IssueScout – Find open source issues worth contributing to

Name: IssueScout – Find open source issues worth contributing to
Availability: InStock
Author: axrisi

by axrisi·Feb 28, 2026·1 point·0 comments

Visit Project View on HN

AI Analysis

●●SolidSolve My ProblemNiche Gem

Health score + AI difficulty tiers solve real friction in GitHub's 'good first issue' search.

Strengths

•Per-user GitHub OAuth tokens scale linearly (no single bottleneck)—clever infrastructure choice
•Two-tier difficulty system (rule-based → GPT-4o fallback) balances speed and accuracy
•7-factor health scoring (CONTRIBUTING.md, CoC, response time, merge rate) is concrete and auditable

Weaknesses

•Caching strategy is standard (24h difficulty, 48h health); no novel insights on stale-while-revalidate
•Market already has Code Triage, First Timers, and GitHub's own 'good first issue' label—differentiation is UX-level, not structural

Post Description

I built IssueScout to solve a problem I had as a beginner looking to contribute to open source: GitHub has hundreds of thousands of "good first issue" labeled issues, but there's no way to know if the repo behind one is actively maintained or if the issue is actually beginner-friendly.

IssueScout adds two things on top of GitHub's search:

1. A Community Health Score (0-100) per repository — computed from 7 factors: CONTRIBUTING.md, license, code of conduct, recent activity, star count, issue response time, and PR merge rate. A score of 80+ means someone will actually review your PR.

2. AI Difficulty Estimation — a rule-based keyword analyzer runs first. If confidence is below 80%, it falls back to GPT-4o-mini. A purple sparkle shows when AI was used.

Architecture choices that might be interesting to HN: - Each user's GitHub OAuth token powers their own API requests (5K/hr per user) instead of a single server PAT. Scales linearly with users. - Two-level caching: issue difficulty cached 24h, repo health cached permanently with stale-while-revalidate at 48h. The IndexedRepo collection grows over time into a shared database of scored repos. - Two-phase progressive loading: raw GitHub results return instantly, enrichment fills in asynchronously. No spinners. - Rule-based AI first, LLM fallback only when needed. Keeps costs near zero for most queries. Stack: Next.js 16, TypeScript, MongoDB Atlas, GitHub GraphQL API, OpenAI GPT-4o-mini, Vercel.

Live at https://issuescout.dev — sign in with GitHub to try it.

MIT licensed. Full architecture docs: https://github.com/turazashvili/issuescout.dev/blob/main/doc...