Review outcome capture (sw_review tool) — pyrite

Problem

The `review → done` and `review → in_progress` transitions exist in the workflow, but there's no tool to execute them with structured data. A reviewer (human or agent) has no way to record:

Approved or changes requested

What was good (reinforcement for future work)

What needs rework (specific, actionable feedback)

Whether Definition of Done criteria were met

Without this, review is just a status toggle. The system can't learn from review feedback, and agents repeating the same mistakes get no signal.

Solution

`sw_review` MCP tool (write tier)

`item_id`, `kb_name` — the item under review

`outcome` — `approved` or `changes_requested`

`feedback` — structured review notes

`reviewer` — who reviewed

If `approved`: transition `review → done` (or `completed`). If `changes_requested`: transition `review → in_progress` with reason (required by workflow).

CLI: `sw review --outcome approved|changes_requested`

Storage

Create a linked `review` entry (or append to work log) capturing the review outcome, feedback, and reviewer. This builds a history of reviews per item.

Integration

`sw_context_for_item` should surface prior review feedback so agents see what was rejected before and why

Review queue (`sw_review_queue`) could show items on their second+ review cycle with a "rework count"

Future: aggregate review patterns to identify recurring issues across the team

Review outcome capture (sw_review tool)backlog_item