Continuous Feedback

Stay on top of code quality with background monitoring and historical trends.

Auto-Review on Save

Passive, always-on code quality feedback without interrupting your workflow. When enabled, every file save triggers a silent AI review in the background.

• How to enable: Command Palette -> Ollama Code Review: Toggle Auto-Review on Save, or set ollama-code-review.autoReview.enabled: true.
• Status Bar: Look for the $(eye) Auto badge in the status bar to see the current status.
• Workflow: Findings appear as inline annotations and optional notifications. You can configure the minSeverity to only be notified of critical or high issues.

Review Quality Scoring & Trends

Track the quality of your code over time with a 0–100 score derived from AI findings.

• Scoring: Points are deducted based on finding severity (Critical: -20, High: -10, etc.).
• History Panel: View your score trends, average quality, and a sortable table of past reviews.
• Trigger: Click the quality score in the status bar or use the command Ollama Code Review: Show Review Quality History.

Review Analytics Dashboard

Get a comprehensive, visual overview of your review history.

• Command: Ollama Code Review: Show Review Analytics Dashboard.
• Insights: Distribution of issues by severity and category, most-reviewed files, and model usage statistics.
• Data Portability: Export your history as CSV or JSON for external analysis.

Review Diff Cache

Avoid paying the same latency and API cost twice when you re-run an unchanged review.

• How it works: The extension hashes the final review prompt plus model, provider, endpoint, and temperature. If those inputs match a previous review, the stored structured result is reused instantly.
• What changes the cache key: Diff content, active profile, selected skills, gathered context, RAG context, model, endpoint, provider, or temperature.
• Visibility: Cached reviews show a Cached badge in the review panel's System Info area, along with when the result was first stored.
• Configuration: Use ollama-code-review.cache.enabled, cache.ttlMinutes, and cache.maxEntries to disable caching, shorten the retention window, or cap local storage.

Cached results stay local in VS Code's extension storage and still update findings, annotations, scores, history, and coverage views like a fresh review.

Review Coverage & Staleness

See which workspace files still need attention without manually digging through review history.

• Where to find it: Open the AI Review activity bar item and select the Coverage view.
• Groups: Files are organized into Never Reviewed, Stale, Reviewed With Findings, and Recently Reviewed.
• Actions: Use the inline actions to review a file or restore its last stored review. The toolbar can refresh the view or copy a Markdown coverage summary.
• Freshness: Files become stale after ollama-code-review.coverage.staleAfterDays days. The default is 14.
• Scope: Configure the tracked file set with coverage.includeGlob, coverage.excludeGlob, and coverage.maxFiles.

Coverage uses the same local review history as the score and analytics views. It works with staged diff reviews and with file, folder, and selection reviews.

Findings Persistence Across Sessions

Your review findings and inline annotations survive VS Code restarts — no need to re-run the AI.

• Automatic: On startup, the most recent review's findings are restored into the Findings Explorer and inline editor decorations.
• Age Indicator: The Findings Explorer shows a "Reviewed Xh ago" label so you know how fresh the data is.
• Opt-out: Set ollama-code-review.restoreLastReview: false to disable auto-restore on startup.
• Manual Restore: You can also restore any past review from the Review History panel — findings, decorations, and the review panel are all rebuilt without calling the AI again.

Code Health Regression Guard

Track per-file health scores over time and catch regressions before they accumulate into technical debt.

• Where to find it: Open the AI Review activity bar → Code Health view.
• Tiers: Files are grouped into Critical Health (score < 60), Needs Attention (60–79), and Healthy (80+).
• Regression Detection: After each review, the extension compares the new score against the file's previous score. A drop exceeding the threshold triggers a notification.
• Block Commits: Optionally block commits via the pre-commit guard when a regression is detected (codeHealth.blockOnRegression: true).
• Hotspot Count: Configure how many worst-scoring files are surfaced with codeHealth.hotspotCount (default: 15).
• Copy Summary: Use the toolbar button to copy a Markdown table of hotspots to the clipboard.

Code Health Settings

Setting	Default	Description
codeHealth.enabled	true	Enable per-file health tracking and regression detection
codeHealth.regressionThreshold	10	Score drop (points) that triggers a regression warning
codeHealth.blockOnRegression	false	Block commits when a file health regression is detected
codeHealth.hotspotCount	15	Number of worst-scoring files to display in the tree view

Findings Management

Efficiently organize and share the results of your reviews.

• Severity Filter: Filter the Findings Explorer to focus on specific severities (e.g., only Critical and High).
• Markdown Export: Export your findings as a Markdown checklist, grouped by file, to share with your team or track in an issue tracker.
• Findings Explorer: A dedicated tree view in the sidebar (AI Review icon) that acts as a "Problems Panel" for AI findings.
• Fix All Findings (Batch Fix): Generate AI fixes for all file-backed findings in one go. The extension processes each finding, generates a code fix, filters out overlapping edits, and presents all candidates in a preview panel where you can accept or dismiss each one.
• Ignore Finding: Dismiss a finding you've already addressed or consider a false positive. The finding is removed from the tree view and inline annotations, and the quality score is recalculated immediately.
• Restore Review from History: Re-apply findings, annotations, and the review panel from any past review in the Review History panel — without calling the AI again.