What credentials are required?

You need a Braintrust API key and must set the TRACE_TO_BRAINTRUST environment variable to true in your configuration.

What does the Maestro Quality skill do?

It provides deep integration with Braintrust to trace Maestro agent sessions, capturing everything from user prompts to sub-agent tool calls and LLM outputs.

Can I trace sub-agents with this skill?

Yes, it includes specific logic to correlate sub-agent runs (which often create separate traces) back to their parent session using metadata and timestamps.

Where are the session logs and state files stored?

The skill maintains state files and logs within the ~/.maestro/state/ directory, including per-session JSON files and a global hook log.

Braintrust Tracing for Maestro

Name: Braintrust Tracing for Maestro
Author: scooter-lacroix

byscooter-lacroix

分析与监控

Enables comprehensive session tracing and sub-agent correlation via Braintrust for deep observability into agent workflows.

关于

The Maestro Quality skill provides a sophisticated integration with Braintrust, allowing developers to trace every aspect of their Claude Code sessions. By leveraging a structured hook architecture, it captures root sessions, individual turns, tool executions, and LLM spans. It specifically addresses the complexity of sub-agent workflows, offering patterns for correlating orphaned sub-agent traces back to parent sessions. This skill is indispensable for developers building production-grade AI agents who require detailed logs, performance metrics, and debugging capabilities to refine agent logic and cost efficiency.

主要功能

0 GitHub stars
Advanced sub-agent correlation for multi-agent workflows
Analysis scripts for replaying and querying traced agent data
Real-time debug logging and session state management
Hierarchical trace mapping for sessions, turns, and tool spans
Automatic capture of tool inputs, outputs, and LLM metadata

使用场景

Monitoring token usage and model performance across distributed sub-agents
Debugging complex multi-step agent interactions and tool failures
Collecting structured datasets of agent runs for evaluation and fine-tuning

关于

主要功能

0 GitHub stars
Advanced sub-agent correlation for multi-agent workflows
Analysis scripts for replaying and querying traced agent data
Real-time debug logging and session state management
Hierarchical trace mapping for sessions, turns, and tool spans
Automatic capture of tool inputs, outputs, and LLM metadata

使用场景

Monitoring token usage and model performance across distributed sub-agents
Debugging complex multi-step agent interactions and tool failures
Collecting structured datasets of agent runs for evaluation and fine-tuning