Develop

Implement Subframe designs in the codebase. Fetch the design via MCP, sync components, and add business logic.

## MCP Authentication

If you cannot find the `get_page_info` tool (or any Subframe MCP tools), the MCP server likely needs to be authenticated. Ask the user to authenticate the Subframe MCP server. If the user is using Claude Code or Codex, instruct them to run `/mcp` to view and authenticate their MCP servers, and then say "done" when they're finished.

## Detect Project State

Before starting, check for `package.json` and `.subframe/` folder in the current directory:

| Condition                                      | Action                                                                    |
| ---------------------------------------------- | ------------------------------------------------------------------------- |
| No `package.json`                              | Run `/subframe:install` first — there's no project to implement into yet. |
| Has `package.json` AND has `.subframe/` folder | Proceed with the workflow below.                                          |
| Has `package.json` but NO `.subframe/` folder  | Ask the user (see below).                                                 |

### Existing non-Subframe project

If the current directory has a `package.json` but no `.subframe/` folder, ask the user which approach they prefer:

- **Use the design as inspiration** — Fetch the design via MCP for reference, but implement the page using the existing styles, components, and patterns already in the repo. Translate the Subframe design's layout and structure into whatever UI framework the project already uses (e.g., existing component library, CSS modules, styled-components). Do NOT install Subframe or sync components. Skip to [Inspiration Workflow](#inspiration-workflow).
- **Use Subframe styles and components** — Install Subframe into the project so the design renders pixel-perfect with Subframe's generated code. Run `/subframe:install` first, then continue with the [Workflow](#workflow) below.

## Workflow

1. **Wait for any in-flight design jobs** — see [Awaiting In-Flight Designs](#awaiting-in-flight-designs)
2. **Fetch the design** — `get_page_info` with the URL, ID, or name
3. **Read design documentation** — `get_project_info` and `get_component_info` return any attached design docs; check them for usage guidance, accessibility notes, or constraints before implementing
4. **Sync any missing components** — Only if components don't exist locally. `npx @subframe/cli sync` for the specific components used in the page
5. **Create the page** — put it in the right place per codebase patterns
6. **Add business logic** — data fetching, forms, events, loading/error states

## Awaiting In-Flight Designs

If a design was just kicked off in the same conversation (via `/subframe:design`), the underlying AI job is likely still running. Reading the result via `get_*_info` too early returns empty or stale code.

`design_page`, `design_component`, and `edit_component` return a `jobId`. Before the first read, call:

```
wait_for_jobs({ jobIds: [jobId1, jobId2, ...] })
```

Each result is `running`, `done` (with optional summary), or `not_found`. Call in a loop until every job is `done`. Surface progress to the user — "Designs are still generating in Subframe…" then "✓ Designs ready, fetching the code now." — so they understand the wait. (Jobs that stall longer than ~10 minutes are surfaced as `done` so the loop never hangs.)

You don't need `wait_for_jobs` when:

- The user came in with an existing Subframe URL (no in-flight job)
- You're only working from `id` or `name` and don't need the generated code.

If the user asks to implement immediately after kicking off a design, batch all relevant `j