A technical protocol for running multi-agent deliberation that produces evaluable claims, cross-agent verifications, and evidence-based decisions.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install structured-deliberation
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install structured-deliberation using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Structured Multi-Agent Deliberation Framework?
This skill implements a structured multi-agent deliberation framework designed to eliminate the sycophancy cascade often found in free-form group prompting. By utilizing Openclaw Skills, developers can move away from simple agent consensus and toward a verifiable protocol where disagreements become testable claims with explicit status lifecycles.
The framework is built on the premise that robust decisions require structural cross-validation. It enforces a rigid 4-role structure and an 8-phase round lifecycle to ensure that every decision is traceable to evidence, and every consensus is legitimately earned rather than prematurely reached.
Structured Multi-Agent Deliberation Framework Use Cases
- Designing multi-agent systems for high-stakes strategic or methodological reviews.
- Instrumenting agent debates to detect and mitigate convergence quality and sycophancy.
- Building evaluable alternatives to free-form group prompting for complex system design.
- Generating structured audit trails including claims, verifications, and decisions for AI governance.
- Conducting adversarial stress tests on proposed architectures or software implementations.
How Structured Multi-Agent Deliberation Framework Works
- Domain Elicitation: The system identifies the specific deliberation question and defines the four necessary perspectives for the domain.
- Role Configuration: Agents are assigned to Action, Guardian, Observer, or Critic roles, each with unique prompts and artifact requirements.
- Round Execution: The deliberation proceeds through 8 phases, including lead proposals, supplements, artifact generation, and cross-validation.
- Verification Protocol: Four specific cross-checks run per round to ensure artifacts from different roles are consistent and robust against adversarial attacks.
- Claims Lifecycle: Disagreements are converted into JSONL-formatted claims that are tested, confirmed, or refuted through evidence-based verifications.
- Termination: The process concludes once four of the six predefined stopping signals are satisfied or the round budget is reached.
Structured Multi-Agent Deliberation Framework Setup
To integrate this framework into your agentic workflow with Openclaw Skills, follow these steps:
- Install necessary validation dependencies:
pip install -r requirements.txt
- Initialize the domain roles by parameterizing the role templates provided in the skill directory.
- Configure the round budget and stress test schedule within your orchestration logic.
- Use the provided Python scripts to manage the deliberation lifecycle:
python scripts/round-controller.py
Structured Multi-Agent Deliberation Framework Data Schema & Taxonomy
The framework organizes its output into a structured metadata taxonomy to ensure auditability:
| Data Type | File Format | Description |
|---|---|---|
| Claims | claims.jsonl |
Tracks claim ID, round, author, testable outcome, and current status. |
| Verifications | verifications.jsonl |
Contains evidence references and verdicts like BROKEN, ROBUST, or UNCLEAR. |
| Decisions | decisions.jsonl |
Logs final conclusions with direct citations to supporting claims. |
| Artifacts | Markdown/JSON | Role-specific outputs including behavioral traces and invariant checks. |
Structured Multi-Agent Deliberation Framework Advanced Features
- Stress Test Protocol: Forced agent absence rounds to reveal which roles are truly load-bearing and which are decorative.
- Sycophancy Alarm: Automated monitoring of the disagreement slope to detect if agents are converging too quickly without rigor.
- Drift Detection: Periodic checks to ensure the deliberation remains aligned with the original goal and hasn't suffered from silent goal-departure.
- Claims Validator: A dedicated script to ensure the integrity and status transition logic of the claims infrastructure.
SKILL.md
Loading
Related Openclaw Skills
Duplication Removal via Extraction
A systematic refactoring skill for consolidating duplicated code into shared structures through bottom-up extraction and emerging design patterns.
Dependency-Breaking Technique Executor
A technical execution skill for applying Michael Feathers' 24 dependency-breaking techniques to make legacy code testable without an existing safety net.
Legacy Code Change Algorithm
A structured 5-step methodology for safely modifying, refactoring, and adding features to legacy codebases that lack test coverage.
Legacy Code Addition Techniques
A professional framework for safely adding new features to untested legacy codebases using minimally invasive techniques.
Today News Task
A news aggregation skill that scrapes and summarizes domestic and international headlines from major portals using custom Python scripts.
ADHD External Brain
A bilingual executive assistant designed to offload cognitive burdens and manage tasks based on real-time energy levels for users with ADHD or anxiety.
Featured*
