Enable Roo Code to Read VS Code Diagnostics via getDiagnostics()

[nader443]

Hi Roo Code Team 👋,

First off, thanks for building such a powerful and developer-friendly AI assistant. Roo Code is shaping up to be a fantastic tool, and I’d like to suggest a feature that could make it even more intelligent and context-aware:

---

🧠 Proposed Feature

Leverage the built-in VS Code API method:

const uri = vscode.window.activeTextEditor?.document.uri;
const diagnostics = vscode.languages.getDiagnostics(uri);

This provides an array of Diagnostic objects with error/warning messages, ranges, severity, and source (e.g., typescript, eslint, etc.).

---

💡 Why This Matters

With access to live diagnostics, Roo Code could:

• 💬 Proactively suggest fixes for ESLint/TypeScript issues
• 🛠️ Identify syntax or semantic problems without prompting
• 🧠 Use diagnostic metadata as part of its AI context (e.g., severity, location)
• ⚡ Speed up debugging and improve code quality guidance

---

✅ Benefits

• Better AI code reviews and lint-fix suggestions
• Seamless integration into existing developer workflows
• Adds a real-time debugging “copilot” vibe to Roo Code

---

🔧 Implementation Idea

The getDiagnostics() method is part of the stable VS Code Extension API, and can be safely used within any code that already has access to vscode.window or vscode.languages.

A possible flow:

1. Collect diagnostics from the active document
2. Pass the relevant messages + metadata into the AI prompt context
3. Roo Code then responds with inline suggestions or quickfixes

---

Would love to hear your thoughts — and happy to help test or iterate on this idea if it's of interest!

Cheers,

[mrubens]

Hi, thanks for the feedback! We do attempt to use it in a few places but I'm sure we could do more with it: https://github.com/search?q=repo%3ARooVetGit%2FRoo-Code%20getdiagnostics&type=code

Anything in particular that seems missing? Thank you!

[nader443]

Thanks for the response!

I wanted to highlight a specific scenario that could really benefit from deeper integration of getDiagnostics():

When I’m using the AI agent in CODE mode, I often prompt it to generate code. It runs and then tells me “Task complete” — but when I check the Problems panel, there’s still a long list of warnings that haven’t been addressed.

Right now, I have to go through those diagnostics manually, copy each message one by one, and then re-prompt the AI to fix them — which breaks the flow.

💡 Suggested Enhancement

It would be incredibly helpful if the AI agent could:
• Automatically check for diagnostics (via getDiagnostics()) right before reporting the task as “complete”
• Include those diagnostics in its decision to stop or continue working
• Optionally: summarize unresolved diagnostics in its final message so I know what’s left

This way, the AI would feel more aware of what’s still broken or needs cleanup — and could even proactively fix the remaining issues without a second prompt.

Happy to help test if this makes it into a future release!

Thanks again 🙌