Pre-Build Idea Validator

The most common failure mode for side projects: you build for a week, then discover three other projects already do exactly your thing, better. A 5-minute reality check up front would have saved the week. The validator sweeps real data sources and gives you a "how crowded is this?" signal before you commit.

What it does

Scans 5 sources: GitHub, npm, PyPI, Hacker News, Product Hunt
Returns a reality_signal score (0-100) — 0 = greenfield, 100 = saturated
Lists the top 5 most relevant existing projects with star counts / download counts / launch dates
Recommends a strategic angle: differentiate vs ignore vs fork-and-extend
Saves the scan to memory so you can re-run quarterly

What you'll need

Web fetch with public API access (no keys needed for any of the 5 sources)
Memory for scan history
A clear, specific idea description (not "I want to build an AI thing")

Setup

1. Define the validator workflow

Send to Flowly

When I say "validate idea: <description>": 1. Extract 3-5 keyword combinations from the description (the distinctive technical terms, not "AI" or "tool") 2. For each source, search: GITHUB: https://api.github.com/search/repositories?q=<keywords>+language:python+language:typescript NPM: https://registry.npmjs.org/-/v1/search?text=<keywords>&size=10 PYPI: https://pypi.org/simple/ search via Google site:pypi.org or /pypi/ JSON API HN: https://hn.algolia.com/api/v1/search?query=<keywords>&tags=story PRODUCT HUNT: https://www.producthunt.com/?q=<keywords> (parse HTML) 3. For each result, extract: - Name, description, URL - Stars / downloads / upvotes (whichever metric the source has) - Last activity (commit, version, comment) 4. Compute reality_signal: - Top GitHub repo > 1k stars: +20 - Top npm/pypi package > 50k downloads/month: +15 - HN front-page story in last 12 months: +15 - Product Hunt top of category in last 6 months: +20 - 3+ active maintainers across results: +10 - All else: small additive contributions (max 100) 5. Output: - reality_signal score with band ("greenfield 0-20", "competitive 20-50", "crowded 50-80", "saturated 80-100") - Top 5 most relevant projects (table) - Recommendation: * 0-20: build it * 20-50: build it but find a clear differentiator first * 50-80: study the leaders, decide if you can do something they don't * 80-100: don't build, contribute or fork 6. Save to memory tagged "idea-validation:<slug>:<date>".

2. Try it

Send to Flowly

validate idea: a CLI that takes a YouTube URL, downloads the audio, transcribes via local whisper, and saves a markdown summary

Within ~30 seconds you get back:

reality_signal: 65 (crowded)

Top hits:

whisper-cli (npm, 12k downloads/wk) — transcription

youtube-dl (PyPI, defacto for download)

summarize-yt (npm, 800 stars) — closest match: download + transcribe + summarise

<2 more>

Recommendation: "summarize-yt" already does this. Either contribute to it (add markdown export?) or differentiate strongly (specific language? batch processing? integration into your editor?). Building it solo from scratch is wasted effort.

3. Re-run quarterly

Send to Flowly

Cron "idea-rescan" monthly: For ideas tagged "idea-validation:*" with status "exploring", re-run the validator. New results since last scan get flagged. This catches: someone shipped what you were planning, OR your idea became more relevant as the space grew.

Tips

Specificity beats generality. "AI agent framework" returns 1000 results, none useful. "Plugin system for personal AI agents with in-process Python execution" returns the actual competitors.
Don't only count stars. Old projects with 10k stars but no activity in 2 years are not real competition; they're abandoned. Filter on last-activity.
Re-running matters. The space shifts. An idea that was crowded 6 months ago might have a clear gap now if the leader stagnated.
Don't trust the score absolutely. It's a signal, not a verdict. A reality_signal of 70 might still be worth building if the leaders are bad and you can do better.
The "fork and extend" path is underrated. When the closest match is open source and not actively developed, forking it gets you 80% of the work for free.