Phase 1: Ideation

Goal: one concept the author is excited about, that a real reader would pay for, that this author can write, and that no existing book already delivers. Everything downstream inherits this decision.

Read PRD.md, STYLE.md, sources/INDEX.md, and ${CLAUDE_PLUGIN_ROOT}/skills/book-{genre}/SKILL.md.

1. Premise stress test

Before generating alternatives, interrogate what the author already has. Answer in writing:

• Why this book? What does it do that the reader cannot get elsewhere in an afternoon of searching?
• Why now? What changed in the world, the field, or the author's life that makes this timely? If nothing, that is fine, but say so.
• Why this author? What access, experience, or angle do they have? (For fiction: what obsession or wound is underneath the premise?)
• What is the reader's objection? The strongest reason a target reader would put it back on the shelf.
• Where is the book most likely to fail? Middle sag, a premise that runs out at 30%, research the author does not have, a voice that will tire.

2. Comparable titles

Research 5-8 real books the target reader has read or would consider instead. Use web search when available; otherwise rely on known titles and mark each as verified or from memory. Never invent a title, author, or sales figure. For each comp: title, author, year, what it does well, what it leaves undone, how this book differs. Write the matrix into ideation.md.

For Korean-market books, include domestic comps (교보/예스24/알라딘 bestseller lists, 밀리의 서재 trends) and translated comps separately.

3. Concept generation

Generate 5-10 concepts. Concepts are angles on the premise, not random alternatives. Each:

• Elevator pitch (two sentences, no adjectives doing the work)
• Reader promise (what they have at the end)
• The first chapter in one paragraph (this exposes weak concepts fast)
• Differentiation against the comps
• What the author would need to research or invent
• Risk (one line)

Score each on: clarity, reader value, differentiation, feasibility for this author, timeliness, and fit with the voice fingerprint. Rank. Recommend one and explain why in three sentences.

4. Voice samples (if STYLE.md has no voice sample)

Write three 200-word openings of the recommended concept in three distinct voices consistent with the fingerprint's constraints. Label the choices in each (distance, sentence rhythm, diction). The author's pick is recorded in STYLE.md under ## Voice Sample and refines the fingerprint.

5. Author checkpoint

Present the ranked concepts, the recommendation, and the voice samples. Ask the author to choose (or merge up to three). This is a required stop. Do not proceed to outlining in the same turn without the author's choice.

Output

ideation.md: Premise stress test, Comparable titles (verified/unverified flagged), Concepts (ranked, scored), Chosen concept with final reader promise and working title, Voice sample decision, Open risks to carry into outlining.

Update PRD.md with the chosen concept, working title, and reader promise.

Gate

ideation.md exists with a ## Chosen Concept section approved by the author. Comps matrix has at least five entries with verification status marked. Reader promise is one sentence.

Post-completion

node ${CLAUDE_PLUGIN_ROOT}/velith.mjs scan [project-dir]