greenfield

superkit: greenfield

Build a product from nothing. Run a dependency check first (bash "${CLAUDE_PLUGIN_ROOT}/scripts/check-deps.sh"); if it fails, route to /superkit:setup.

Phase 1 — Define the product

1. Invoke speckit-constitution to set governing principles (once).
2. With the user, author .superkit/product.md: the overall what/why plus a feature backlog — an ordered list of features to spec, each with a short id, title, and any dependencies. Use AskUserQuestion to decompose if the product is large.
3. For each backlog item, register a placeholder: bash "${CLAUDE_PLUGIN_ROOT}/scripts/registry.sh" add <id> "<title>" <depends-on>

Phase 2 — Spec everything (loop over the backlog)

For each backlog item, in dependency order:

1. Note the current base branch: git rev-parse --abbrev-ref HEAD.
2. Invoke speckit-specify (creates specs/<NNN>-name/ and branch <NNN>-name). Record spec-path + branch: registry.sh set <id> spec-path <path> and registry.sh set <id> branch <branch>; registry.sh set-status <id> specified.
3. Invoke speckit-clarify → set-status clarified.
4. Invoke speckit-plan → set-status planned.
5. Invoke speckit-tasks → set-status tasked. Gate: specs/<id>/tasks.md non-empty.
6. Branch hygiene: return to the base branch (git checkout <base>) before the next item, so specs don't drift across branches.

When the backlog is fully specified, run /superkit:status to show the batch.

Phase 3 — Execute the batch

Ask the user (AskUserQuestion) how to execute:

• Continuous — one worktree, all specs' tasks back-to-back, one review + one finish.
• Sequential, isolated — per spec: worktree → TDD → review → finish, then next.
• Parallel where independent — use the registry's depends-on to build a graph; run independent specs via superpowers:dispatching-parallel-agents, serialize dependents.

Then drive superpowers accordingly (using-git-worktrees, test-driven-development, subagent-driven-development, superkit:iterate-to-green, superkit:conformance-check, requesting-code-review, finishing-a-development-branch), updating each spec's status (executing → in-review → done) as you go. Run superkit:iterate-to-green per strategy: continuous = one whole-suite pass after all specs; sequential = one pass per spec before its review/finish; parallel = one pass per agent, plus a final whole-suite pass after merges to catch cross-spec regressions.

Contract (enforced throughout)

Do NOT re-plan — each specs/<id>/tasks.md is the contract. Pass each subagent ONLY its task. Never hand a subagent the whole .superkit/ or specs/ tree.