Reviewloop

Each agent only writes to its own file and reads the other's.

Flow

1. Run specdev reviewloop <phase> — lists reviewers
2. Ask user which reviewer to use
3. Run specdev reviewloop <phase> --reviewer=<name>
4. Command spawns reviewer, waits for completion
5. Reads verdict from review/{phase}-feedback.md
6. Pass → auto-approves phase, proceed to next phase
7. Fail → run specdev check-review to read findings, fix issues, write {phase}-changelog.md
8. Re-run specdev reviewloop for next round

Hard Rules

1. Never skip check-review — always read findings before the next round
2. Never argue with findings — fix what the reviewer says or escalate to the user
3. Never exceed max rounds — when max is reached, stop and defer to the user