Claudewatch

关于

Claudewatch addresses the lack of feedback loops in AI development by providing comprehensive observability for Claude Code sessions. It offers developers structured visibility into critical metrics like friction patterns, cost-per-outcome, and agent success rates, using local session data without telemetry. Simultaneously, it empowers Claude itself with real-time access to its own session history, performance, and project health via an MCP server, enabling more informed in-session decision-making. This tool transforms raw session data into actionable insights, helping to optimize AI workflows and measure the effectiveness of changes.

主要功能

• Multi-Agent Workflow Analytics: Reconstruct agent lifecycles to analyze success rates, parallelization, and token costs by agent type.
• Automated CLAUDE.md Patch Generation: Generate data-driven improvements to CLAUDE.md files, optionally AI-powered.
• Structured Session Analytics: Gain visibility into friction patterns, cost-per-outcome, and agent success rates from local Claude Code data.
• CLAUDE.md Effectiveness Scoring: Measure the impact of prompt changes with before/after comparisons on key metrics.
• Real-time AI Self-Observability (MCP): Provide Claude with in-session access to its own history, performance, and project health.
• 0 GitHub stars

使用案例

• Empowering Claude with self-awareness for better in-session decision-making to enhance productivity.
• Optimizing Claude Code prompts and CLAUDE.md files for improved efficiency and reduced friction.
• Monitoring and improving the performance and cost-effectiveness of AI agent workflows.