# __APP_NAME__ — map for AI agents

> A pointer map, not a rulebook. Every rule lives in one place, fetched live —
> nothing here is restated, so nothing here can go stale.

## Start here, every session

1. **`project_state`** (ai-builder MCP) — the app, its modules/models, the
   installed layer versions, and the state of [`tracking/plan.md`](tracking/plan.md).
   It returns a `nextAction`. Do that.
2. No MCP server connected? `npx -y @dreamtree-org/ai-builder init --ai <provider>`.

## Where the answers are

| Question | Source |
| --- | --- |
| What am I working on, and where did the last session stop? | [`tracking/plan.md`](tracking/plan.md) — the Resume-here block |
| What is this app, architecturally? | [`tracking/big-picture.md`](tracking/big-picture.md) |
| Known bugs, quality debt, ideas | [`tracking/backlog.md`](tracking/backlog.md) |
| The binding build rules + RBAC + how to add a feature | MCP `build_rules` |
| How to read/write data; schema column format | MCP `data_contract` |
| A twreact-ui component's real props | MCP `layer_docs` — `mode: "component"` |
| A KORM option at the version we installed | MCP `layer_docs` — `package: "@dreamtree-org/korm-js"` |
| Am I on current versions? | MCP `check_version` |
| Running the app, scripts, Docker | [`README.md`](README.md) |
| What the tests cover | [`tests/USER-JOURNEYS.md`](tests/USER-JOURNEYS.md) |

## Two rules worth repeating here

- **Plan before you build, and record progress as you go.** `tracking/plan.md`
  is rewritten at every task transition, not at session end — a session can end
  mid-task, and that file is all the next one gets.
- **Never state a layer API from memory.** Component props and KORM options come
  from `layer_docs`, which reads the package this app actually installed.

This map is provider-neutral — read it whichever assistant you are. The skill pack
itself was installed for `__AI_PROVIDER__` at scaffold; add another with
`npx -y @dreamtree-org/ai-builder init --ai <provider>`.