Incident Investigator — Data & AI agent for Claude Code
Investigates an LLM-serving incident (vLLM or SGLang) in an isolated context.
How to install Incident Investigator
Installs to ~/.claude/agents/lenialno-roofline-incident-investigator.md
mkdir -p ~/.claude/agents && curl -fsSL https://raw.githubusercontent.com/lenialno/roofline/HEAD/agents/incident-investigator.md -o ~/.claude/agents/lenialno-roofline-incident-investigator.md Restart Claude Code, or start a new session, for it to be picked up.
What Incident Investigator does
name: incident-investigator description: "Investigates an LLM-serving incident (vLLM or SGLang) in an isolated context. Reads long runtime logs, runs the toolbox health, metrics and gpu tasks, checks competing hypotheses and returns a short verdict with the root cause and the next action. Use for long logs, incidents spanning hours, or several hypotheses, so the main conversation receives only the summary. Also used for Russian requests about инцидент, OOM, очередь, p95." tools: Bash, Read,
Alternatives in Data & AI
- Lemonade Specialist — Lemonade Server and SDK specialist for local LLM on AMD hardware 1.5k ★
- Rlm Subcall — Acts as the RLM sub-LLM (llm_query) 382 ★
- Part 11: Auto-Capture Hook (Stop Manually Writing Memory) — By Terp - Terp AI Labs --- The vault system from Part 9 is powerful, but there's a gap: you have to manually w 232 ★
Full documentation available on GitHub
View Source RepositoryRelated Agents
Context Isolator
Runs data-heavy read-and-summarize operations in an isolated context window to prevent bloating the main conve
Python ML Expert
Senior ML engineer. Use proactively for classical ML work — training pipelines, model evaluation, feature engi
Data Investigator
Newsletter Competitive Analyst
Use this agent when you need to analyze competing AI newsletters to identify content gaps, differentiation opp
Bead Recovery Specialist
Use this agent for bd/Dolt incident response — a dolt-server that won't start or has orphaned, server sprawl,
Jev Analyst
Deep read-only analyst for Jev-routed subtasks. Investigates and explains behavior that spans many files (end-
Related Skills
Inference Serving
ai-research-skills vLLM, SGLang, TensorRT-LLM, llama.cpp
ML Production
Use when shipping ML models (classical or LLM) to production - quantization (torch.ao.quantization, AWQ/GPTQ/A
Universal Debug Mode Skill
Evidence-driven debugging skill for AI agents: generate competing hypotheses, instrument with isolated runtime
