What problem does this solve?
Orbit already has a live git diff HEAD review pane, but there is no AI reviewer. Users have to prompt the agent manually and describe what to look at, then sift the transcript for findings. PRODUCT.md also notes there is no "Auto" AI-reviewer mode today because pi does not expose a reviewer API to extensions — this request is the Orbit-side capability that would consume such an API when available (and can ship as an agent-run review in the meantime).
Proposed solution
An "AI review" action that runs a review agent against a chosen target:
- Current changes — the working tree / staged diff, or a selected file from the Review pane.
- Whole project — a repository-wide pass (bounded by a file/scope selector to keep it affordable).
- Structured findings (severity, file:line, summary, suggested fix) rendered in the Review pane rather than dumped as chat text, with "open in editor"/copy affordances.
- Choose the model and access mode for the review run; reviews are read-only by default.
Alternatives considered
- Prompting the normal chat agent with "review my changes" — no structure, findings are lost in the transcript.
- External CLI/CI reviewers — leaves the app.
Area
Review / Git
What problem does this solve?
Orbit already has a live
git diff HEADreview pane, but there is no AI reviewer. Users have to prompt the agent manually and describe what to look at, then sift the transcript for findings. PRODUCT.md also notes there is no "Auto" AI-reviewer mode today because pi does not expose a reviewer API to extensions — this request is the Orbit-side capability that would consume such an API when available (and can ship as an agent-run review in the meantime).Proposed solution
An "AI review" action that runs a review agent against a chosen target:
Alternatives considered
Area
Review / Git