockers, session memory
config.json workflow configuration
Already have a codebase?
Run/devflow:map-codebase first. It
analyses stack, architecture, conventions and concerns so the roadmap is grounded
in what you actually have.2. Build the first objective
The single command that does everything:
/devflow:build 1
That chains three phases:
- Plan — the
planneragent breaks objective 1 into atomic jobs, each with success criteria and a dependency order. Thejob-checkerreviews the plan before it runs and asks whether it would actually achieve the objective. - Execute — independent jobs are dispatched to
executorsubagents in parallel waves, each in a fresh context window. Every task produces one atomic commit. - Verify — the
verifieragent tests the result against the objective’s goal and writesVERIFICATION.md.
If you would rather drive each phase yourself:
/devflow:plan-objective 1
/devflow:execute-objective 1
/devflow:verify-work 1
3. See where you are
/devflow:status
Reads every state file and tells you your position and the next action. It is the command to run when you have lost the thread.
4. Keep going
/devflow:build 2
Repeat until the milestone’s objectives are done, then:
/devflow:milestone audit
/devflow:milestone complete
audit checks the milestone actually delivered its requirements and reports gaps.
complete archives it and opens the next.
Choosing the right entry point
Not every change deserves the full loop. DevFlow has four tiers, and picking the right one is the single biggest lever on cost.
| Change | Command | Cost |
|---|---|---|
| Typo, rename, import, dependency bump — sub-30 LOC, one file | /devflow:micro | ~2k tokens |
| Small feature or isolated fix — under 5 files, under 200 LOC, no new abstractions | /devflow:quick | one executor, no planner or verifier |
| Multi-file feature touching more than one subsystem | /devflow:build | full plan → execute → verify |
| Something is broken and you don’t know why | /devflow:debug | structured investigation that survives context resets |
When in doubt, go smaller
/devflow:quick that turns out to be too big will tell you so and suggest
/devflow:build. A /devflow:build on a one-line fix just burns tokens.What to read next
- The loop — what each phase actually produces.
- The
.planningdirectory — every file and who writes it. - Commands — the full command surface.