Filtre de tri des PR pour les mainteneurs open source débordés

Une étape CI qui note et filtre les PR à faible effort ou générées en masse par IA avant qu'elles ne consomment le temps de revue des mainteneurs open source.

I want to build a small, bootstrappable product. Help me plan and start the MVP.

Product: PR triage gate for overwhelmed OSS maintainers
What it is: A CI gate that scores and filters low-effort/AI-dumped PRs before they consume open-source maintainer review time.
Who pays: Maintainers and engineering leads at funded open-source projects and foundations (e.g., projects like Plotly, QGIS-scale, or corporate OSS offices) with an infrastructure/community budget, paying $200–800/mo for CI add-ons.
Closest alternatives: CodeRabbit, GitHub Copilot code review, Greptile, human triage / stale-bot automation
The gap: Existing AI reviewers review code quality; none act as a maintainer-side contribution gate that enforces a project's declared AI policy, flags low-effort/unverifiable submissions, and requires submitter-side proof (repro, test, video walkthrough) before any human time is spent.
Why now: The LLM contribution flood is a 2025–26 phenomenon: major projects (QGIS now, others considering) are enacting blanket bans as a stopgap precisely because no tooling exists to ration reviewer attention, and maintainers publicly say they'd rather keep contributing open if filtering were cheap.
Moat to build toward: Policy-as-code rule sets tuned per project plus accumulating labeled data on low-effort vs. genuine contribution patterns across many repos; workflow lock-in once embedded in a project's merge pipeline.

Un brief prêt à coller dans Claude Code, Cursor ou tout agent de code (en anglais).