Export Collapsible Tool Output

Scope

Verify that long shell output and long think tool content keep their full text in the shared WebUI while remaining collapsed by default. This covers the HTML export viewer and the other consumers of @qwen-code/webui.

Baseline

A released global qwen executable was not available in this environment, so the before-change /export dry-run could not be performed. Source inspection on upstream/main reproduces the loss: ShellToolCall and ThinkToolCall replace content after 500 characters with an ellipsis.

Manual Scenario

1. Export a session containing a successful Bash or Execute tool call whose output is longer than 500 characters and ends with a unique marker.
2. Open the exported HTML and find the tool call.
3. Confirm the output is collapsed and has a Show more control.
4. Expand it and confirm the unique tail marker is visible, then collapse it again.
5. Repeat with a think tool call longer than 500 characters.

Regression Checks

• Content of exactly 500 characters has no new toggle.
• Both Bash and Execute variants preserve the full output.
• A long single-line Shell result remains horizontally scrollable after it is expanded.
• A collapsible Shell result does not widen the enclosing message container, and its toggle remains inside the visible message bounds.
• Clicking a shell toggle does not invoke the surrounding OUT-row action.
• Clicking the shell OUT row still opens the complete output.
• Error, empty-output, and short-thinking paths remain unchanged.
• Toggle buttons expose their expanded state to assistive technology.

Automated Verification

cd packages/webui
npx vitest run \
  src/components/toolcalls/shared/CollapsibleOutput.test.tsx \
  src/components/toolcalls/GenericToolCall.test.tsx \
  src/components/toolcalls/ShellToolCall.test.tsx \
  src/components/toolcalls/ThinkToolCall.test.tsx
npm test
npm run typecheck
npm run lint
npm run build

cd ../sdk-typescript
npm run build

cd ../web-shell
npm run lint
npm run typecheck
npm run test:e2e:smoke

Results

• Failure-first run: 3 long-output cases failed because their tail markers were absent; the 500-character boundary and short-thinking cases passed.
• Focused component regression run: 4 files and 9 tests passed.
• Failure-first Playwright run: both Bash and Execute failed because their output boxes expanded to roughly 5.7k pixels, leaving no internal horizontal overflow to scroll.
• Post-fix Playwright run: both Bash and Execute passed in real Chromium. The output scroll width exceeds its client width in collapsed and expanded states, the enclosing message has no horizontal overflow, the toggle remains visible, and the output accepts a non-zero horizontal scroll position.
• Full web-shell Playwright smoke run: 26 tests passed, including both new WebUI layout cases.
• Full WebUI run: 31 files and 400 tests passed with Node.js 24.
• WebUI lint, typecheck, and production build passed.
• Web-shell lint and typecheck passed; the latter was run after generating the SDK declarations required by that package. All changed files pass Prettier.
• A normal local /export smoke test was not used as proof because the generated page loads the published WebUI package from unpkg rather than the worktree bundle. Component tests and the local WebUI build cover this change until a release build is available for the manual scenario.