Review Loop

Run an implement → test → independent review → fix loop. The main agent only coordinates; delegate every implementation and review pass to a fresh worker.

Run independently from a goal and repository. A reviewed plan from ck-plan or another source is optional input, never a prerequisite or permission to act.

Inputs

Use user values or these defaults: goal = current conversation (ask only if absent); implementation harness/model = current; review harness/model = current; review parallelism = 1; greenfield = false; test lifecycle = parallel; max cycles = unlimited within the run budget; critic = auto (alternative: always); propose_improvements = false; fix_scope = root_cause (alternative: immediate).

Treat greenfield: true as permission to make breaking changes. Otherwise, preserve compatibility; ask before a breaking change not already authorized by the goal.

“Unwrap the onion” means fix_scope: root_cause: solve at the root so extra layers are unnecessary. If the user says that phrase, use root_cause even if immediate was set earlier.

When propose_improvements: true, collect reusable instruction changes that could prevent similar findings. These are advisory; do not modify this skill or repository instructions unless separately asked.

Before drafting

Before launching workers, read Evidence states and prompt assembly. Load other procedures when their step applies. If a reviewed plan was supplied, apply the optional handoff; derive missing or stale inputs locally without requiring another skill.