René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"Add starter prompt suggestions to the empty conversation state","articleBody":"## Problem or Motivation\n\nWhen you open a new conversation, OpenWork shows a centered composer under the heading **\"What should we build?\"** — but nothing else. New users (and returning ones facing a blank canvas) get no hint about what the agent can do or how to phrase a good first request.\n\nEvery comparable desktop client fills this gap with **prompt suggestions**: Claude Code Desktop shows curated \"prompt suggestions\", and ChatGPT / Codex show example starter prompts on their empty state. OpenWork has a rotating placeholder in the input, but no clickable starting points.\n\n## Proposed Solution\n\nRender a small row of **starter prompt suggestion chips** directly beneath the centered composer on the empty conversation state (only while the composer is empty). Clicking a chip **populates the composer** with a fuller prompt and focuses the input — it does **not** auto-send, so the user can edit before submitting.\n\n- 4 generic, build-oriented suggestions that match the \"What should we build?\" framing: *Explain this codebase*, *Build a feature*, *Fix a bug*, *Write tests*.\n- Each chip shows a short localized label + icon; clicking seeds the composer with the fuller localized prompt.\n- Suggestions hide as soon as the composer has content (typed or seeded), matching the empty-state-only behavior in other clients.\n\n### Feasibility\n\n**Frontend-only.** The empty-state composer already lives in `apps/electron/src/renderer/components/app-shell/ChatDisplay.tsx` (`showCenteredEmptyComposer`). Seeding the composer reuses the existing controlled-draft channel (`onInputChange` → `inputValue`), the same path drafts already use — no backend/qwen-code changes. Content is static and localized via i18n. No new IPC, no new persistence.\n\n### Acceptance criteria (what the CDP assertion checks)\n\n1. On the empty conversation, a `[data-testid=\"empty-suggestions\"]` surface renders with 4 chips.\n2. The composer starts empty.\n3. Clicking the first chip populates the composer with a prompt that is longer than the chip's short label (i.e. it seeds the *prompt*, not just the title).\n4. The suggestions surface disappears once the composer has content.\n\n## Alternatives Considered\n\n- **Auto-send on click** — rejected; seeding + letting the user edit is safer and matches Claude Code Desktop / ChatGPT.\n- **Workspace-type-aware or dynamically generated suggestions** — deferred; a small static, localized set is lower-risk and fully frontend.\n\n## Additional Context\n\nSource references: Claude Code Desktop \"prompt suggestions\"; ChatGPT / Codex empty-state example prompts. Part of the autonomous desktop-feature parity loop (`loop-bot`).","author":{"url":"https://github.com/DragonnZhang","@type":"Person","name":"DragonnZhang"},"datePublished":"2026-07-09T01:37:11.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":0},"url":"https://github.com/72/openwork/issues/72"}