Local-first AI agent workflow

Make your AI agents argue before they ship.

Round Table Workspace adds a decision layer for Codex, Claude Code, and CLI agents: explore with /room, review with /debate, and run a ship-check before trusting AI-generated work. 

$ ./rtw ship-check "Should we merge this AI-generated feature?"

Decision: revise
Confidence: medium
Panel: product, engineering, risk, user-advocate

Why:
- useful direction, but evidence is incomplete
- public claims need tighter boundaries
- a visible demo should exist before launch

Next:
1. run tests and doctor
2. add a concrete demo transcript
3. keep claims local-first unless validated

/room

Explore ambiguous ideas with a structured multi-agent discussion.

/debate

Escalate higher-stakes choices into a formal decision review.

ship-check

Get a ship / revise / reject gate for AI-generated work.

evidence

Keep JSON and Markdown artifacts for auditability.

Decision: revise

Not another chat UI.

The goal is to make AI-assisted decisions reviewable: source docs, schemas, prompts, fixtures, runtime scripts, tests, and claim boundaries live in the repo.

  • local-first by default
  • fixture-backed demo without provider setup
  • no host-live/provider-live claims without evidence
Productrevise
Engineeringship after tests
Riskrevise claims
User advocateadd demo