cs-build-plan-app
Plan an app from an idea
There is no code yet. Produce the same artefact cs-build-import-codebase produces — core features → sub-features → notes → PRDs → tasks — but derived from a person instead of a repo.
Two phases with a gate between them:
| A. Think (§1–§6) | Conversational, nothing is written | the job, the journey, every kind of user, the platform call, the hard part and what solves it, the systems checks, the v1 cut |
| GATE | Read the plan back and get an explicit yes | §6a |
| B. Write (§7) | CodeSpring gets touched | the map and the notes |
| C. Show (§7a) | Hand off to cs-build-ui-mockup |
the cheapest possible disagreement — corrections come back into the notes |
| D. Instruct (§8–§9) | Only once the mockup is agreed | PRDs, then tasks |
Never write to CodeSpring during phase A. A map is expensive to unpick and it makes people feel committed to a plan they were still arguing with. Phase A is meant to be argued with.
This is the no-code-yet counterpart to cs-build-import-codebase and it deliberately does its own writing rather than handing over — that skill's engine is reading a repo, and there is no repo. What the two share is the destination, and that is already factored out: the node model and map rules in cs-build-import-codebase/references/target-map.md, the schema in mindmap-structure.md, the CLI in commands.md. Use those; do not re-derive them and do not fork them.
Shared knowledge, read it: codespring skill references/project-state.md (enter-from-anywhere, the three stages), references/mindmap-structure.md (the exact node/edge schema), references/prd-management.md, references/pitfalls.md. And cs-build-import-codebase/references/target-map.md — the node model, the one-sentence-card rule and the 4–8 core feature ceiling are identical here; do not restate or re-derive them.