Brainstorm

Generate ideas that are actually worth having, then leave behind a vault that shape can turn into a plan. Read .claude/skills/shared/vault-contract.md before writing anything to .idea/.

Plain language. Everything you say to the person — questions, plans, summaries, reports — follows .claude/skills/shared/plain-language.md: assume they are not a developer, use simple words, recommend an option when you ask, and explain any technical word the first time it appears. Notes in .idea/ may stay technical; explain them, don't paste them.

Phase 1 — Ground yourself

1. ls -R .idea/ 2>/dev/null. If it exists: read INDEX.md, every themes/*.md, and any notes/*.md whose title touches the topic. Do not re-propose an idea already in the vault — build on it, sharpen it, or argue against it, and say which note you are extending.
2. Read SPECS.md and ROADMAP.md if present. Ideas that contradict a committed phase are still welcome, but you must flag the contradiction explicitly.
3. If the repo has code, skim README, the package manifest, and the main entry points. Ideas ungrounded in what exists are cheap and mostly useless.

Phase 2 — Predict the intent before generating

This is the step that makes the rest accurate. Before any ideas, state in 4 short bullets:

• The real goal — what the user is actually trying to achieve, beneath the literal ask.
• Success in 6 months — what would have to be true for this to have been worth doing.
• The constraint that binds — time, money, skill, distribution, or an existing commitment.
• What this is probably not — the adjacent thing you are deliberately not going to explore.

Then say: "Correct me on any of these and I'll re-aim." Do not wait for a reply — proceed on your stated reading, and revise if corrected. Record these as the Intent ledger in the session note; shape reads it to write the SPECS rationale.

Phase 3 — Diverge

Load references/divergence.md and work through the axes there. Non-negotiables:

• 8-12 distinct ideas before evaluating any of them. Quantity first, judgment later.
• Every axis in the reference gets at least one idea, including the ones that are uncomfortable: the contrarian, the 10x-simpler, the premise-questioning, and the horizon bet — what becomes possible or obsolete in 12-24 months as costs, models, and platforms shift.
• One line of pitch plus one line of why-it-might-work, each. No essays at this stage.
• Never suppress an idea for being hard. Mark it #stretch and keep it.

Phase 4 — Group, break down, converge

1. Cluster the ideas into 3-6 themes by the problem they solve, not by technology. A theme with one member is a signal you mis-clustered or found a genuine outlier — say which.
2. Break down anything that is really a container: if an idea implies three separable pieces of work, it is a theme, and its pieces are the notes.
3. Score each idea impact / effort / confidence 1-5 and assign a horizon. The numbers are a sort order and a conversation starter, not a verdict — state your reasoning in one line.
4. Recommend the 2-3 strongest with actual reasoning, name the riskiest assumption in each, and list the open questions.

Phase 5 — Discuss

Ask for reactions and keep iterating as long as the user wants. Ideas the user contributes are captured the same way and attributed to them in the session note. New ideas mid-conversation go through Phase 4 too — do not let late ideas skip grouping and scoring.

Phase 6 — Capture

Checkpoint after every round; do not save only at the end of a long session.

Write, per references/note-templates.md:

• .idea/notes/<Idea>.md — one per idea, atomic, scored, linked to its theme.
• .idea/themes/<Theme>.md — the grouping layer, with the theme's thesis and its member notes.
• .idea/sessions/<YYYY-MM-DD> Brainstorm - <Topic>.md — intent ledger, every idea produced as wikilinks, decisions reached, and next steps as - [ ] tasks.
• .idea/INDEX.md — regenerated, never appended to.

Get the date from date +%Y-%m-%d.

Handoff

Close by telling the user what changed on disk, then: "Run /shape when you want these turned into a spec and a phased roadmap." If 5+ notes now sit at status: promising, say so — that is the signal that shaping is worth doing.