scot/diligence
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
diligence/frontend/public/.well-known/skills/diligence-fitness/SKILL.md

2.6 KiB
Raw Blame History

Diligence — Agent Skill Definition

What Is Diligence?

Diligence is a self-hosted fitness rewards platform. Users earn points through fitness activities (workouts, food logging, step goals) and spend them on configurable rewards (gaming time, screen time, etc.). A daily gate keeps rewards locked until enough points are earned. Points reset weekly.

Architecture

  • Backend: FastAPI (Python 3.12) + PostgreSQL 16
  • Frontend: React 18 + Vite, served by nginx
  • MCP Connector: FastMCP (Streamable HTTP/SSE) on port 3001, proxied via nginx at /mcp
  • Deployment: Docker Compose (4 services: frontend, backend, mcp-connector, fitness-db)

Connecting

Point your MCP client to:

  • Development: http://localhost:3001/sse
  • Production: https://your-domain/mcp

Available Tools (14)

Tool Category Description
get_context() Status Full profile, motivation, program, rules, rewards
get_today(date?) Status Daily points, gate status, activities
get_week(start_date?) Status Weekly summary, active days
log_activity(category, title, duration_minutes, ...) Activity Log workout, earn points
log_food(meal_type, food_name, ...) Food Log food with macros
search_food(query) Food Search Open Food Facts + USDA
get_program_schedule(date?) Programs Today's workout from active program
redeem_reward(reward_name) Rewards Spend points on a reward
load_meal_plan(name, duration_days, meals, ...) Meal Plans Create a full meal plan
get_meal_plan(date?) Meal Plans View today's planned meals
update_meal_compliance(plan_item_id, status, ...) Meal Plans Mark meal followed/skipped
get_plan_progress(plan_id?) Meal Plans Compliance statistics
configure_integration(provider, credentials) Integrations Store encrypted credentials (write-only)
get_integration_status() Integrations Check provider connection status

Key Design Decisions

  • The agent generates meal plans — the app only stores and tracks them. Zero AI API cost.
  • Integration credentials are write-only through MCP — agents can set but never read them back.
  • The app collects BREQ-2 motivation data during onboarding. Use get_context() to read the user's motivation profile and calibrate tone accordingly.
  • Points reset weekly. Each week is a fresh start. No guilt carry-over.

Source