worker-critic

An agentic loop improves task refinement by orchestrating automated worker and critic agents to systematically draft, critique, and revise outputs. This iterative cycle ensures worker agents address specific critic feedback until task quality assurance standards are met.

You automate feedback integration by running up to three revision cycles where a worker agent drafts content and a critic agent evaluates it against a defined rubric. The worker then iteratively refines the draft by directly addressing the provided feedback.

The maximum number of revision cycles supported for iterative task automation is three. The worker-critic loop manages these drafting and critique rounds to ensure outputs meet the defined project configuration and phase-specific requirements.

Yes, you can use task automation to refine specific project phases like research paper sections. The worker agent drafts the section, the critic provides feedback based on a rubric, and the worker revises until the phase-specific requirements are satisfied.

Critic feedback and rubric requirements are applied by deriving a defined rubric from project configuration and phase-specific needs. The critic agent evaluates drafts against this rubric, and the worker agent must address this specific feedback in subsequent revisions.

A key limitation is the strict cap of three revision cycles for automated output vetting. If the worker agent cannot satisfy the critic's rubric-derived feedback within these three iterations, the refinement loop terminates without further automated attempts.