ActorCritic

ActorCritic pairs two agents: an actor that writes a draft and a critic that looks for problems in it, with a different prompt and ideally a different model. The actor revises until the critic’s score clears a threshold. Use it for quality-critical work such as code generation, security review or important written communication, where a second, differently-built reviewer catches mistakes a model misses in its own output.

The critic returns structured JSON ({score, issues, summary}); the actor refines below threshold.

Why a separate Self-Refine?

Self-Refine runs critic + refiner with the parent’s same model and same prompt template. ActorCritic earns its complexity only when the actor and critic have different blind spots, which usually means different models. ActorCritic requires both actor and critic Agent instances; for same-model self-critique, use Self-Refine.

Usage

from loomflow import Agent
from loomflow.team import Team
 
actor = Agent(
    "Generate the code.",
    model="claude-opus-4-7",
)
 
critic = Agent(
    "Find bugs, security issues, and edge cases. Return strict JSON: "
    '{"score": 0..1, "issues": [...], "summary": "..."}',
    model="gpt-4o",  # different model on purpose
)
 
team = Team.actor_critic(
    actor=actor,
    critic=critic,
    max_rounds=3,
    approval_threshold=0.9,
    model="claude-opus-4-7",  # coordinator
)
 
result = await team.run("Write a function that sanitizes user input.")

When to use

• Code generation: Actor generates, critic looks for bugs / security issues / missing edge cases.
• Security review: Actor drafts a policy; critic challenges it against an attacker model.
• Important written communication: Actor drafts; critic checks tone / accuracy / sensitive content.

Cost: 2–4× a single-actor ReAct run, depending on rounds. Pays off when shipping a wrong answer is more expensive than the extra calls.