airbone42/360-data-athlete

360 Degree Data Athlete is an experimental project that turns Claude Code, a coding assistant tool, into an AI training coach for endurance and strength athletes. It is distributed as a Claude Code plugin and is clearly labeled as not a replacement for a real coach or medical advice, meant only for people with a solid training background who accept the risk of using it, with no warranty or support. The interesting idea behind the project is less about sport and more about a use of Claude Code itself. The author is not trying to build another multi-agent framework, since tools like LangChain, AutoGen, or CrewAI already exist for that. Instead, the project treats Claude Code as a general purpose agent system that can run a long, file backed workflow with many specialized helpers, applied to a real world problem instead of software development. Training planning was chosen as the test case because it involves conflicting constraints, like heart rate recovery versus a race schedule, or injury versus progress, plus a real person, the maintainer, who pushes back when the system gets something wrong. In practice, the plugin includes a team of specialized sub-agents: a planner, three different workout specialists, a mental coaching agent, a video analyst, a post activity analyst, a data scientist, a plan validator, a configuration auditor, and clinical consultants. Together they build a daily training plan that takes into account heart rate variability, training load measures, sleep, weather, upcoming races, and any current injuries, then pushes the finished plan to intervals.icu, a training log service, and can also connect to Strava, Garmin devices, and optionally Telegram. Before a plan is shown to the athlete, it goes through both a rule based mechanical check and a separate AI based semantic reviewer. The system also tries to avoid overreacting to a normal drop in heart rate variability after a hard workout, using a small model fitted on the athlete's own recent training history to judge whether that day's reading is expected or worth flagging. There is also a video based form check for strength and running movements, muscle level fatigue tracking, and an audit feature that looks for inconsistencies across the project's own configuration files and agents. This is aimed at technically minded endurance or strength athletes who are comfortable running and configuring Claude Code themselves. The full README is longer than what was shown.