platform-retrofit: 8 stories / 18 tasks (spec bundle into docs/product, single-version release-please before deletions, demo feature removal, web-next auth shell, optional cores, work-tree archive, library traces, agent-doc rewrite). walking-skeleton: 11 stories / 28 tasks across workspaces, discovery, core-runner-protocol, runner app, editor, and the flagship connect-to-selection e2e; builds on platform-retrofit (prose dependency note — cross-epic frontmatter unsupported). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016j8z4VHjedXDTjEDNg7qHK
3.4 KiB
3.4 KiB
id, epic, title, type, status, feature, depends-on, blocks, created
| id | epic | title | type | status | feature | depends-on | blocks | created | ||
|---|---|---|---|---|---|---|---|---|---|---|
| 05-runner-scan-and-preview-adapter | walking-skeleton | runner scan stage + embedded-Vite preview adapter + canvas agent v0 | technical-story | todo | runner |
|
|
2026-07-12T10:24:14Z |
Goal
Complete apps/runner: a minimal react-docgen-typescript scan that returns ≥1 component (name + props) from the fixture, and an embedded-Vite preview adapter rooted at the clone that serves a Veect frame-host page — importing the component requested via render-frame and mounting the injected, Veect-authored canvas agent script v0.
Why
This is invariant #1 made real: the component renders through the repo's own toolchain, not a Veect re-implementation (ADR-028 — no NodeRenderer on the board path). The agent script is the editor's only eyes inside the cross-origin iframe; it must be adapter-injected and never sourced from the repo (ADR-028 consequences — security floor even at fixture scale).
Done when
- Scan stage runs react-docgen-typescript over the fixture's component glob and emits a scan-result message with ≥1 component carrying name + props (skeleton fidelity; full
ComponentMetacomes withdiscovery-and-library). - Preview adapter starts embedded Vite rooted at the clone on a dedicated port = dedicated origin (spec §9), serving a frame-host entry that imports the requested component with default props (
render-frame) and mounts the agent script. - Canvas agent v0 reports
runtime.ready, component bounding boxes (postfonts.loaded), and click-target resolution;targetOriginpinned in both directions (ADR-028). - Agent script is Veect-authored and adapter-injected — never sourced from the repo.
- Protocol integration suite extends to: scan returns ≥1 component → adapter serves the frame host with the agent script → adapter origin liveness (PRD Success criteria).
pnpm typecheck && pnpm lint && pnpm test && pnpm conformance && pnpm fallow:audit && pnpm coverage:diffall pass after each task.
In scope
- Scan stage (react-docgen-typescript, minimal glob) + scan-result message.
- Embedded-Vite adapter, frame-host entry,
render-framehandling, adapter-ready/status events. - Canvas agent script v0 (ready, geometry, click-target resolution; pinned
targetOrigin). - Integration tests for scan payload, adapter origin liveness, and agent-script presence.
Out of scope
- The editor-side canvas-protocol client (story 09).
- Watch/HMR, perf-500 budgets (
discovery-and-library); Next.js adapter (preview-and-nextjs-depth). - Multi-view frames, isolation, code panel (later PRDs).
Tasks
- Add the scan stage: minimal react-docgen-typescript scan over the fixture's component glob → scan-result message with ≥1 component (name + props); integration test asserts the
Buttonpayload arrives over real WS. - Add the preview adapter: embedded Vite rooted at the clone on a dedicated port/origin (spec §9), frame-host entry importing the component requested via
render-frame(default props) and mounting the injected Veect-authored canvas agent script v0 (runtime.ready, bounding boxes postfonts.loaded, click-target resolution; pinnedtargetOriginboth directions; never sourced from the repo); adapter-ready/status events; integration test asserts adapter origin liveness and that the frame host serves with the agent script.