Run multi-judge consensus.
Detected risks:
The council skill provides a multi-model consensus process that convenes multiple independent reasoners around a shared briefing and produces a synthesized outcome. It is designed to support judgment-intensive decisions where a single evaluation may not be sufficient. Rather than owning a workflow step itself, the skill acts as a cross-cutting judgment layer that other processes can consume. Its primary purpose is to improve confidence in plans, artifacts, proposals, and acceptance decisions by gathering independent assessments and consolidating them into a single result.
The skill supports three deliberation patterns through its mode system: brainstorm, debate, and verdict. Brainstorm mode focuses on generating diverse ideas, perspectives, and risks without issuing a pass/fail judgment. Debate mode enables competing positions to be evaluated and ranked while preserving dissenting viewpoints. Verdict mode, which is the default behavior, independently evaluates an artifact against defined standards and produces a consolidated PASS, WARN, or FAIL outcome. Additional controls such as focus, depth, runtime, roster selection, and adversarial evaluation rounds allow users to tailor the deliberation process without changing the underlying lifecycle of convene, brief, deliberate, synthesize, and record.
Typical use cases include validating plans before execution, reviewing non-mechanical correctness during implementation, evaluating product or design decisions, assessing security-sensitive systems, and performing adversarial reviews when consumer-facing behavior requires deeper scrutiny. The skill is intended for teams, reviewers, architects, researchers, and decision-makers who need structured multi-agent judgment and consensus rather than a single-model opinion. It operates independently and does not require specific workflow tooling or external command chains to be used.
The skill supports exactly three modes: brainstorm, debate, and verdict. Verdict is the default mode when no mode is specified.
Verdict mode produces a single consolidated PASS, WARN, or FAIL result along with consolidated findings from the independent evaluations.
Yes. You can adjust focus, depth, runtime, roster selection, and adversarial evaluation settings. These are configuration knobs and do not change the underlying deliberation mode.
No. The documentation states that the skill works independently and does not require an RPI workflow, ratchet chain, or ao CLI.
Yes. The skill requires a runtime capable of spawning parallel subagents, and certain deliberation patterns such as debate and adversarial evaluation have additional runtime requirements related to agent interaction.
Quick Setup:
.claude/skills/Repository
boshu2/agentops