feat: evaluate review area options by CamilleLetavernier · Pull Request #325 · eclipsesource/theia

CamilleLetavernier · 2026-07-10T07:17:02Z

What it does

Add an AI Review view in the AI First perspective (First draft, to be refined and extended)
The AI can review a change set (Currently only supporting the active Worktree, but more providers can be added)
AI Review groups changes by topic and annotates each file to explain changes one by one
AI Review can optionally take an "Intent" input describing the task being reviewed (Task Context(s), Chat Session(s) summary or free text from the user)
(Placeholder) user can mark changes as Reviewed/Needs Work/Dismissed (These actions are not wired to anything)

How to test

Open a Workspace with a dirty worktree
Open the AI First perspective (may need to explicitly open AI Review view if the perspective was persisted with an older layout)
Optionally add an intent describing the task to review
Press review
Once the review is complete, check the description of each "Area" (Topic) and navigate to the related files
The files should be annotated with comments (bullet points on the left side) describing each section

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
User-facing text is internationalized using the nls service (for details, please see the Internationalization/Localization section in the Coding Guidelines)

Reminder for reviewers

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

EclipseSourceAI

Note

Autonomous AI review. Generated automatically, so it may contain mistakes; feel free to ignore any comment you disagree with (noting why helps future reviews).

An approval here does not mean the change is free of architectural issues; the overall architecture and design should still be reviewed by a human.

To get an updated review after pushing changes, re-request a review from this account.

Adds an AI Review view to the AI First perspective: reviews the git worktree change set, groups changes into "Areas", annotates files with per-hunk editor decorations, and accepts an optional "Intent" input. Solid draft: clean provider abstraction (ReviewChangeSetProvider via ContributionProvider), reuse of @theia/scm DiffComputer for hunks, and good test coverage (tests + lint pass).

Focus areas for a human: (1) review-diff-decorator.ts leaks decorations for diff editors, since clearDecorations can't reach decorations applied via applyToMonacoDiffEditor, so re-reviews accumulate stale annotations; (2) review-summary-agent.ts registers a prompt fragment it never uses and calls the model outside the Agent framework under an unregistered agent id, so its model/prompt aren't configurable; (3) the review storage preference is unregistered and ignores workspace scope. Remaining notes are i18n gaps and a minor filename-truncation bug. Nothing here blocks the draft, but the decoration leak is worth addressing before this is refined further.

EclipseSourceAI · 2026-07-10T07:26:57Z

+            }
+        }
+        this.appliedDecorations.clear();
+    }


clearDecorations can't clear diff-editor decorations. getDecorationTarget returns undefined for MonacoDiffEditor, and applyToMonacoDiffEditor stores IDs under a diff-modified:* key that never matches decorationKey(target). So the map is emptied but the monaco decorations stay on the modified editor, and the next applyToMonacoDiffEditor reads oldDecorations: [], leaving the old ones orphaned. Result: stale/duplicated annotations on re-review with a diff open (the common path).

EclipseSourceAI · 2026-07-10T07:26:58Z

+            return undefined;
+        }
+        return editor;
+    }


This 15-line stream-of-consciousness comment describing what MonacoDiffEditor does internally should be trimmed to one line. The method just returns undefined for diff editors.

EclipseSourceAI · 2026-07-10T07:26:58Z

+        this.promptService.addBuiltInPromptFragment({
+            id: REVIEW_SUMMARY_PROMPT_ID,
+            template: reviewSummaryPromptTemplate.template,
+        });


This fragment is registered on every review but never used: buildPrompt sets systemMessage to reviewSummaryPromptTemplate.template directly, so PromptService and this call are dead. Either resolve the system prompt through the prompt service (so it becomes customizable) or drop the registration and the injection.

EclipseSourceAI · 2026-07-10T07:26:59Z

+
+    async reviewChangeSet(cs: ReviewChangeSet, intents?: ReviewIntent[]): Promise<ReviewResult> {
+        const diffs = this.collectDiffs(cs);
+        const model = await this.languageModelRegistry.selectLanguageModel({ agent: 'review-summary', purpose: 'chat', identifier: 'default/code' });


agent: 'review-summary' refers to an agent that is never registered (this service is bound only toSelf, not as Agent), so its model and prompt aren't configurable in AI settings the way the sibling pr-reviewer agent is (

theia/packages/ai-ide/src/browser/review/pr-review-agent.ts

Lines 26 to 41 in 5459778

export class PRReviewAgent extends AbstractStreamParsingChatAgent {

id = PRReviewAgentId;

name = 'PR Reviewer';

languageModelRequirements: LanguageModelRequirement[] = [{

purpose: 'chat',

identifier: 'default/code',

}];

protected defaultLanguageModelPurpose: string = 'chat';

override description = nls.localize('theia/ai/ide/prReviewAgent/description',

'An AI-powered PR review agent that orchestrates a full code review workflow: fetches PR info, explores the codebase, ' +

'performs structured analysis, interactively walks the user through findings with diff viewers, and optionally creates a pending review on GitHub.');

override iconClass: string = 'codicon codicon-git-pull-request-go-to-changes';

override prompts = [{ id: PR_REVIEW_SYSTEM_PROMPT_ID, defaultVariant: prReviewSystemPrompt, variants: [] }];

protected override systemPromptId: string = PR_REVIEW_SYSTEM_PROMPT_ID;

override tags: string[] = [...this.tags, 'Alpha'];

). Worth registering as an Agent so it fits the framework. The file is also named *-agent.ts but exports ReviewSummaryService.

EclipseSourceAI · 2026-07-10T07:27:00Z

+            return undefined;
+        }
+        const values = this.preferenceService.inspect(REVIEW_STORAGE_DIRECTORY_PREF);
+        const configuredPath = values?.globalValue ?? values?.defaultValue ?? REVIEW_STORAGE_DEFAULT_DIR;


ai-features.review.storageDirectory isn't registered in any preference schema, so it can't be discovered/set via settings, and reading globalValue ?? defaultValue skips a workspace-level value. Register it in the schema and use preferenceService.get(REVIEW_STORAGE_DIRECTORY_PREF) which resolves scope precedence.

EclipseSourceAI · 2026-07-10T07:27:01Z

+
+    protected reviewFilename(id: string): string {
+        const sanitized = id.replace(/[^\p{L}\p{N}]/ug, '-').replace(/^-+|-+$/g, '');
+        const truncated = sanitized.length > 32 ? sanitized.slice(0, sanitized.indexOf('-', 32) || 32) : sanitized;


sanitized.indexOf('-', 32) || 32 returns -1 when there is no dash past index 32, so slice(0, -1) drops the last character instead of truncating to 32. Guard the -1 case.

EclipseSourceAI · 2026-07-10T07:27:02Z

+                }));
+                changeSets.push({
+                    id: `git-worktree:${repo.provider.rootUri}`,
+                    label: 'Working tree changes',


'Working tree changes' is user-facing but not localized. Wrap with nls.localize.

EclipseSourceAI · 2026-07-10T07:27:03Z

+                <div className='ai-review-changeset-header' onClick={() => this.toggleExpand(cs.id)}>
+                    <span className={codicon(isExpanded ? 'chevron-down' : 'chevron-right')} />
+                    <span className='ai-review-changeset-label'>{cs.label}</span>
+                    <span className='ai-review-changeset-badge'>{cs.files.length} {cs.files.length === 1 ? 'file' : 'files'}</span>


'file'/'files' are hardcoded. Localize the count label (nls supports plural placeholders).

CamilleLetavernier added 13 commits July 9, 2026 15:16

feat: ai review

ad4592d

fix: decorators

f2730e4

fix: improve code hunks

64ea7a0

fix: delete stale review on refresh

1f497e7

fix: fallback for missing hunk refs

f6f94a3

fix: untracked files

ec0224f

fix: new files decorators

8ebd96d

feat: support per-hunk/file review comments

3f40198

fix: file name rendering + tooltip

8688b58

feat: make sure file comments are always visible

d927e7c

fix: avoid full rerender on updates

16afeda

feat: add support for intents

cb312ba

move the AI Review to the right

408f50a

CamilleLetavernier changed the title ~~Issues/290 reviews~~ feat: evaluate review area options Jul 10, 2026

CamilleLetavernier changed the base branch from ai-first to issues/287-sessionsList July 10, 2026 07:18

EclipseSourceAI reviewed Jul 10, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat: evaluate review area options#325

feat: evaluate review area options#325
CamilleLetavernier wants to merge 13 commits into
issues/287-sessionsListfrom
issues/290-reviews

CamilleLetavernier commented Jul 10, 2026

Uh oh!

EclipseSourceAI left a comment

Uh oh!

EclipseSourceAI Jul 10, 2026

Uh oh!

EclipseSourceAI Jul 10, 2026

Uh oh!

EclipseSourceAI Jul 10, 2026

Uh oh!

EclipseSourceAI Jul 10, 2026

Uh oh!

EclipseSourceAI Jul 10, 2026

Uh oh!

EclipseSourceAI Jul 10, 2026

Uh oh!

EclipseSourceAI Jul 10, 2026

Uh oh!

EclipseSourceAI Jul 10, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

	export class PRReviewAgent extends AbstractStreamParsingChatAgent {
	id = PRReviewAgentId;
	name = 'PR Reviewer';
	languageModelRequirements: LanguageModelRequirement[] = [{
	purpose: 'chat',
	identifier: 'default/code',
	}];
	protected defaultLanguageModelPurpose: string = 'chat';
	override description = nls.localize('theia/ai/ide/prReviewAgent/description',
	'An AI-powered PR review agent that orchestrates a full code review workflow: fetches PR info, explores the codebase, ' +
	'performs structured analysis, interactively walks the user through findings with diff viewers, and optionally creates a pending review on GitHub.');

	override iconClass: string = 'codicon codicon-git-pull-request-go-to-changes';
	override prompts = [{ id: PR_REVIEW_SYSTEM_PROMPT_ID, defaultVariant: prReviewSystemPrompt, variants: [] }];
	protected override systemPromptId: string = PR_REVIEW_SYSTEM_PROMPT_ID;
	override tags: string[] = [...this.tags, 'Alpha'];

Uh oh!

Conversation

CamilleLetavernier commented Jul 10, 2026

What it does

How to test

Follow-ups

Breaking changes

Attribution

Review checklist

Reminder for reviewers

Uh oh!

EclipseSourceAI left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants