Questions about adopting Cloverleaf

How is Cloverleaf different from Scrum?

Scrum assumes humans are the slow part of building software. Cloverleaf assumes AI is the primary author and humans are the slow part of judgment. The result is a different bottleneck and a different optimisation: Cloverleaf shifts human attention upstream into strategy and review, where the leverage now lives.

How is Cloverleaf different from Shape Up?

Shape Up keeps humans implementing during the cycle and uses appetites as a soft cap. Cloverleaf removes humans from implementation entirely (agents do that) and uses gates as hard caps on strategy and merge.

Can a single developer use Cloverleaf?

Yes. The 9 agents are recommended roles, not staffing. A solo dev can switch personas in a single Claude session and still get the structural benefits of fresh-eyes review and typed failure handling.

What about teams that aren't using AI yet?

Cloverleaf is specifically for teams where AI is the primary code author. If your team writes most code by hand, Scrum or Shape Up will fit better.

Is the methodology stable?

Standard v0.8.3 (this version). The architecture (two tracks, 4 work items, 9 agents) is settled. Some details may evolve based on real-team feedback.

Is there a reference implementation?

Yes — @cloverleaf/reference-impl on npm. Both tracks ship as Claude Code skills. Delivery runs through the orchestrator /cloverleaf-run: the Implementer works the Task, then a single delivery council convenes. The Reviewer always sits; the Security Reviewer joins when the Task's security classification is high, and a high-risk Task additionally seats QA and — when the diff touches UI — the UI Reviewer. Documentation is a step in the Task lifecycle rather than a council seat; the Documenter writes those commits when the Task is high-risk. Discovery runs through /cloverleaf-discover, which sequences the RFC, Spike, gate and breakdown skills from brief to materialised Tasks.

How do I cite Cloverleaf?

Cite the live guide at https://cloverleaf-org.github.io/cloverleaf/ and pin to the Standard version shown in the footer (currently v0.8.3). A DOI may follow.

When should I use a Plan vs go RFC → Task directly?

Plans are checkpoints. The task_batch_gate exists so a human reviews the decomposition before any task materialises — it's the load-bearing review point of the Discovery track. Use a Plan when the work spans three or more related tasks AND there's real value in reviewing the decomposition as a unit. The cost of forming a Plan pays for itself in the form of catching bad scope decisions early.

RFC-direct tasks (/cloverleaf-new-task --rfc=<RFC-ID>) skip the Plan and the gate. That's the right call for hotfixes (one task; nothing to review at the decomposition level) and for incremental RFC progress where the operator hasn't yet committed to a batch shape. The walker treats RFC-direct tasks as first-class — they participate in the RFC's auto-advance check alongside sibling Plans (see chapter 4 § "RFC-direct tasks"). Skip the Plan only when the checkpoint wouldn't earn its keep.