sglang-prod-incident-triage

Name: sglang-prod-incident-triage
Availability: InStock
Author: annealing-inversion

Replay live SGLang serving faults into reproducible debug paths.

Updated Jun 4, 2026

Search Tags:#workflow #replay #production-ops #sglang #incident-triage #serving-debug

Ready to run

npx skills add https://github.com/annealing-inversion/sglang-kimi-deferral --skill sglang-prod-incident-triage-annealing-inversion

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: sglang-prod-incident-triage
Source: https://github.com/annealing-inversion/sglang-kimi-deferral/tree/main/.claude/skills/sglang-prod-incident-triage
Command: npx skills add https://github.com/annealing-inversion/sglang-kimi-deferral --skill sglang-prod-incident-triage-annealing-inversion

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires requests, and includes scripts (resource) and references (resource) components.

What problem does it solve?

Replay-first debug flow to convert live SGLang serving problems into reproducible debugging pathways, enabling rapid diagnosis and containment.

Core Features & Use Cases

Replay-first incident triage workflow: collect a baseline bundle, capture failing request or crash dump, replay on a clean target, and progress tool-switching as the issue narrows.
Examples and guidance to drive decisions about which focused tools to use next.
Works with /health, /health_generate, /v1/loads, /metrics, and crash dumps.

Quick Start

Collect a baseline bundle from the live server, save the failing request or crash dump, and replay on a clean target before applying deeper diagnostics.

Frequently Asked Questions about sglang-prod-incident-triage

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I debug SGLang serving issues like queue growth and latency regressions?▼

Debug SGLang serving issues by collecting a baseline bundle, saving the failing request or crash dump, replaying it on a clean target, and switching to focused diagnostic tools as the problem narrows.

What is replay-first incident triage for production LLM serving?▼

Replay-first incident triage converts live serving faults into reproducible debug paths. It enforces a structured workflow: capture a baseline bundle, save the failing request, replay on a clean target, and apply deeper diagnostics.

Which SGLang endpoints and artifacts do I need to collect for an incident baseline?▼

Collect baseline data from SGLang endpoints including /health, /health_generate, /v1/loads, /metrics, along with any crash dumps, to establish a baseline bundle for reproducing the serving issue.

Does this incident triage workflow work with distributed serving stalls and timeouts?▼

Yes, the incident triage workflow applies to distributed stalls, timeouts, throughput regressions, and crash dumps. It guides you from baseline collection to replay and focused tool-switching to isolate the root cause.

What's the best way to reproduce a live SGLang crash dump for debugging?▼

The best way to reproduce an SGLang crash dump is to save the failing request from the live server, isolate it, and replay the request on a clean target environment before applying deeper diagnostic tools.

Why does my SGLang production server experience health-check failures after a deployment?▼

Health-check failures require a structured triage workflow. Collect a baseline bundle from the live server, capture the failing request, replay it on a clean target, and switch to appropriate debugging tools to diagnose the regression.