How it Works
The extension automatically detects when you accept AI-generated code. It supports both Inline Suggestions (ghost text) with deterministic detection, and Agentic AI interactions (Copilot Chat, Inline Chat, Agent Mode) with confidence-based detection.
When an AI suggestion is confirmed:
- Detection: For inline suggestions, the extension intercepts the Tab key. For agentic AI, it uses a multi-signal confidence scoring system.
- Tracking: It sets a temporary flag in your local git repository.
- Git Integration: A custom
commit-msg hook is automatically installed in your repository. When you commit your changes, this hook checks for the flag and appends an "Impacted by AI" footer to your commit message.
Features
- Inline Suggestion Tracking: Deterministic, zero false positives. Tracks Tab, Ctrl+Right (word), and Ctrl+Shift+Right (line) acceptances.
- Agentic AI Detection: Confidence-based scoring system detects Agent Mode, Chat Apply, and autonomous code insertions.
- Configurable Threshold: Adjust the confidence threshold (50-100%) to balance detection vs false positives.
- Marker Consolidation: When both inline and agentic AI are used before a commit, the message shows "Impacted by AI (Inline + Agentic)".
- Git Integration: Automatically tags commits that contain AI-generated code with an "Impacted by AI" trailer.
- Privacy Focused: No code is sent to external servers. All processing happens locally within VS Code and your Git hooks.
Extension Settings
This extension contributes the following settings:
copilotInsightTracker.agenticConfidenceThreshold: Minimum confidence percentage (50-100) to mark as agentic (default: 70).
Agentic Detection Signals
The confidence scoring system uses 14 signals to detect AI-generated code.
Changes are only marked when confidence exceeds the threshold (default: 70%).
Commit Message Examples
After accepting inline AI suggestions:
feat: add user authentication
Impacted by AI
After Agent Mode changes (with confidence score):
fix: resolve memory leak
Impacted by AI (Agentic - 85% confidence)
After using both inline and agentic in the same session:
refactor: improve performance
Impacted by AI (Inline + Agentic)
Threshold Guidelines
- 50%: More detections, higher false positive risk
- 70% (default): Balanced detection, recommended for most users
- 85%+: Conservative, only marks high-confidence changes
Requirements
- VS Code 1.85 or later
- Git installed and repository initialized
