UI Action

SWTBot Probe Scripts

Use this skill to create or update JSON probes for com.microsoft.copilot.eclipse.swtbot.test. Probes drive the workbench through SWTBot actions and run as Tycho tests; no Java is compiled per test case.

Quick start

1. Create one focused JSON script at com.microsoft.copilot.eclipse.swtbot.test/probe-scripts/<plan-slug>-<tc-id>.json.

2. Start by settling the workbench and opening the view under test:

   [
     { "action": "waitForIdle" },
     { "action": "screenshot", "id": "01-startup" },
     { "action": "showView", "idRef": "com.microsoft.copilot.eclipse.ui.chat.ChatView" },
     { "action": "waitForIdle" },
     { "action": "screenshot", "id": "02-chat-open" }
   ]
   

3. Add user-level steps only, such as click, typeIn, clearElement, pressKey, invokeCommand, waitFor, waitForMethod, assertExists, screenshot, or dumpUi. Prefer widgetId locators for widgets the plugin owns.

4. Run from the repository root:

   ./mvnw clean verify -Dprobe.script=probe-scripts/<name>.json
   

   Use root clean verify as the default. Tycho can reuse stale bundle jars when only the SWTBot module is run, making source edits appear ignored. Use xvfb-run -a on Linux headless.

5. Read com.microsoft.copilot.eclipse.swtbot.test/target/probe-results/results.json, workspace.log, screenshots/, and ui-dumps/.

If -Dprobe.script is unset, ProbeRunner skips itself so ordinary ./mvnw verify runs are unaffected.

Authoring workflow

• Convert one scenario/test case into one probe. Split unrelated behaviors so each failure screenshot has one meaning.
• Drive the UI the way a user does. invokeCommand is allowed for registered Eclipse command IDs; do not call OSGi services, private methods, or test-only shortcuts.
• Pair screenshots with machine-checked assertions. Screenshots never fail a step; assertExists, waitFor, and waitForMethod decide pass/fail.
• Wait for asynchronous UI state before acting. For chat, wait for the model-picker selected model before Send, then wait for user-turn and model-info-label.
• When a locator fails, insert dumpUi before the failure and inspect ui-dumps/*.xml for widget class names and SWTBot widget IDs.
• For unauthenticated chat probes, assert the "Sign in to GitHub" UI. Do not poll internal auth managers.

Chat send skeleton

[
  { "action": "waitForIdle" },
  { "action": "showView", "idRef": "com.microsoft.copilot.eclipse.ui.chat.ChatView" },
  { "action": "waitFor", "locator": { "by": "styledText" }, "timeoutSec": 30 },
  { "action": "clearElement", "locator": { "by": "styledText" } },
  { "action": "typeIn", "locator": { "by": "styledText" }, "text": "hello" },
  { "action": "waitForMethod", "locator": { "by": "widgetId", "value": "model-picker" },
    "method": "getSelectedItemId", "timeoutSec": 60 },
  { "action": "click", "locator": { "by": "buttonWithTooltip", "tooltip": "Send" } },
  { "action": "waitFor", "locator": { "by": "widgetId", "value": "user-turn" }, "timeoutSec": 30 },
  { "action": "waitFor", "locator": { "by": "cssClass", "value": "model-info-label" }, "timeoutSec": 120 },
  { "action": "assertExists", "locator": { "by": "widgetId", "value": "copilot-turn" } },
  { "action": "screenshot", "id": "agent-response" }
]

Pass/fail judgment

A probe passes only when Maven exits 0 and results.json has "failed": 0. On failure, inspect the failed step message, FAILED-stepNN-*.png, nearest ui-dumps/*.xml, and workspace.log.

Extending probes

The vocabulary lives in com.microsoft.copilot.eclipse.swtbot.test/src/.../probe/. Add UI-level primitives in ProbeStep, Locator, and StepExecutor, then update REFERENCE.md.

See REFERENCE.md for action/locator tables, platform notes, result queries, troubleshooting, and current stable widget IDs.