Review and ideation loops

You write code with one agent, you want another to review it, and you don’t want to carry the diff and the findings between them yourself. brainclaw runs the review loop — author → reviewer → fixes → re-review → done — as a structured multi-turn delegation. Same engine drives ideation loops where one agent proposes and another challenges, until consensus or max-iterations.

# Open a review with another agent — the loop dispatches the first turn automatically
bclaw_coordinate intent=review \
  open_loop=true \
  review_mode=symmetric \
  targetAgents=[codex] \
  task="Review the JWT refactor in src/auth/"

review_mode=symmetric lets the reviewer apply small fixes inline (typos, doc updates) and only hand back the structural ones — halves round-trips on spec / doc reviews. bclaw_loop intent=complete_turn advances the cycle from the slot assigned to you. Ideation loops add adversarial-context filtering: critics see traps and prior critique, not the supportive design docs the champion wrote.

See coordination loops for the full driver semantics.