Make a blank Flect workspace immediately understandable and useful

[robinbraemer]

Parent epic: #40

Outcome

On first open, a person understands in plain language that Flect can create and refine a useful running interface from what they need. They can begin without choosing a framework, agent role, project type, or setup mode.

Product-quality coverage

FQ-01.7, FQ-02.1, FQ-02.5, FQ-02.6, FQ-03.1, FQ-03.3, FQ-19.1, and FQ-19.2.

Dependencies

• Make first-run Pi sign-in and model readiness zero-friction #19 owns real provider readiness and authentication.
• Persist workspace, conversation, and canvas state across reloads and failures #21 owns durable resume behavior.
• Complete accessibility, reflow, and adaptive appearance gates #22 owns accessibility across the result.

Required behavior

• A blank workspace has one calm, outcome-oriented starting point that explains the next useful action without documentation.
• Provide a small set of representative, editable starting requests grounded in outcomes, not implementation jargon. Choosing one only fills or sends the same protected composer; it never creates a hidden second workflow.
• Clearly distinguish an answer, an edit, and a request for information before taking action. Questions must not accidentally change the canvas.
• Explain unavailable model, workspace, or capability state at the point of need with one safe next action.
• Preserve a draft across readiness and recovery paths; never submit it twice.
• On an opened interface, show an authoring entry point only when Flect can actually activate against that interface. Its label describes the user outcome, not Flect internals.
• Never show a selection instruction until there is an active, selectable interface target.

Acceptance criteria

• In a bounded first-use study, participants can state what Flect does and begin a meaningful task without external help or product-internal vocabulary.
• A clean browser and packaged macOS path reaches a usable composer or a precise recovery action from no-auth, offline, unavailable-model, and existing-auth states.
• A participant can create a small working interface from a blank workspace with one ordinary prompt, then ask a factual question without changing it.
• The start surface does not offer an action that lacks a current target or silently opens an unrelated workflow.
• Keyboard, screen reader, 320 CSS px, 200% zoom, light/dark, increased text, reduced motion, and focus states remain complete.
• Production Chromium and packaged macOS tests exercise the public flow; the evidence includes an Experience observation, not only automated checks.

Constraints

• Keep the main canvas and the person’s intended work more prominent than Flect chrome.
• Do not require framework selection, a project wizard, a role picker, or a terminal.
• Do not leak authentication details, credentials, or provider payloads into the visible shell, logs, or test artifacts.

Non-goals

• A generic template marketplace.
• Claiming that any arbitrary backend, package, or operating-system action is available.
• Replacing Make first-run Pi sign-in and model readiness zero-friction #19's credential and provider lifecycle implementation.

Owning documentation

VISION.md, PRODUCT.md, DESIGN.md, and docs/product-quality.md; record proof only in a dated docs/verification/ report.

[robinbraemer]

Implemented the first vertical slice in 3e53a2a.

• Static and hydrated blank workspaces now state the outcome in product language and offer three editable example prompts.
• Selecting an example fills and focuses the single protected composer; it does not submit or activate work.
• Canvas-only controls remain absent before a live target exists.

Evidence: bun run check (898 passed, 1 skipped) and bun run build passed. The production preview was exercised in Chromium: all examples rendered; selecting one preserved the static shell, focused the composer, filled the prompt, and produced no console errors. The Playwright run is currently blocked by an existing local process on port 3210, so this is implementation evidence, not a closure claim.