test: add endgame test cases for 0.19.0

com.microsoft.copilot.eclipse.swtbot.test/test-plans/cls-session-persistence/cls-session-persistence.md

@@ -181,6 +181,34 @@ This feature integrates CLS (Copilot Language Server) server-side session persis
 
 ---
 
+### TC-008: Cancelled terminal tool call stays cancelled after restoration
+
+**Type:** `Regression`
+**Priority:** `P0`
+
+#### Preconditions
+- Copilot Chat is open in Agent mode
+- A workspace is open so the `run_in_terminal` tool can execute
+- A model that supports tool calls is selected
+
+#### Steps
+1. Send a message that makes Copilot use the `run_in_terminal` tool to perform a longer-running task (e.g. "run a command that takes a while, such as `ping -n 30 127.0.0.1`")
+2. While the terminal tool call is still running (spinner active on the tool status), click Cancel on the chat response
+3. Verify the terminal tool call status updates to the **cancelled** icon (no longer spinning)
+4. Click "New Chat" to start a fresh conversation
+5. Open chat history and select the original conversation to restore it
+
+#### Expected Result
+- After restoration, the terminal tool call still shows the **cancelled** icon
+- The tool call status does **not** revert to running/spinning (no rotating/ongoing indicator reappears)
+- The cancelled status is consistent between the live cancel and the restored view
+
+#### 📸 Key Screenshots
+- [ ] **After cancel** — Terminal tool call showing the cancelled icon (not spinning)
+- [ ] **After restore** — Same terminal tool call restored with the cancelled icon, not reverted to running
+
+---
+
 ## Screenshots Checklist
 > Consolidated list of all key screenshot moments.
 
@@ -195,3 +223,5 @@ This feature integrates CLS (Copilot Language Server) server-side session persis
 - [ ] `TC-006` Restored history with successful follow-up and no 400 error
 - [ ] `TC-007` Original conversation showing assistant reply
 - [ ] `TC-007` Restored conversation with correctly attributed messages and no duplicates
+- [ ] `TC-008` Cancelled terminal tool call showing cancelled icon after cancel
+- [ ] `TC-008` Restored terminal tool call still cancelled, not reverted to running

com.microsoft.copilot.eclipse.swtbot.test/test-plans/editor-indentation-preferences/editor-indentation-preferences.md

@@ -0,0 +1,168 @@
+# Editor Indentation Preferences for Inline Completions
+
+## Overview
+When Copilot produces an **inline completion (ghost text)**, the indentation it
+suggests (tab character vs. spaces and the tab width) should follow the
+platform's **default text editor preferences** instead of always defaulting to
+spaces with a tab size of 4.
+
+The formatting options are resolved by `FormatOptionProvider` and sent on each
+completion request as `CompletionDocument.insertSpaces` / `tabSize` by
+`CompletionProvider`. For Java and C/C++ projects the language-specific
+formatter settings continue to apply. For every other case — an unknown
+language, a file with no extension, or a file with no project — the provider
+now reads the Eclipse text editor preferences
+`org.eclipse.ui.editors/spacesForTabs` and `org.eclipse.ui.editors/tabWidth`.
+If those preferences cannot be read it falls back to the previous hardcoded
+defaults (spaces, tab width 4).
+
+> **Important:** these options affect **inline completions only**, not the
+> Agent-mode file create/edit tools. The verification must therefore be done
+> by triggering ghost-text completions in an editor, not by asking Agent mode
+> to write a file.
+
+These preferences are configured under
+**Preferences → General → Editors → Text Editors**:
+- **"Insert spaces for tabs"** maps to `spacesForTabs`.
+- **"Displayed tab width"** maps to `tabWidth`.
+
+Entry points:
+- Window → Preferences → General → Editors → Text Editors
+- Typing in a text editor to trigger an inline completion (ghost text)
+
+Not exercised:
+- Direct unit-level invocation of `FormatOptionProvider` (covered by
+  `FormatOptionProviderTests`).
+- Agent-mode file create/edit tools (they do not consult these options).
+- Java/C-specific formatter configuration screens.
+
+---
+
+## Prerequisites
+
+- Eclipse IDE with the GitHub Copilot for Eclipse plugin installed and
+  activated.
+- The user is signed in to GitHub Copilot and inline completions are enabled.
+- A workspace with at least one open project that is **neither a Java nor a
+  C/C++ project** (a General/empty project), so the editor text preferences
+  are the deciding factor.
+- Knowledge of the current values of **Preferences → General → Editors →
+  Text Editors → "Insert spaces for tabs"** and **"Displayed tab width"** so
+  they can be restored after testing.
+- Because completion content is model-influenced, prefer a prompt/context that
+  reliably forces an **indented multi-line suggestion** (e.g. an opening
+  brace/block), and inspect the **raw whitespace** of the accepted text rather
+  than relying on the displayed tab width.
+
+---
+
+## 1. Spaces configuration
+
+### TC-001: Inline completion uses spaces when "insert spaces for tabs" is on
+
+**Type:** `Happy Path`
+**Priority:** `P0`
+
+#### Preconditions
+- A non-Java/non-C project is open in the workspace.
+- A file whose language is unknown / has no extension (e.g. a file named
+  `sample` with code-like content) is open in the editor.
+
+#### Steps
+1. Open **Preferences → General → Editors → Text Editors**, check
+   **"Insert spaces for tabs"**, set **"Displayed tab width"** to `2`, then
+   click **Apply and Close**.
+2. In the editor, type content that forces an indented multi-line
+   continuation, for example an opening block and a newline:
+   `function greet() {` then press Enter so Copilot suggests the indented
+   body as ghost text.
+3. Accept the inline completion (Tab).
+4. Reveal whitespace ("Show whitespace characters") or inspect the raw bytes
+   of the inserted lines.
+
+#### Expected Result
+- The accepted completion indents the body with **spaces** (not tabs), at
+  **2 spaces** per level, matching the configured tab width.
+- No error dialog appears and the Eclipse error log has no formatting
+  exception.
+
+#### 📸 Key Screenshots
+- [ ] **Editor preferences** — "Insert spaces for tabs" checked, tab width 2.
+- [ ] **Accepted completion whitespace** — Indentation shown as 2-space groups.
+
+---
+
+## 2. Tabs configuration
+
+### TC-002: Inline completion uses tabs, and switching the preference updates the next request
+
+**Type:** `Happy Path`
+**Priority:** `P0`
+
+#### Preconditions
+- A non-Java/non-C project is open in the workspace.
+- An unknown-language / extensionless file is open in the editor.
+
+#### Steps
+1. Open **Preferences → General → Editors → Text Editors**, **uncheck**
+   "Insert spaces for tabs", set **"Displayed tab width"** to `4`, then click
+   **Apply and Close**.
+2. In the editor, trigger an indented multi-line completion (as in TC-001) and
+   accept it. Reveal whitespace / inspect raw bytes.
+3. Change editor preferences to **spaces**, tab width `2`. Apply and close.
+4. Trigger another indented completion in the same or a new unknown-language
+   file and accept it. Reveal whitespace.
+
+#### Expected Result
+- The first accepted completion indents with **tab characters** (`\t`).
+- After switching the preference, the next completion indents with **2
+  spaces**.
+- The provider reads the current editor preference value on **each completion
+  request** rather than caching the first observed value.
+- No error dialog appears and the Eclipse error log has no formatting
+  exception.
+
+#### 📸 Key Screenshots
+- [ ] **Tabs completion** — Tab indentation under the tabs preference.
+- [ ] **Spaces completion** — 2-space indentation after switching the preference.
+
+---
+
+## 3. Regression — language-specific projects are unaffected
+
+### TC-003: Java completion still uses the Java formatter settings, not editor preferences
+
+**Type:** `Regression`
+**Priority:** `P0`
+
+#### Preconditions
+- A **Java** project is open in the workspace with its own
+  formatter/indentation settings (e.g., tabs for Java).
+- Editor preferences (General → Text Editors) are set to a **different**
+  value than the Java formatter (e.g., spaces, tab width 2).
+- A `.java` source file is open in the editor.
+
+#### Steps
+1. In the Java file, trigger an indented multi-line inline completion (e.g.
+   inside a method body) and accept it.
+2. Reveal whitespace / inspect raw bytes of the inserted lines.
+
+#### Expected Result
+- The Java completion follows the **Java formatter** indentation settings, not
+  the generic text editor preferences.
+- Only unknown/extensionless/projectless cases consult the editor
+  preferences; Java (and C/C++) behavior is unchanged.
+
+#### 📸 Key Screenshots
+- [ ] **Java completion whitespace** — Indentation matches the Java formatter,
+      independent of the General text-editor preference.
+
+---
+
+## Screenshots Checklist
+
+- [ ] TC-001 — Editor preferences (spaces, width 2) + accepted completion with
+      2-space indentation.
+- [ ] TC-002 — Tabs completion vs. spaces completion after a preference switch.
+- [ ] TC-003 — Java completion indentation unchanged by the General
+      text-editor preference.

com.microsoft.copilot.eclipse.swtbot.test/test-plans/file-system/local-file-edit-and-create-tools.md

@@ -219,3 +219,68 @@ Not exercised:
 #### Key Screenshots
 - [ ] **Local file tool link** -- The tool result shows a clickable absolute path outside the workspace.
 - [ ] **External local file editor** -- The external local file opens in an Eclipse editor.
+
+### TC-007: Workspace directory and project links reveal in the Project Explorer
+
+**Type:** `Happy Path`
+**Priority:** `P1`
+
+#### Preconditions
+- The Eclipse workbench is open with at least one project (e.g., `demo`) that contains a sub-folder (e.g., `src`).
+- The **Project Explorer** view is open.
+- Copilot Chat is open in Agent mode.
+
+#### Steps
+1. Send a prompt that causes Agent mode to reference the workspace sub-folder by path (for example, ask it to list the
+   contents of the `src` folder so the tool result renders a directory link).
+2. When the tool call appears in the Chat view, click the directory link for the `src` folder.
+3. Verify the `src` folder is selected and revealed in the **Project Explorer** (no external browser opens).
+4. Send a prompt that causes Agent mode to reference the project root, then click the project link in the tool result.
+5. Verify the project is selected and revealed in the **Project Explorer**.
+
+#### Expected Result
+- Clicking a workspace folder or project link reveals and selects that resource in the Project Explorer.
+- No external browser or web page is opened for directory/project links.
+- No error dialog is shown and the Eclipse error log has no navigation exception.
+
+#### Key Screenshots
+- [ ] **Directory tool link** -- The tool result shows a clickable workspace folder link.
+- [ ] **Folder revealed** -- The `src` folder is selected and revealed in the Project Explorer.
+- [ ] **Project revealed** -- The project root is selected and revealed in the Project Explorer.
+
+#### Notes on failure modes
+- Clicking the directory link opens a browser or does nothing -- the folder/project branch may not route through
+  `UiUtils.revealInExplorer`.
+- The resource is opened in an editor instead of being revealed -- folder/project types may be incorrectly treated as
+  files.
+
+### TC-008: File URI link with a line-number fragment opens the external local file
+
+**Type:** `Edge Case`
+**Priority:** `P2`
+
+#### Preconditions
+- The Eclipse workbench is open.
+- Copilot Chat is open in Agent mode.
+- The local test directory outside the workspace exists and contains `existing-local-file.txt` with multiple lines.
+
+#### Steps
+1. Send a prompt that causes Agent mode to reference `existing-local-file.txt` by absolute path with a line-number
+   fragment (for example, a link ending in `#L10` or a `file:` URI with a `#L10` fragment).
+2. When the tool call appears in the Chat view, click the file path link.
+3. Verify Eclipse opens `existing-local-file.txt` in an editor (the trailing line-number fragment is treated as a
+   fragment, not part of the file name).
+
+#### Expected Result
+- The line-number fragment is stripped when resolving the local path, so the correct external file opens.
+- Eclipse does not report a missing file named `existing-local-file.txt#L10` or a path-resolution error.
+- No error dialog is shown and the Eclipse error log has no local file navigation exception.
+
+#### Key Screenshots
+- [ ] **Fragment file link** -- The tool result shows a clickable local path with a `#L10` fragment.
+- [ ] **External file opened** -- The external local file opens in an Eclipse editor despite the fragment.
+
+#### Notes on failure modes
+- Eclipse reports the file cannot be found -- the `#` fragment may not be stripped before resolving the local path.
+- A relative-path or `https:` link is unexpectedly opened as a local file -- the URI-scheme guard in
+  `FileUtils.getLocalFilePath` may be bypassed.

com.microsoft.copilot.eclipse.swtbot.test/test-plans/subagent/subagent.md

@@ -15,6 +15,7 @@ Entry points exercised:
 - Agent mode with a prompt that triggers `run_subagent`.
 - Chat history panel: switching to a different conversation and back.
 - Cancel (stop) button during subagent execution.
+- Cancel button on a subagent tool-confirmation dialog.
 - `conversation/destroy` LSP call on session switch.
 
 Not exercised:
@@ -222,3 +223,47 @@ Not exercised:
 
 #### Key Screenshots
 - [ ] **After restore** — conversation unchanged, no extra messages.
+
+---
+
+## 6. Tool-confirmation dialog cancel layout
+
+### TC-007: Subagent panel collapses after confirmation dialog is cancelled
+
+**Type:** `Regression`
+**Priority:** `P1`
+
+#### Preconditions
+- The Chat view is open in Agent mode.
+- A new conversation (fresh session).
+- The selected agent triggers a tool that requires confirmation
+  (e.g., a command that needs to run in the terminal).
+
+#### Steps
+1. Send a prompt that triggers subagent execution where the subagent
+   wants to run a command (e.g., `scan for CVEs`, or
+   `run ls in the terminal`).
+2. Wait for the subagent's tool **confirmation dialog** to appear inside
+   the subagent card (Allow / Skip buttons).
+3. Click **Skip** on the confirmation dialog.
+4. Observe the subagent card area after the dialog disposes.
+
+#### Expected Result
+- The subagent panel resizes/collapses to fit its remaining content.
+- **No empty/blank gap** is left at the bottom of the subagent area where
+  the confirmation dialog used to be (the bug in #169).
+- A skip label/message is shown where appropriate, and the scroller
+  minimum height is recomputed in a single layout pass.
+- The send button returns to the send state.
+
+#### Key Screenshots
+- [ ] **Confirmation dialog visible** — Allow / Skip buttons shown
+  inside the subagent card.
+- [ ] **After skip** — subagent panel collapsed with no empty area below.
+
+#### Notes on failure modes
+- Empty area left below the subagent card → `cancelConfirmation()` did not
+  trigger `requestRefreshScrollerLayout()`; the `ScrolledComposite`
+  `setMinHeight()` was never recomputed after the dialog disposal.
+- Verify the Allow path is unaffected: clicking **Allow** on a
+  separate invocation should still lay out correctly (no regression).

com.microsoft.copilot.eclipse.swtbot.test/test-plans/thinking-effort/thinking-effort.md

@@ -79,3 +79,48 @@ Entry points:
 - Model A still shows **`High`**; Model B still shows **`Low`**.
 - Each model's effort is independent — changing one did not affect the other.
 
+---
+
+## TC-003: Thinking effort descriptions are fully readable in the hover card
+
+**Type:** `Regression`
+**Priority:** `P1`
+
+Regression guard for
+[issue #233](https://github.com/microsoft/copilot-for-eclipse/issues/233) /
+[PR #234](https://github.com/microsoft/copilot-for-eclipse/pull/234). The hover
+card previously clamped its width to a fixed maximum (`LONG_POPUP_WIDTH = 300`),
+which truncated the longer per-level description text in the **Thinking effort**
+section. The fix removes that cap so the hover shell uses its natural packed
+width (only enforcing a `250` px minimum), letting every description wrap and
+render in full.
+
+#### Preconditions
+- A reasoning-capable model (e.g. Claude Sonnet 4.6) advertising multiple
+  effort levels, each with a descriptive sub-label (e.g.
+  `Low — Faster responses, less reasoning`,
+  `Medium — Balanced reasoning and speed`,
+  `High — Deeper reasoning, slower responses`).
+
+#### Steps
+1. Open the Copilot Chat view and click the model picker to open the dropdown.
+2. Hover over the reasoning-capable model to open its hover card.
+3. Locate the **Thinking effort** section and read each effort level's
+   description line in full.
+4. Visually compare the right edge of the longest description against the hover card border.
+
+#### Expected Result
+- The hover card is wide enough to display the full description for every effort
+  level — no text is clipped, cut off, or replaced with an ellipsis at the right
+  edge.
+- Each description ends with its natural last word (or wraps onto a second line)
+  rather than being truncated mid-word against the card border.
+- The hover card width is never narrower than the `250` px minimum, so short
+  content still renders cleanly.
+
+#### 📸 Key Screenshots
+- [ ] **Hover card — full descriptions** — Thinking effort section showing each
+  level's complete, untruncated description text.
+- [ ] **Right-edge close-up** — the longest description's final word fully
+  visible inside the card border (no clipping/ellipsis).
+