Improve VS Code extension reliability, status, and diagnostics

[codeprakhar25]

What problem does this solve?

The VS Code extension is currently a minimal capture surface. To make AgentDiff reliable for Copilot and editor-native agents, the extension needs better UX, diagnostics, and capture semantics.

What would you like to see?

Improve the VS Code extension around reliability and observability.

Suggested enhancements:

• Add a visible AgentDiff status item:
  • enabled/disabled
  • repo initialized/not initialized
  • current agent capture status
  • last capture time/error
• Add commands:
  • AgentDiff: Show Status
  • AgentDiff: Open Logs
  • AgentDiff: Toggle Capture
  • AgentDiff: Record Context
• Improve capture reliability:
  • debounce buffer changes safely
  • distinguish save events vs accepted AI edits where possible
  • include document version/timestamps to avoid stale ranges
  • avoid recording huge generated changes without a flag
• Better logging:
  • extension output channel
  • structured debug events
  • clear errors when repo is not initialized
• Better workspace behavior:
  • multi-root workspace handling
  • WSL/remote workspace path handling
  • workspace trust behavior
• Packaging/testing:
  • extension integration tests
  • documented local install/dev loop
  • compatibility notes for Copilot/Cursor/VS Code versions

Acceptance criteria

• A user can tell from VS Code whether AgentDiff capture is active.
• Capture errors are visible without digging through hidden files.
• Multi-root and WSL paths are handled deliberately.
• The extension has test coverage for basic edit/capture flows.
• Docs explain which editor/agent events AgentDiff can and cannot reliably capture.

Alternatives considered

Keep VS Code support as a thin script bridge only. That is easier, but it leaves Copilot/editor-native attribution unreliable and hard to debug.