Documentation

This is the index. Start with whichever section matches what you're looking for.

Product

Overview — the big picture, components, data flow, stack
Agent Runtime — how runSkill works
Execution Engine — the safety boundary between AI and exchange
Simulation — backtest replay loop
Live Runtime — Fly-based per-skill processes
Chat Agent — single product-wide chat agent: authoring, coaching, prepare-action confirm-cards
Chat Agent Prompts — exact persona, tier addenda, mode addenda, trust rules, dialogue flows, suggested prompts
Tools & MCP — tool registry pattern and MCP integration
Data Model — Supabase schema

Architecture Decision Records — one per non-trivial choice we've made, with context and consequences.

Environments — dev / preview / production matrix
Launch checklist — five-phase path from code-in-repo to first real-money tick, with verification queries and incident response

One topic per file. If a doc grows past ~400 lines, split it.
Lead with the decision or summary. Reader should know the gist in 30 seconds.
Diagrams in ASCII in the doc itself. Avoid binary diagrams that drift from the code.
Cross-link liberally. Use relative links between docs.
Date the doc if it captures a point-in-time decision. ADRs include a date in the header.
Update or delete. Stale docs are worse than missing docs.