Context:
From @karthiknadig: The chat agent can run long-lived or error-prone terminal commands, and Copilot remains unaware of what’s happening in the terminal. This limits its ability to provide timely suggestions or detect inefficiencies.

Problem:

• Inefficient commands (e.g. redundant builds) go unnoticed.
• Repeated errors or warnings receive no guidance.
• Copilot has no visibility into command failures unless prompted.

Proposal:

Allow Copilot to (optionally) observe terminal output to:

• Detect repeated errors or warnings and suggest fixes.
• Recommend command optimizations (e.g., caching, watch mode).
• Proactively offer help when commands fail (e.g., “Want help debugging this?”).
• Reply to in terminal prompts like Y/N

Questions:

• We already have diagnostic awareness for tasks. Should we introduce dynamic problem matchers for terminals?
• How would users feel about terminal output sharing?
• How would we parse the output?