- Tests DOM serializer extracts elements from:
- Regular DOM (buttons, inputs, links)
- Shadow DOM (elements inside shadow roots)
- Same-origin iframes
- Cross-origin iframes
- Tests clicking elements using tools.click(index) for each category
- Test fails if clicks don't work (currently failing - needs fix)
- Verifies 12 total interactive elements are found
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Simplified test implementations by replacing ActionModel usage with direct calls to Tools methods (e.g., navigate, click, upload_file).
- Enhanced readability and maintainability of test code by removing unnecessary model definitions.
- Utilized new helper methods for element lookup by ID, improving efficiency in finding elements during tests.
This change aligns with recent updates to the Tools API, making tests cleaner and more intuitive.
Fixes dropdown selections not persisting in Vue.js, React, and other
reactive frameworks. The issue was that only 'change' events were
dispatched, but Vue's v-model requires 'input' events to register changes.
**Root Cause:**
Dropdown selection handler only dispatched 'change' events, missing the
'input' event required by reactive frameworks.
**Solution:**
Enhanced event dispatching sequence in default_action_watchdog.py:
1. Focus element first
2. Set value and selected state
3. Dispatch 'input' event (critical for Vue v-model and React)
4. Dispatch 'change' event (standard for select elements)
5. Blur element for validation
**Tests:**
Added comprehensive integration tests in tests/ci/interaction/:
- test_dropdown_vue_submit.py: Vue.js 3 with v-model
- test_dropdown_react_submit.py: React 18 with controlled components
Both tests verify dropdown selections persist through the reactive
framework's state management and form submission works correctly.
Closes#3415🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Prevents 40 parallel runners from racing to install chromium simultaneously on cache miss.
Before: 40 runners × 2min = 80 runner-minutes wasted on first run
After: 1 runner installs (2min), then 40 runners use cached version (10s each)
Savings on cache miss: ~78 runner-minutes per workflow run
Previously, done() calls inside if/else/elif blocks would only show a warning
but still execute, potentially causing tasks to never complete if the condition
wasn't met. This changes the behavior to raise a RuntimeError, forcing the LLM
to restructure code properly.
The proper pattern is:
```python
# Validate and set result variables
if condition:
result = "success"
else:
result = "failure"
# Then call done() unconditionally
await done(result, success=True)
```
Changes:
- Upgraded warning to RuntimeError in namespace.py
- Added comprehensive test suite with 5 test cases
- Tests cover if/else/elif blocks and verify standalone done() still works
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Magnus Müller