ideate
Take a fuzzy idea (or an existing thing you want to improve) through a disciplined funnel — explore → pressure-test → converge — and produce one living CONCEPT_BRIEF.md: a locked concept + phased roadmap. Acts as a blunt, honest co-founder (permission to overrule you), forces a success metric and a kill criterion before any roadmap, and hands off cleanly to the prompt-pack skill for building. ALWAYS invoke when the user says any of "help me figure out what to build", "I have an idea for…", "is this idea any good", "pressure-test this idea", "should I build / should I rebuild X", "evaluate whether my app/idea is worth building", "where are we really at", "is this idea or app worth pursuing", "iron out / scope / spec this concept", or "turn my idea into a plan/roadmap". Also invoke proactively when someone is reasoning about WHAT to build or WHETHER an idea is worth it — before any code or prompt pack exists.
nelsonwerd · AI review on file
Installation
Read the source and dependencies before running a command. Copying does not install or execute anything.
npx skills add nelsonwerd/idea-to-ship-skills --skill ideateAbout this skill
name: ideate description: >- Take a fuzzy idea (or an existing thing you want to improve) through a disciplined funnel — explore → pressure-test → converge — and produce one living CONCEPT_BRIEF.md: a locked concept + phased roadmap. Acts as a blunt, honest co-founder (permission to overrule you), forces a success metric and a kill criterion before any roadmap, and hands off cleanly to the prompt-pack skill for building. ALWAYS invoke when the user says any of "help me figure out what to build", "I have an idea for…", "is this idea any good", "pressure-test this idea", "should I build / should I rebuild X", "evaluate whether my app/idea is worth building", "where are we really at", "is this idea or app worth pursuing", "iron out / scope / spec this concept", or "turn my idea into a plan/roadmap". Also invoke proactively when someone is reasoning about WHAT to build or WHETHER an idea is worth it — before any code or prompt pack exists.
Ideate — fuzzy idea → locked concept + roadmap
ideate is the upstream partner to the prompt-pack skill. It takes an idea that's still fuzzy — or an existing project you want to evaluate and improve — and runs it through a disciplined funnel that ends in one durable artifact: a CONCEPT_BRIEF.md holding the locked concept, the honest verdict, and a phased roadmap. That brief is exactly what prompt-pack consumes to produce build prompts.
The whole pipeline:
ideate →
CONCEPT_BRIEF.md→ prompt-pack → execute
This skill encodes a method reverse-engineered from real, successful idea→ship journeys — including the mistakes those journeys made, so it improves on them rather than repeating them.
Why this exists (the problem it solves)
Left unstructured, ideation with an AI fails in predictable ways:
- It specs too early. A full data model and endpoints get written before anyone asks "is this even a good idea?" — so effort pours into an unvalidated concept.
- It forgets. The AI re-derives the same "is this a business?" assessment every session; the user re-pastes context; decisions never settle. (This is the single biggest failure.)
- It has no finish line. No success metric, no kill criterion — so the concept mutates forever and no one can say whether it's working.
- It over-commits. A giant roadmap gets locked before the first piece is validated.
- It drifts (when improving an existing thing) away from the original it was supposed to preserve.
ideate fixes each of these with a specific mechanism (see Pitfalls). Its center of gravity is one living brief, edited in place — never regenerated.
The prime directive (the ordering rule)
Never produce implementation detail, a giant roadmap, or a name before the concept has survived an honest pressure-test. The order is explore → pressure-test → converge → then roadmap/handoff. If the user asks to skip ahead ("just give me the build plan"), name the risk, lock the minimum needed, and offer to proceed — but do not silently spec an unvalidated concept.
This is non-negotiable and stated first because violating it is the most common, most expensive mistake.
When to use this
Strong triggers — invoke without asking:
- "Help me figure out what to build" / "I have an idea for…" / "give me a few ideas for…"
- "Is this idea any good?" / "pressure-test this" / "be honest about whether this is worth it"
- "Should I build X?" / "should I rebuild / refactor / improve X?"
- "Evaluate whether to keep building my app / repo / idea" / "where are we really at?" / "is this thing's direction still right?"
- "Iron out / scope / spec this concept" / "turn my idea into a plan"
Do NOT use this for:
- A settled concept that just needs building → go straight to
prompt-pack. - Pure factual research with no decision attached → use a research skill.
- A soundness/quality audit of code or a strategy with no "should I pursue this" question → that's the
deep-diveskill. (ideatejudges worth pursuing;deep-divejudges is it sound. For high-stakes validation,ideatedelegates todeep-divemid-funnel.) - Writing or executing build prompts → that's
prompt-pack.ideatestops at the brief. - Routing tie-breaker: ideate answers "is this worth building?" If the real question is "is this correct / safe / sound?" on something that already exists, that's a rigorous audit — hand to the
deep-diveskill if you have it (ideate already delegates heavy validation to it mid-funnel); if you don't, run a lighter honest pressure-test inline and label the gaps. If the concept is already settled and the user just wants the build sequenced, that'sprompt-pack. When it's genuinely unclear, ask one question: viability direction, rigorous audit, or execution-planning?
This is an interactive loop, not a one-shot
ideate is a multi-turn facilitation, not a single generated document. Run it as a real back-and-forth:
- On re-entry, load the existing brief first. Before anything else, check for an existing
CONCEPT_BRIEF.md(indocs/, else the repo root); if it exists, read it and continue editing that file. Never start a fresh brief over an existing one — that destroys the decision log this skill exists to preserve. - Stop-and-confirm at every gate. Don't run the whole funnel in one turn and dump a finished brief. Lock decisions one at a time, as the user reacts.
- Edit the brief every turn — then show it. After each meaningful exchange, actually write the new decision/finding into the brief (diff into it; don't regenerate the whole file). Then mirror the key change inline in chat so the user sees it without needing the file open — and never claim a brief update you didn't actually make.
- Ask, then wait. When you pose options or a pressure-test verdict, stop and let the user respond before converging.
The two modes
Detect the mode in the first exchange; it's the first branch. Full p
Read the complete source ↗