mirror of
https://github.com/different-ai/openwork
synced 2026-04-25 17:15:34 +02:00
* refactor(repo): move OpenWork apps into apps and ee layout Rebase the monorepo layout migration onto the latest dev changes so the moved app, desktop, share, and cloud surfaces keep working from their new paths. Carry the latest deeplink, token persistence, build, Vercel, and docs updates forward to avoid stale references and broken deploy tooling. * chore(repo): drop generated desktop artifacts Ignore the moved Tauri target and sidecar paths so local cargo checks do not pollute the branch. Remove the accidentally committed outputs from the repo while keeping the layout migration intact. * fix(release): drop built server cli artifact Stop tracking the locally built apps/server/cli binary so generated server outputs do not leak into commits. Also update the release workflow to check the published scoped package name for @openwork/server before deciding whether npm publish is needed. * fix(workspace): add stable CLI bin wrappers Point the server and router package bins at committed wrapper scripts so workspace installs can create shims before dist outputs exist. Keep the wrappers compatible with built binaries and source checkouts to avoid Vercel install warnings without changing runtime behavior.
2.9 KiB
2.9 KiB
Session Flow UX: Human Narrative Pass
What changed
- Removed user-facing meta labels in session message blocks (no
Plan,Activity, orAnswertags). - Kept grouped execution blocks, but changed their collapsed copy to natural language (
Working on it,Hide details). - Updated grouped execution summaries to avoid raw tool-centric wording.
- Added richer reasoning summaries in
summarizeStep()so reasoning can render as:Thinking: <specific headline>- optional paragraph detail under it.
- Removed the extra freeform run-footer detail paragraph to avoid duplicate/noisy narration.
Why
- Users were seeing internal/meta framing in the timeline instead of a human flow.
- The previous footer detail line looked detached from the timeline and could surface markdown-like noise.
- The new pass keeps the same underlying events while presenting a cleaner narrative sequence.
End-to-end evidence (Chrome MCP)
Screenshot 1: Two prompts in one session, with grouped execution rows
Verified:
- Multiple convo turns render in one timeline.
- Grouped execution rows use user-friendly copy (
Working on it). - No
Plan/Activity/Answertags appear.
Screenshot 2: Second prompt with grouped steps expanded
Verified:
- Expanded block reveals concrete steps while preserving the simplified outer copy.
- The group can be collapsed/expanded with
Hide details.
Screenshot 3: First prompt with grouped steps expanded
Verified:
- Earlier turn remains readable and expandable.
- Step grouping and history remain stable across multiple prompts.
Conversation prompts used during verification
smoke test 1: explain why grouped tool calls appear in session viewsmoke test 2: think out loud about one possible UX redesign and then run checks
Test commands run
pnpm typecheck(pass)pnpm --filter @different-ai/openwork-ui test:health(fails in this environment:Unauthorizedon/global/health)pnpm --filter @different-ai/openwork-ui test:events(sameUnauthorizedfailure)pnpm --filter @different-ai/openwork-ui test:sessions(sameUnauthorizedfailure)pnpm --filter @different-ai/openwork-ui test:todos(sameUnauthorizedfailure)pnpm --filter @different-ai/openwork-ui test:permissions(sameUnauthorizedfailure)pnpm --filter @different-ai/openwork-ui test:session-switch(sameUnauthorizedfailure)pnpm --filter @different-ai/openwork-ui build(fails in this environment: missing optional@rollup/rollup-darwin-arm64)
Environment notes
- Docker stack started via
packaging/docker/dev-up.shin this worktree. - Web UI validated at the emitted URL with Chrome MCP.


