Skip to content

parcadei/Continuous-Claude-v3 / .claude/plugins/braintrust-tracing/skills/trace-claude-code

trace-claude-code

Unverified

Automatically trace Claude Code conversations to Braintrust for observability. Captures sessions, conversation turns, and tool calls as hierarchical traces.

Trace Claude Code to Braintrust

Automatically send Claude Code conversations to Braintrust for tracing and observability. Get full visibility into your AI coding sessions with hierarchical traces showing sessions, turns, and every tool call.

What you get

Claude Code Session (root trace)
├── Turn 1: "Add error handling"
│   ├── Read: src/app.ts
│   ├── Edit: src/app.ts
│   └── Response: "I've added try-catch..."
├── Turn 2: "Now run the tests"
│   ├── Terminal: npm test
│   └── Response: "All tests pass..."
└── Turn 3: "Great, commit this"
    ├── Terminal: git add .
    ├── Terminal: git commit -m "..."
    └── Response: "Changes committed..."

How it works

Four hooks capture the complete workflow:

HookWhat it captures
SessionStartCreates root trace when you start Claude Code
PostToolUseCaptures every tool call (file reads, edits, terminal commands)
StopCaptures conversation turns (your message + Claude's response)
SessionEndLogs session summary when you exit

Quick setup

Run the setup script in any project directory where you want tracing:

bash /path/to/skills/trace-claude-code/setup.sh

The script prompts for your API key and project name, then configures all hooks automatically.

Manual setup

Prerequisites

Configuration

Create .claude/settings.local.json in your project directory:

{
  "hooks": {
    "SessionStart": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "bash /path/to/hooks/session_start.sh"
          }
        ]
      }
    ],
    "PostToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "bash /path/to/hooks/post_tool_use.sh"
          }
        ]
      }
    ],
    "Stop": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "bash /path/to/hooks/stop_hook.sh"
          }
        ]
      }
    ],
    "SessionEnd": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "bash /path/to/hooks/session_end.sh"
          }
        ]
      }
    ]
  },
  "env": {
    "TRACE_TO_BRAINTRUST": "true",
    "BRAINTRUST_API_KEY": "sk-...",
    "BRAINTRUST_CC_PROJECT": "my-project"
  }
}

Replace /path/to/hooks/ with the actual path to this skill's hooks directory.

Environment variables

VariableRequiredDescription
TRACE_TO_BRAINTRUSTYesSet to "true" to enable tracing
BRAINTRUST_API_KEYYesYour Braintrust API key
BRAINTRUST_CC_PROJECTNoProject name (default: claude-code)
BRAINTRUST_CC_DEBUGNoSet to "true" for verbose logging

Viewing traces

After running Claude Code with tracing enabled:

  1. Go to braintrust.dev
  2. Navigate to your project (e.g., claude-code)
  3. Click Logs to see all traced sessions

Each trace shows:

  • Session root: The overall Claude Code session
  • Turns: Each conversation exchange (user input → assistant response)
  • Tool calls: Individual operations (file reads, edits, terminal commands)

Trace structure

Traces are hierarchical:

  • Session (root span)

    • span_attributes.type: "task"
    • metadata.session_id: Unique session identifier
    • metadata.workspace: Project directory
  • Turn (child of session)

    • span_attributes.type: "llm"
    • input: User message
    • output: Assistant response
    • metadata.turn_number: Sequential turn number
  • Tool call (child of turn or session)

    • span_attributes.type: "tool"
    • input: Tool input (file path, command, etc.)
    • output: Tool result
    • metadata.tool_name: Name of the tool used

Troubleshooting

No traces appearing

  1. Check hooks are running:

    tail -f ~/.claude/state/braintrust_hook.log
    
  2. Verify environment variables in .claude/settings.local.json:

    • TRACE_TO_BRAINTRUST must be "true"
    • BRAINTRUST_API_KEY must be valid
  3. Enable debug mode:

    {
      "env": {
        "BRAINTRUST_CC_DEBUG": "true"
      }
    }
    

Permission errors

Make hook scripts executable:

chmod +x /path/to/hooks/*.sh

Missing jq command

Install jq:

  • macOS: brew install jq
  • Ubuntu/Debian: sudo apt-get install jq

State issues

Reset the tracing state:

rm ~/.claude/state/braintrust_state.json

Hook logs

View detailed hook execution logs:

# Follow logs in real-time
tail -f ~/.claude/state/braintrust_hook.log

# View last 50 lines
tail -50 ~/.claude/state/braintrust_hook.log

# Clear logs
> ~/.claude/state/braintrust_hook.log

File structure

hooks/
├── common.sh          # Shared utilities (logging, API, state)
├── session_start.sh   # Creates root trace span
├── post_tool_use.sh   # Captures tool calls
├── stop_hook.sh       # Captures conversation turns
└── session_end.sh     # Finalizes trace

Alternative: SDK integration

For programmatic use with the Claude Agent SDK, use the native Braintrust integration:

import { initLogger, wrapClaudeAgentSDK } from "braintrust";
import * as claudeSDK from "@anthropic-ai/claude-agent-sdk";

initLogger({
  projectName: "my-project",
  apiKey: process.env.BRAINTRUST_API_KEY,
});

const { query, tool } = wrapClaudeAgentSDK(claudeSDK);

See Braintrust Claude Agent SDK docs for details.

reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 463kUpdated Claude CodeCodexVerifiedInstall
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 463kUpdated Claude CodeCodexVerifiedInstall
discernment-nudgeanthropics/skillsAfter you give a substantive answer or draft that the user may act on — advice or recommendations, drafted artifacts such as goals, plans, pitches, proposals, or emails, estimates or projections, analysis or interpretation of data, factual claims they may rely on, or a multi-step argument — invoke this skill BEFORE finalizing your reply and then, if it applies, append 2-3 short follow-up questions, each tied to something specific in what you just produced, that help the user check key facts, probe the reasoning or assumptions, and notice missing context. Do this at most once per conversation. Skip it when the user asked a trivial how-to or simple lookup, wants a purely educational explanation, asked you only to format, convert, or assemble a file from content they provided, is writing code they will run, is doing creative writing or casual chat, or already asked you to double-check, cite, or review — the skill file explains these boundaries and the exact output format.Stars: 195kUpdated Claude CodeCodexVerifiedInstall
internal-commsanthropics/skillsA set of resources to help me write all kinds of internal communications, using the formats that my company likes to use. Claude should use this skill whenever asked to write some sort of internal communications (status reports, leadership updates, 3P updates, company newsletters, FAQs, incident reports, project updates, etc.).Stars: 195kUpdated Claude CodeCodexVerifiedInstall