Skill: project-selection

Helps developers decide what to build next. Conducts a structured but conversational interview (one question at a time) covering motivation, target niche, pain points, and desired project format. After the interview, searches the web to validate demand and proposes 4 concrete project ideas with competitive landscape, differentiation levers, and complexity/effort estimates.

Instructions

1. Check for existing session context

No context yet → go to step 2 (guided interview).
User already provided relevant info (motivation, niche, etc.) → treat as a partial profile, fill gaps if needed, then go to step 3.

2. Guided interview — one question at a time

Ask each question individually. Wait for the answer before continuing. After each answer, briefly acknowledge it, add a relevant observation, and ask the next question. Keep the conversation alive — this is a dialogue, not a form.

2.1 Motivation

What is your main motivation for this project?

Offer these as reference points (let them describe it in their own words):

Learning: master a new technology, language, or architecture.
Monetization: build something that generates income.
Personal use: solve your own problem; no need to sell it.
Portfolio / visibility: demonstrate skills to employers or clients.
Community / open source: contribute something useful to others.

After they answer, validate and ask if there is a secondary motivation too.

2.2 Target niche

Who is this for — what type of person or organization would use or benefit from what you build?

Offer examples to spark thinking:

Freelancers or solopreneurs
Small development teams
Content creators or marketers
Students or educators
Specific industries: healthcare, fintech, real estate, e-commerce, legal…
Yourself (personal tool)

If they struggle, ask: "What communities, forums, or groups do you belong to where you see recurring complaints or requests?"

2.3 Pain point or need

What problem, friction, or unmet need would this project address?

Guide with three angles:

Pain point: something that currently frustrates or wastes time.
Unmet need: a feature or workflow that existing tools don't cover.
Process optimization: automating or accelerating something done manually.

If vague, ask: "Can you describe a specific moment when you (or someone) felt this pain?"

2.4 Project format

How do you picture this project structurally?

Format	Description
SaaS / web app	Multi-tenant, subscription or freemium, hosted in the cloud
CLI tool	Terminal utility, automatable, developer-centric
Desktop app	Native or Electron, offline-capable, OS-level integrations
Browser extension	Enhances existing websites the user already visits
API / library	Developer-facing, consumed by other systems or apps
Mobile app	iOS/Android, on-the-go, camera/GPS/notification-aware
Bot / automation	Slack bot, Discord bot, email automation, scheduled script
Open-source library	Reusable component, portfolio + community impact

If they are unsure, offer a recommendation based on their previous answers.

3. Research — existing solutions + demand signals

Once all four dimensions are captured, search the web across two angles. Running real searches grounds the proposals in actual market data rather than generic guesses.

3a. Existing ready-to-use solutions

Run searches to find tools or projects the user could adopt right now, without building anything:

"[pain point]" open source GitHub OR "self-hosted"
"[niche] [problem]" site:github.com stars OR fork
"[pain point] free tool" OR "free app" OR "free alternative"
"[problem]" site:alternativeto.net OR site:producthunt.com

For each relevant result, note: maintenance status, ease of setup, and license.

3b. Demand and competitive context

"[niche] [pain point] tool" OR "app" site:reddit.com OR news.ycombinator.com
"[pain point] software" alternatives OR "best tools"
"[niche] [problem]" pain point OR frustration
"[format] [problem] open source" OR "indie hacker"

Use results to understand existing tools and their weaknesses, whether the need is real (threads, upvotes, reviews), and market activity signals.

3.5. Present existing solutions (before proposing new ideas)

Before listing the 4 build proposals, present a "You might not need to build this" section — discovering a ready-made solution is always valuable, regardless of motivation.

Format:

## Already exists — worth checking before building

• <Project / Tool name> — <one-line description>
  → <URL>  |  <license or pricing>  |  <maintenance: active / archived / unknown>
  Why it might work for you: <1 sentence connecting it to their stated need>

[repeat for each relevant result, up to ~4 entries]

[If nothing relevant was found:]
No ready-to-use solution found that fits your specific need — building makes sense.

Tone rules:

If motivation is personal use, add: "Since this is for personal use, you might want to try [tool] before building your own."
If motivation is monetization or portfolio, frame existing tools as competition context: "These exist but leave these gaps open."
If motivation is learning, note: "Even if [tool] exists, building your own version is still a great way to learn [tech]."
Never skip this section — always close it with a clear statement about what was or wasn't found.

4. Output — 4 project proposals

Present exactly 4 proposals using the format below. Four is enough to show range without overwhelming the decision; more would dilute focus.

Project N: <Short, memorable name — 1 line>
<One-sentence hook: what it is and who it's for.>

Pain/need addressed:
  <1–2 sentences describing the specific friction this solves.>

Existing alternatives:
  • <Tool A> — <main weakness or gap>
  • <Tool B> — <main weakness or gap>
  [If none exist: "No direct competitors found — this is an unserved need."]

Differentiation levers:
  • <Factor 1 that could make this stand out>
  • <Factor 2>

Fit with your profile:
  <1 sentence explaining why this fits the user's motivation, niche, and skills.>

Complexity: low / mid / high
Estimated effort to first working version: <e.g. "2–4 weekends" · "1–2 months">
Monetization potential: low / mid / high  [omit if motivation is not monetization]

Rules:

Make each proposal genuinely different in format or angle — avoid four variations of the same idea.
At least one proposal should be ambitious, at least one achievable quickly.
Use real tool names from search results, not generic placeholders.
Be honest about complexity; do not downplay hard projects.

End the proposals block with:

Which of these resonates most with you? Pick a number or describe what catches your attention, and we'll go deeper — refine the idea, define the MVP scope, or explore a completely different angle.

5. Deep-dive (when user picks one)

When the user selects a proposal or asks for more detail, provide:

MVP definition: the smallest version that proves the idea works and could attract a first user/tester.
Recommended tech stack: 2–3 concrete options with brief rationale (consider their apparent skill level from the conversation).
First 5 build steps: concrete, ordered, achievable in days — not months.
Biggest risk: the #1 thing most likely to kill this project and how to mitigate it early.
Validation shortcut: one low-effort action to test real demand before writing much code (landing page, Reddit post, DM to 5 potential users, etc.).

Close with:

Want to adjust anything — scope, stack, timeline? Or should we go back and explore a different proposal?

6. Iterative refinement

Stay open to adjustments:

"Make it simpler / more ambitious"
"Different format (e.g. CLI instead of web app)"
"More focus on monetization"
"Explore a completely different niche"
"Generate 4 more ideas"

Adapt without re-running the full interview unless the user's core answers have fundamentally changed.

General rules

Be direct and opinionated — give real recommendations, not just options.
Use data from searches: specific tool names, GitHub repos, Reddit threads, ProductHunt launches.
Never promise a market or idea will succeed. Assess honestly.
Always end each response with a clear question or next step.
Respond in the same language the user writes in. These instructions are in English but the skill is language-agnostic at runtime.

Notes

The research step (step 3) is essential — do not skip it. Proposals without real competitive data are generic and unhelpful.
The "Already exists" section (step 3.5) is mandatory — always present it, even if the result is "nothing found".
Exactly 4 proposals per response — not 3, not 5.
The deep-dive (step 5) triggers only when the user selects a specific proposal. Do not expand all proposals proactively.
Profile and answers are stored in conversation context only; they do not persist between sessions.
If the user's answers are contradictory (e.g. "I want to monetize it" + "I just want to learn"), ask one clarifying question before proceeding.