@2008muyu/pi-stats-insight

This Skill addresses the lack of fine-grained observability data when using the pi coding agent. While pi's built-in footer displays token and context usage, developers often need key metrics like cache hit rate, tool call statistics, average response time, and provider errors—critical data for optimizing AI call costs and troubleshooting performance bottlenecks.

Usage is straightforward: after installation, no configuration is required. It automatically adds a compact status bar at the bottom of pi, showing cache hit rate level, tool call count, average latency, and error count. Additionally, entering the /stats command opens an interactive menu to drill down into token and cost details, cache performance, per-tool call statistics, context window usage, and HTTP error specifics. To reset session statistics, simply enter /stats reset.

It's ideal for agent developers, backend engineers, and full-stack engineers using the pi coding agent for daily development—especially teams already integrated with multiple AI providers (like Anthropic, OpenAI, Google, Mistral, etc.) who need unified monitoring of cache effectiveness and costs across providers. This Skill directly fills that gap.

It's recommended to monitor cache hit rate data after major refactors or model switches to evaluate cache strategy effectiveness. Note that it only tracks data within the current session; statistics reset after restarting the session, but historical data is persisted locally in JSONL format for later analysis. Also, the Skill relies on pi's ctx.ui.setStatus interface and does not replace pi's native footer—they coexist without interference.