sport-agent

Sport Agent

Expandable training + nutrition framework. Currently two athletes (Denis, Alicja); adding another is copying a folder — see people/_template/README.md.

Structure

sport-agent/
├── README.md                       ← this file
├── CLAUDE.md                       ← standing instructions loaded every session
├── STATUS.md                       ← short-lived open items (FTP test pending, etc.)
├── .claude/
│   ├── settings.json               ← tool permissions (which git commands are pre-approved)
│   └── skills/                     ← the four recurring workflows; each SKILL.md is the live prompt
│       └── README.md               ← how they fit together + restore steps for a new machine
│
├── shared/                         ← methodology & reference (applies to everyone)
│   ├── recommendation_protocol.md  ← the recommendation engine spec (inputs → outputs → guardrails)
│   ├── training_principles.md      ← readiness, deload, session types, removal hierarchy, mobility
│   ├── calisthenics_ladders.md     ← skill-progression ladder definitions
│   └── keto_nutrition_reference.md ← per-100 g food table, TKD fueling, electrolytes
│
├── people/
│   ├── index.json                  ← list of person ids (dashboard person switcher)
│   ├── _template/                  ← copy this folder to add a person
│   ├── denis/
│   │   ├── data.json               ← ALL structured data: profile, goals, week, sessions, nutrition
│   │   ├── training_plan.md        ← reasoning: priorities, guardrails, weekly logic
│   │   ├── gym_training_plan.md    ← gym split detail & progression rules
│   │   ├── calisthenics_status.md  ← SKILL_STATE (current levels) + focus
│   │   └── workout_log.md          ← history; logged via dashboard
│   └── alicja/
│       ├── data.json
│       ├── training_plan.md
│       ├── gym_training_plan.md
│       ├── mobility_splits.md
│       ├── running_plan.md
│       └── workout_log.md          ← history; dashboard needs a token on her device
│
├── tools/
│   └── dashboard.html              ← ONE dashboard for everyone: ?person=<id>

Design rules

• data.json = data, markdown = reasoning. Anything a tool renders (schedule, exercises, targets, macros, supplements) lives in data.json. The why lives in markdown.
• One dashboard, config-driven. tools/dashboard.html?person=denis — tabs, sessions, and colors come from that person’s data.json. Nutrition tab is read-only by design.
• Readiness is subjective and identical for everyone: morning 1–5 ratings for sleep, soreness, energy, motivation → green/amber/red (rule in shared/training_principles.md). No wearable required.
• The engine is the assistant. shared/recommendation_protocol.md defines how plans, daily briefs, and weekly reviews are generated from data.json + logs + readiness.

Dashboard

Open tools/dashboard.html?person=<id> (GitHub Pages or locally via any static server). Reading works without credentials on Pages; saving log entries needs a GitHub fine-grained token (repo contents read/write) entered via ⚙︎ — stored only in that browser.

Using the assistant

Say whose plan you mean. The assistant follows CLAUDE.md (loaded automatically every session) and generates recommendations per shared/recommendation_protocol.md.

The recurring workflows are project skills, invocable by name in any session:

Command	What it does
/daily-session-brief	Today’s session for Denis with last loads, readiness, fueling. Read-only.
/weekly-review-and-plan	Sunday review for both athletes; progresses loads and plans next week. Commits.
/start-training-block	Builds Denis a 4-week periodized block. Manual, commits.
/monthly-nutrition-review	Supplement + macro audit for both. Read-only.

weekly-review-and-plan (Sunday 18:00) and monthly-nutrition-review (1st of the month, 19:00) also run on a schedule. daily-session-brief is registered but paused — activate it when a training block starts. See .claude/skills/README.md.