Starter tier

The disciplined, batteries-included AI workflow for developers who want structure without learning a six-role vocabulary on day one. 60 seconds from install to first task. Defaults are opinionated; nothing to configure.

Audience #

Pick Starter if any of these describe you:

You're new to structured AI development workflows and want a sensible default rather than a configuration menu.
You've been writing your own CLAUDE.md for a few weeks and want something more disciplined without committing to the full system.
You're evaluating aidokit and don't want to learn the full vocabulary before deciding whether it's worth your time.
You're scaffolding a small project (solo or 2–3 people) and don't yet need role separation, MCP scoping, or audit artifacts.

If you find yourself wanting more after a few weeks, see Upgrade path at the bottom of this page.

Install #

npx aidokit init --tier starter

That's the whole thing. No interactive prompts (defaults are taken), no optional pieces to choose.

Or, if you run npx aidokit init with no flags, the interactive picker will offer Starter as the default choice — accepting it puts you on this tier.

What you get #

Component	Starter
`CLAUDE.md`	Short, opinionated. Three slash commands, no role list.
Slash commands	`/intake` (work entry), `/implement-task` (run one task)
Roles	Implicit — the orchestrator decides; no role files to read
Task graph	Beads if available, flat-file `agent-artifacts/tasks.md` fallback
MCP servers	None by default. Add later with `aidokit mcp add`.
Watchdog hooks	Basic scope check; no capability declarations
`docs/` skeleton	Minimal — project brief + decisions log
Validation commands	Whatever your stack pack suggests (lint, typecheck, test)

What you do not get (and why) #

The full six-role workflow. Researcher / Architect / Planner / Builder / Tester-Reviewer / Maintainer are deliberately hidden at this tier. They're useful, but they're vocabulary cost — and Starter exists to delay that cost until you decide it's worth paying.
Per-role MCP scoping. All MCPs you install are available to the orchestrator. Standard tier adds the scoping discipline.
Capability declarations / audit export. Strict-tier features. If you need an audit trail today, install Strict directly.

Upgrade path #

When Starter feels too thin (typical signs: you're hitting scope-leak problems the basic watchdog doesn't catch, or you want different MCPs visible to different agents), upgrade:

aidokit sync --tier standard --dry-run    # preview the diff
aidokit sync --tier standard               # apply

sync is non-destructive: your CLAUDE.md edits, Beads database, and docs/ content are preserved. Only the engine directory (.claude/) gets the additional Standard-tier emission.

Honest limitations of Starter #

The flat-file Beads fallback is fine for one project but does not scale across repos. If you have more than one project, install Beads (per the prereq-check hint).
"Implicit roles" means the orchestrator makes role-like decisions internally; you cannot inspect them as separate files. Standard tier makes them explicit if that's important to you.
The watchdog is permissive at Starter — it warns on out-of-scope edits but does not block them. Standard tier blocks; Strict tier blocks plus records.