sglang-prod-incident-triage

Convert live SGLang incidents into replayable debugging workflows.

721|65|Updated Apr 1, 2026

Search Tags:#debugging #diagnostics #incident-management #serving #bundles #replay #sglang

Ready to run

npx skills add https://github.com/BBuf/AI-Infra-Auto-Driven-SKILLS --skill sglang-prod-incident-triage

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: sglang-prod-incident-triage
Source: https://github.com/BBuf/AI-Infra-Auto-Driven-SKILLS/tree/main/skills/sglang-prod-incident-triage
Command: npx skills add https://github.com/BBuf/AI-Infra-Auto-Driven-SKILLS --skill sglang-prod-incident-triage

SYSTEM DOCUMENTATION & REQUIREMENTS

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

What problem does it solve?

Replay-first debug workflow to investigate and stabilize SGLang serving problems by turning live incidents into repeatable, replayable debugging paths.

Core Features & Use Cases

Collect a healthy baseline bundle from a running server and summarize the state.
Save the failing request or crash/dump for replay on a clean target.
Replay on a controlled environment and branch into targeted debug tools (debug-cuda-crash, debug-distributed-hang, llm-torch-profiler-analysis) as the root cause becomes clearer.
Use references and scripts to automate bundle collection, dump summarization, and replay.

Quick Start

Start by collecting a healthy bundle, then capture the trigger and replay on a clean target to reproduce and diagnose the issue.

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 latency or throughput regressions?▼

Debug SGLang serving latency regressions by collecting a healthy baseline bundle, preserving the failing request, and replaying it on a controlled target to systematically diagnose the root cause.

What is a replay-based debugging workflow for LLM serving incidents?▼

Replay-based debugging converts live LLM serving incidents into repeatable workflows by capturing baseline states and failing triggers, allowing you to reproduce and analyze issues in a controlled environment.

How do I reproduce an SGLang crash or timeout after a deployment?▼

Reproduce an SGLang crash after deployment by saving the crash dump or failing request, then replaying it on a controlled target environment to isolate the trigger and branch into targeted diagnostic tools.

Can I automate diagnostic bundle collection for SGLang incidents?▼

Yes, you can automate bundle collection and dump summarization for SGLang incidents using the integrated scripts and references, which streamline the extraction of server states and failing requests.

Does SGLang incident triage require specific dependencies to function?▼

SGLang incident triage requires the requests dependency to automate bundle collection and interact with the running server, ensuring you can capture health states and trigger data for replay analysis.

When should I use replay-first debugging for LLM serving issues?▼

Use replay-first debugging for LLM serving issues when health checks fail, latency or throughput regress, or when you experience crashes, timeouts, and misbehavior after deploys needing a repeatable diagnostic path.