Skip to content

Code Assist: screen-aware pair-programming and learning #9

Description

@collinsadi

Problem

When working through code, debugging one's own project, learning, or practicing problems, it helps to get a concise approach or hint grounded in what is on screen and in the user's own codebase.

Proposed solution

An advanced mode that captures a user-selected screen region (ScreenCaptureKit), reads it with OCR or a multimodal model, sends the problem or code to the agent, and streams back a concise approach or hint. Provide an adjustable hint level from a Socratic nudge to a full worked approach. Add practice-set integration that tracks progress and explains thoroughly after an attempt. Add codebase awareness for real work. Intended for the user's own development, debugging, and self-directed practice, not for completing an evaluated coding assessment while concealing assistance.

Privacy considerations

Screen reading is highly sensitive; it can capture whatever is on screen. Restrict capture to a user-selected region, make active capture obvious, require explicit opt-in, and send only the captured region to the agent. Combining this with Invisible Mode is not a way to defeat oversight in evaluated settings.

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Projects

    No projects

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions