feat: migrate Claude Code to Theia native slash commands and modes by planger · Pull Request #16541 · eclipse-theia/theia

planger · 2025-11-05T20:32:12Z

What it does

This PR migrates Claude Code to use Theia's native slash command and mode infrastructure, replacing custom Monaco-based completion providers with platform-standard implementations.

Native Slash Command Support: Migrates from a custom Monaco completion provider to Theia's PromptService for registering slash commands
- Static commands (clear, compact, config, init, memory, review, resume) are registered at startup
- Dynamic commands are loaded from .claude/commands/*.md files with live file watching
Dynamic Command Management: Implements file system watching for .claude/commands/ directory
- Automatically registers new commands when .md files are added
- Updates commands when files are modified
- Removes commands when files are deleted
- Handles workspace changes by clearing and re-registering commands
Mode Support: Adds three operational modes to the Claude Code chat agent
- "Ask before edit" (default): Requests permission for file modifications
- "Edit automatically": Automatically applies suggested changes
- "Plan mode": Plans changes without executing them
- Integrates with Theia's native ChatAgent.modes interface

How to test

Testing Slash Commands:

Open the chat view
Type @ClaudeCode / and verify all static commands appear (clear, compact, config, init, memory, review, resume)
Create a .claude/commands/ directory in your workspace
Add a markdown file (e.g., test.md) with command description
Verify the command appears in the slash command list without restarting
Modify the file name and verify the command updates
Delete the file and verify the command is removed
Switch to a different workspace and verify commands update accordingly
Type @ClaudeCode / and verify all static commands work (clear, compact, config, init, resume, ...)
Try if custom commands work

Testing Modes:

Open the chat view and type @ClaudeCode
Locate the mode selector (should show three modes)
Test "Ask before edit" mode: Send a request that would modify files and verify permission prompts appear
Test "Edit automatically" mode: Send a request and verify changes are applied without prompting
Test "Plan mode" mode: Send a request and verify it plans without executing

Follow-ups

Breaking changes

This PR introduces breaking changes and requires careful review. If yes, the breaking changes section in the changelog has been updated.

Attribution

Review checklist

As an author, I have thoroughly tested my changes and carefully followed the review guidelines

Reminder for reviewers

As a reviewer, I agree to behave in accordance with the review guidelines

sdirix

The change works for me. Great work 👍

I left some comments about the code. All of them are nitpicks and can be ignored, handled later or be applied to this PR.

I performed all suggested tests. Generally speaking they succeeded for me, but there are some rough edges:

Changing modes

The Claude Code modes work, however changing the mode while a request is in process does not work. Even if the request is stopped because Claude Code asks for an approval, the new mode will not be respected by Claude Code until there is an actual new request from the user.

As we start in "Ask before edit" mode, the user will typically encounter an approval message pretty quick. Then they change the mode and click approve, still all further changes need to be manually improved

ExitPlanTool

Claude Code has an ExitPlanTool it tries to use to exit plan mode (after approval from the user). This fails and the user has to change the mode manually afterwards to continue

"Ask before edit" does not populate the change set

The "Ask before edit" asks for edit approval before doing any edits. However the approved edits do not show up in the changeset. Therefore there is no convenient way of directly seeing the edits of what Claude Code did compared to the "Edit automatically" mode

Custom commands request rendering

This "regression" is not caused by this PR. However it seems very weird to me that the Claude Code commands are rendered as #prompt in the request, e.g.

Mode preference?

Personally I don't like that I need to change the mode for Claude Code in every new chat session. There should be some way to avoid this, for example via a preference.

None of these are deal breakers, so we can merge as is. However please capture these remarks as follow ups in case you don't tackle them here.

sdirix · 2025-11-25T21:50:33Z

packages/ai-claude-code/src/browser/claude-code-chat-agent.ts

+        const modeId = request.request.modeId ?? 'default';
+        switch (modeId) {
+            case 'acceptEdits':
+                return 'acceptEdits';
+            case 'plan':
+                return 'plan';
+            case 'bypassPermissions':
+                return 'bypassPermissions';
+            case 'default':
+            default:
+                return 'default';
+        }


This looks a bit clunky... should we instead have an array of allowed modes and check whether the request.modeId is contained in them?

sdirix · 2025-11-25T21:52:11Z

packages/ai-claude-code/src/browser/claude-code-slash-commands-contribution.ts

+                commandAgents: [CLAUDE_CHAT_AGENT_ID]
            });
+        } catch (error) {
+            console.error(`Failed to register dynamic command ${commandName}:`, error);


We should mention claude code in this error message so it's easier to understand for readers what is going on. Bonus points for using a named logger instead.

sdirix · 2025-11-25T21:54:58Z

packages/ai-claude-code/src/browser/claude-code-slash-commands-contribution.ts

+
+    protected getWorkspaceRoot(): URI | undefined {
+        const roots = this.workspaceService.tryGetRoots();
+        return roots.length > 0 ? roots[0].resource : undefined;


We should create a follow up to properly support multi root workspaces

sdirix · 2025-11-25T22:11:27Z

packages/ai-claude-code/src/browser/claude-code-chat-agent.ts

+            case 'bypassPermissions':
+                return 'bypassPermissions';


The bypassPermissions is currently not on any code path, correct?

- Replace custom Monaco completion provider with PromptService-based commands - Add support for dynamic .claude/commands/*.md files with live file watching - Implement workspace-aware file monitoring - Add mode support in Claude Code chat agent

planger · 2025-11-26T20:46:40Z

Thanks for the great review @sdirix!

I've addressed your inline comments and rebased and tracked all follow-ups in #16334.

github-project-automation bot added this to PR Backlog Nov 5, 2025

github-project-automation bot moved this to Waiting on reviewers in PR Backlog Nov 5, 2025

planger mentioned this pull request Nov 10, 2025

Add docu on slash commands eclipse-theia/theia-website#877

Merged

ndoschek requested a review from sdirix November 10, 2025 13:14

ndoschek mentioned this pull request Nov 25, 2025

Endgame Plan for 1.67.x #16643

Closed

29 tasks

sdirix approved these changes Nov 25, 2025

View reviewed changes

github-project-automation bot moved this from Waiting on reviewers to Needs merge in PR Backlog Nov 25, 2025

planger added 2 commits November 26, 2025 21:41

review: address review feedback

e3496d1

planger force-pushed the planger/add-claude-code-modes branch from b554327 to e3496d1 Compare November 26, 2025 20:45

planger merged commit f320692 into master Nov 26, 2025
10 checks passed

github-project-automation bot moved this from Needs merge to Done in PR Backlog Nov 26, 2025

planger deleted the planger/add-claude-code-modes branch November 26, 2025 21:09

github-actions bot added this to the 1.67.0 milestone Nov 26, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: migrate Claude Code to Theia native slash commands and modes#16541

feat: migrate Claude Code to Theia native slash commands and modes#16541
planger merged 2 commits intomasterfrom
planger/add-claude-code-modes

planger commented Nov 5, 2025

Uh oh!

sdirix left a comment •

edited

Loading

Uh oh!

sdirix Nov 25, 2025

Uh oh!

sdirix Nov 25, 2025

Uh oh!

sdirix Nov 25, 2025

Uh oh!

sdirix Nov 25, 2025

Uh oh!

planger Nov 26, 2025

Uh oh!

planger commented Nov 26, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

planger commented Nov 5, 2025

What it does

How to test

Follow-ups

Breaking changes

Attribution

Review checklist

Reminder for reviewers

Uh oh!

sdirix left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Changing modes

ExitPlanTool

"Ask before edit" does not populate the change set

Custom commands request rendering

Mode preference?

Uh oh!

sdirix Nov 25, 2025

Choose a reason for hiding this comment

Uh oh!

sdirix Nov 25, 2025

Choose a reason for hiding this comment

Uh oh!

sdirix Nov 25, 2025

Choose a reason for hiding this comment

Uh oh!

sdirix Nov 25, 2025

Choose a reason for hiding this comment

Uh oh!

planger Nov 26, 2025

Choose a reason for hiding this comment

Uh oh!

planger commented Nov 26, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

sdirix left a comment •

edited

Loading