Round Table Workspace：本地优先的多智能体决策工作流框架

Round Table Workspace is a local-first multi-agent decision workflow framework designed for Codex, Claude Code, and other local CLI agents. It features two core modes—/room (exploratory discussion) and /debate (formal decision review)—to manage the full process from fuzzy problem exploration to high-stakes decision-making. Adhering to local-first philosophy and evidence-driven principles, it ensures data control, reproducible discussions, and reliable decision outputs.

In the age of AI programming assistants, a key problem arises: how to ensure orderly, deep, and reproducible collaboration among multiple AI agents. Round Table Workspace addresses this with a local-first approach—discussion data stays local, agents run locally, and users retain full control, contrasting with mainstream cloud-native AI services.

Two core commands power the workflow:

1. /room: For exploratory discussions on fuzzy/complex issues. Mechanisms include expert panel selection, structured rounds, deep questioning, and focus control. Example scenario: exploring an AI learning product for students with product strategists, architects, etc.
2. /debate: For formal high-stakes decisions (product direction, tech architecture). Features adversarial argumentation, clear decisions (allow/reject/follow-up), and traceable records. Advanced usage: /debate --with Jobs,Taleb to simulate specific experts.

Local-first architecture details:

• Data Sovereignty: All records are stored locally, no cloud dependency, Git version control supported.
• Modular Runtime: Folder structure includes prompts, agent skills, reports, etc.
• Multi-platform Adaptation: Native Codex support, Claude Code via skill discovery, and adapters for other CLI agents. Each support claim requires validation evidence.

The project emphasizes evidence over claims:

• Release Check: Generate verification reports (e.g., live_lane_evidence_report.py).
• Regression Test: Local Codex regression tests (e.g., local_codex_regression.py).
• Release Candidate Review: Strict checks (fixture runs, Git cleanliness) via release_candidate_report.py.

Workflow transition: Use /upgrade-to-debate to move from /room to formal review (e.g., from exploring an AI product to debating its viability). Quick start steps:

1. Clone repo: git clone https://github.com/MarkDonish/round-table-workspace.git
2. Health check: ./rtw doctor
3. Start /room: ./rtw room "I want to discuss a college student-oriented AI learning product, covering direction, entry points, and risks step by step"
4. Upgrade to /debate: ./rtw debate "Is this startup direction worth pursuing?"

Best Scenarios: Startup direction exploration, tech architecture selection, product priority ranking, risk assessment. Limitations: Requires local CLI agents (Codex/Claude Code), some features are stubbed, cloud integration needs extra configuration.

Round Table Workspace represents AI tools evolving from answering questions to organizing discussions. It helps teams make complex decisions via local-first, evidence-driven collaboration. Compared to Cursor Agent Workflow (focused on code workflow automation), this framework focuses on decision-making processes—both reflect AI workflow evolution towards better human-AI collaboration.