Claude Code Observability Checklist — Quick Reference

One-page quick reference for setting up Langfuse observability on Claude Code. For the full walkthrough with code examples, read the complete Build Log.

---

Prerequisites

• Docker Desktop (or Docker Engine + Compose) installed

• Python 3.11+ (3.12 recommended)

• Claude Code CLI installed and working

• ~4-6GB RAM available for the Langfuse stack

Setup (5 commands)

git clone https://github.com/doneyli/claude-code-langfuse-template.git ~/langfuse-local
cd ~/langfuse-local
./scripts/generate-env.sh
docker compose up -d
./scripts/install-hook.sh

Verify

• docker compose ps — all 6 services show Up (healthy)

• curl -s http://localhost:3050/api/public/health — returns {"status": "OK"}

• Open http://localhost:3050. — Langfuse UI loads

• Change the default password on first login

Services Running

Daily Operations

# Start (data persists in Docker volumes)
cd ~/langfuse-local && docker compose up -d

# Stop (when done coding)
cd ~/langfuse-local && docker compose down

# Check if traces are being captured
tail -5 ~/.claude/state/langfuse_hook.log

# Check offline queue (traces captured while Langfuse was down)
wc -l ~/.claude/state/pending_traces.jsonl 2>/dev/null

# Update Langfuse
cd ~/langfuse-local && docker compose pull && docker compose up -d

Opt Out a Specific Project

Add to the project’s .claude/settings.local.json:

{
  "env": {
    "TRACE_TO_LANGFUSE": "false"
  }
}

What You Get

• Full prompt + response capture (no truncation)

• Tool call tracing (Read, Write, Edit, Bash, Grep, Glob)

• Session grouping across conversation turns

• Cross-project unified dashboard

• Offline queue with automatic drain on reconnect

• Queryable via Web UI and API

Troubleshooting

---

This checklist is a subscriber-only quick reference from Signal over Noise. I publish practitioner build logs every two weeks — real architectures from running production AI systems.

Next build log to read: The 3-Layer Claude Code Configuration That Runs 10 Projects