Idea Generator

Idea Generator

Produce a big, well-structured batch of ideas fast — in the same format as the existing idea docs, so they slot right into the pipeline. This skill diverges (creates options); /idea-pipeline converges (grills, researches, ranks). Generate here, then hand off.

Phase 0 — Mode

• Fresh batch: a domain/seed was given (or ask for one).
• --expand <theme-slug>: read that theme's README in 00-inbox/… (or any doc) and generate more like these / fill the gaps it doesn't cover. Read existing inbox themes first to avoid duplicating ideas already captured.
• --to-inbox: after generating, write the batch into the ideate workspace as a new theme folder (scripts/new.sh or direct), not just chat.

Phase 1 — Seed me (short interview)

Use AskUserQuestion, one batch. Skip anything already given as args / inferable from a doc.

1. Domain / market — where to hunt (e.g. local SMB marketing, consumer mobile, dev tools, hockey, fintech). Broad is fine; the frameworks will fan out.
2. Your angle (multi-select) — what should bias generation: unfair advantage / existing audience · skills you already have · a customer you can reach · a workflow you live daily · ride a trend (AI made X 10× cheaper) · pure market size.
3. Constraints — solo vs team · bootstrap cash-machine vs venture-scale (or both) · build-time ceiling (weekend / weeks / months) · monetization (subscription / marketplace take-rate / usage / one-time) · hard avoids (no mobile, no ML, no cold sales, etc.).
4. Count & spread — how many (default 40), and how many themes to bucket into (default 4–6).

Echo a 3-line generation brief; proceed (don't wait for a thumbs-up unless they want one).

Phase 2 — Generate (use frameworks, don't freestyle)

Spread ideas across these lenses so the batch is diverse, not 40 variations of one:

• Pain mining — concrete recurring frustrations for the target user; what they do today.
• Jobs-to-be-done — the job they "hire" a product for; underserved steps.
• "X for Y" — proven product transplanted to a new audience/vertical (e.g. "Ahrefs for local").
• Unbundle an incumbent — take one feature a bloated/expensive tool does and do it 10× better/cheaper.
• Re-bundle / aggregate — combine fragmented point-tools into one sticky system.
• Trend-ride — capability that just got cheap (LLMs, cheap transcription, vision) → newly viable.
• Manual → automatic — a task people do by hand or hire out; automate it.
• Expensive → affordable — enterprise-only category, build the dead-simple $29/mo SMB tier.
• Audience-first — start from a reachable audience/advantage, work back to what to sell them.

For each idea, write this exact block (matches the existing idea docs):

### <Idea Name>
- **Problem:** who hurts + the specific recurring pain.
- **MVP:** the smallest thing that delivers the core value.
- **Missing features:** what nobody does well today (the wedge).
- **Subscription potential / model:** $X–$Y/mo, or take-rate, or usage.
- **Difficulty:** Easy / Medium / Hard (MVP vs full product if they differ).
- **Notes:** the one sharp insight, risk, or positioning angle.

Group ideas under ## Theme: <name> headers. Tag each idea internally as bootstrap cash-machine or venture-scale if the user asked for that split.

Aim for the requested count; if a lens runs dry, lean on another rather than padding clones.

Phase 3 — Light pre-screen (NOT validation)

Before presenting, cheaply prune the obviously-dead so you don't raise false hope:

• Dedup against ideas already in the inbox (read existing theme READMEs) and within the batch.
• Saturation flag — if a category is obviously locked (a dominant incumbent does exactly this for this exact user), mark it ⚠ crowded rather than dropping silently.
• Optional quick name sanity: if a working title is clearly a known app, note it (full naming is /brand-name-research's job).

Do not do deep market research or scoring here — that's /idea-pipeline.

Phase 4 — Deliver

• Present the themed batch in the block format above.
• Add a 1-line "start here" call-out: the 3 ideas that best fit the brief (advantage × feasibility × pull), and why.
• If --to-inbox (or the user says yes): create a theme folder per ## Theme in 00-inbox/ (frontmatter kind: theme, per-idea checklist + the blocks verbatim), then run scripts/board.sh. Mirror the structure of existing imported themes.
• Handoff: "Run /idea-pipeline --rank to grill, research, and rank these; then /brand-name-research on a winner."

Notes

• This skill is the divergent front door; keep it generative and fast. Resist the urge to evaluate — capturing a bad idea is cheap, killing it later is the pipeline's job.
• Re-runnable: call again with a different angle/lens to get a fresh, non-overlapping batch (it reads what's already in the inbox to avoid repeats).