GitHub Issue-Triage KI Starter-Kit GitHub Issue Triage AI Starter Kit

Workflow-Spec: Label-Taxonomie · Duplikat-Erkennung · Confidence-Schwellen · Webhook-Handler · Eskalations-Regeln. Markdown + GitHub-Actions-YAML-Stub. Keine E-Mail, keine Anmeldung. Workflow spec: label taxonomy · duplicate detection · confidence thresholds · webhook handler · escalation rules. Markdown + GitHub Actions YAML stub. No email, no signup.

Verwandt: DraftCat-Guide · GitHub Issue Management AI Artikel Related: DraftCat guide · GitHub issue management AI article

This starter kit is the workflow spec for a Claude-powered issue-triage agent. Same shape as the GitHub issue management AI article and the DraftCat guide — what runs on my own repos, packaged so you can deploy it on yours.

Related: Production AI Agent Architecture, How to choose an LLM for production.

ErforderlichRequiredRepo-KontextRepo context

›
Unter 20: manuelle Triage reicht. Ab 50: dieses Kit zahlt sich aus.Below 20: manual triage is fine. Above 50: this kit pays off.

Label-TaxonomieLabel taxonomy

›
Bei Monorepo: ein Label pro Komponente erlaubt CODEOWNERS-Routing.In a monorepo: one label per component enables CODEOWNERS routing.

Duplikat-ErkennungDuplicate detection

›

Confidence-SchwellenConfidence thresholds

›

Webhook-EventsWebhook events

›

Eskalations-RegelnEscalation rules

›
Wenn ein Issue nach N Stunden noch needs-info oder unlabeled ist, eskalieren.If an issue is still needs-info or unlabeled after N hours, escalate.

Generierte SpecGenerated spec

Triage-Agent auf Ihren Repos in 3 WochenTriage agent on your repos in 3 weeks

Sie haben die Spec. Wenn Sie wollen, dass ich sie baue — GitHub-App + Claude-Backend + Webhook-Handler + Dashboard — ich scope in 24 Stunden und liefere in 3 Wochen. Human-in-the-Loop by default. You have the spec. If you want me to build it — GitHub App + Claude backend + webhook handler + dashboard — I scope in 24 hours and ship in 3 weeks. Human-in-the-loop by default.

Mein Konzept anfragen Request my scope
KopiertCopied
The context layer for your AI agents

Your agents answer from whatever the retriever finds, and too often that is last quarter's truth. I build the context layer they answer and act from: a temporal knowledge graph that keeps every fact with its source and the time it held, reads with each person's own permissions, and writes nothing without a person's approval. On your own tenant, billed by the hour, step by step.