CodeBuddy never modifies your code without your knowledge. Every change goes through the DiffReviewService, which computes line-level diffs, groups them into reviewable hunks, and lets you accept or reject changes selectively.

How it works

When CodeBuddy proposes code changes, the DiffReviewService:

1. Computes a longest common subsequence (LCS) diff between the original and proposed content
2. Groups consecutive changes into hunks with 3 lines of context around each change
3. Presents the diff in the editor using a custom TextDocumentContentProvider (scheme: codebuddy-diff)
4. Tracks each change so it can be undone later

The service maintains a history of the 50 most recent changes for undo support.

Review options

Action	Description
Accept all	Apply every proposed change across all files
Accept file	Apply changes for a specific file only
Accept hunk	Apply a single diff hunk while leaving others pending
Reject	Discard all changes
Edit	Modify the proposed changes before applying

Each hunk shows:

• Added lines in green
• Removed lines in red
• Context lines (3 above and below each change)

Auto-approve mode

For workflows where you trust CodeBuddy, enable auto-approve:

{
  "codebuddy.autoApprove": true
}

With auto-approve enabled, safe operations (file reads, searches, diagnostics) proceed automatically. Write operations still show diffs unless codebuddy.requireDiffApproval is set to false.

Inline review comments

When you run Review Selected Code or a PR review, CodeBuddy can annotate your files with inline comment threads using the Comments API.

How it works

The InlineReviewService parses the LLM’s review output and creates comment threads directly in your editor:

• Comments appear as gutter annotations with severity-colored indicators
• Each thread shows the severity label, title, and full body
• Comments are read-only (no reply widget) — they’re informational

Severity levels

Severity	Icon	Meaning
Critical	🔴	Security vulnerability, data loss risk, crash
Moderate	🟡	Logic error, performance issue, bad practice
Minor	🔵	Style issue, naming, minor improvement
Info	ℹ️	Suggestion, alternative approach

Parsing review output

The service uses a two-pass parser:

1. JSON extraction — Looks for a REVIEW_COMMENTS JSON block in the LLM output
2. Markdown fallback — If no JSON found, parses markdown headers with severity indicators (🔴, 🟡, 🔵 emojis, or “Critical Issues” / “Moderate Issues” headers)

The parser stops at non-issue sections (strengths, recommendations, optimizations) to avoid false positives.

Settings

Setting	Type	Default	Description
codebuddy.review.inlineComments	boolean	true	Enable inline comment thread annotations

Commands

Command	What it does
Clear Inline Review Comments	Removes all comment threads from all files

Thread count is capped at 500 to prevent performance degradation on large reviews.

Pull request review

The Review Pull Request command runs a comprehensive review of the current branch’s changes:

1. Change detection — Identifies modified files by comparing against the base branch
2. Branch selection — Auto-detects the base branch or prompts you to select one
3. Diff generation — Computes diffs for all changed files
4. LLM review — Sends the diff to the LLM with a review-specific prompt
5. Inline annotations — Results appear as inline comment threads (if enabled)
6. Summary — A high-level summary appears in the chat panel

The PR review prompt builder structures the diff for optimal LLM analysis, including file paths, change types (added/modified/deleted), and surrounding context.

Undo

All changes made by CodeBuddy are tracked. Undo the last set of changes with:

CodeBuddy: Undo Last Changes

Or use the editor’s built-in undo (Cmd+Z) for individual file changes. The diff service emits events (added, applied, rejected) that other services can subscribe to for audit logging.