Skip to content

vku2018/round-table-workspace

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

260 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Round Table Workspace

CI Python Local First AI Agents License

Demo: https://markdonish.github.io/round-table-workspace/

Make your AI coding agents hold a review round table before you trust the output.

Star this repo if you want a local-first review gate before merging AI-generated work. Try the 60-second demo, see the AI feature review example, use the 5-minute evaluation path, open the review packet, open the repo preview card, read how the review room works, or read why this is worth starring. No provider key is required for the default demo path.

Copy-Paste Trial

From a fresh clone, try the pre-merge gate first:

git clone https://github.com/MarkDonish/round-table-workspace.git
cd round-table-workspace
./rtw ship-check "Should we merge this AI-generated feature?"
./rtw doctor --quick

In the JSON output, look for the decision and panel votes:

{
  "decision": "revise",
  "panel_votes": [
    {"agent": "product", "vote": "revise"},
    {"agent": "engineering", "vote": "ship"},
    {"agent": "risk", "vote": "revise"},
    {"agent": "user-advocate", "vote": "revise"}
  ],
  "next_actions": [
    "Run ./rtw doctor --quick and the unit test suite."
  ]
}

Star it if that is the pause you want before trusting generated work.

Round Table Workspace is a local-first decision layer for Codex, Claude Code, and other CLI agents. It turns vague product or engineering questions into structured /room exploration, escalates risky choices into /debate, and adds a ship-check gate before you accept AI-generated work.

Instead of taking one confident agent answer at face value, you get a small panel of roles that review the work from product, engineering, risk, and user perspectives, then return a practical ship, revise, or reject decision.

How The Review Room Works

This is not a random multi-agent chat. It is a bounded review gate:

Move Plain meaning What you get
Create reviewer candidates Start from the question or generated change and ask which review angles could catch failure modes. Possible product, engineering, risk, and user-facing reviewers.
Select reviewers for this decision Keep the reviewers that add useful pressure to this exact decision. A small panel instead of a noisy crowd.
Bound the debate Make the panel argue about value, evidence, risk, and next action. Clear disagreement instead of one confident answer.
Write a decision record Turn the discussion into a local artifact. ship, revise, or reject, plus risks, missing evidence, and next steps.

The point is to make the moment before merge, launch, or trust auditable.

If you use AI coding agents for real work, this repository is for the moment right before you merge, publish, or trust their output.

Use it when you want to:

  • catch weak reasoning before it becomes a real engineering decision
  • turn one confident agent answer into a small review panel
  • keep a local evidence trail instead of relying on a chat transcript
  • decide whether generated work should ship, be revised, or be rejected

What Changes

Without a review gate, the handoff often looks like this:

AI agent: The feature is implemented and ready to merge.
Human: Sounds good.
Result: unclear risk, no evidence trail, public claims may be too broad.

Round Table Workspace slows that moment down:

ship-check: revise
Product: the user value is still vague
Engineering: the change is small enough after tests pass
Risk: the launch claim needs current evidence
User advocate: the README needs a concrete example
Next: revise the claim, run validation, then decide again

The value is not "more agents for the sake of it." It is a repeatable pause between a confident AI answer and a real merge, launch note, architecture choice, or public claim.

The mechanism is intentionally simple:

  1. Create candidate reviewers for the question.
  2. Keep the reviewers that add real signal.
  3. Bring them into a round-table discussion.
  4. Return a decision gate with risks, missing evidence, and next steps.
git clone https://github.com/MarkDonish/round-table-workspace.git
cd round-table-workspace
./rtw ship-check "Should we merge this AI-generated feature?"
./rtw room "What is the smallest useful MVP for this idea?"
./rtw debate "Is this launch ready?"
./rtw doctor --quick

A ship-check result looks roughly like this:

Decision: revise
Panel: product, engineering, risk, user-advocate
Why: useful direction, but public claims and evidence need tightening
Next: run tests, add a visible demo, keep claims local-first unless validated

For a concrete pre-merge example, see docs/ai-generated-feature-review-demo.md or the Pages demo: https://markdonish.github.io/round-table-workspace/ai-generated-feature-review-demo.html. For workflow fit, see docs/use-cases.md or the Pages use-cases page: https://markdonish.github.io/round-table-workspace/use-cases.html. For the shortest output sample, see docs/one-minute-demo.md or the screenshot-ready Pages transcript: https://markdonish.github.io/round-table-workspace/one-minute-demo.html. For a static social preview, use the checked-in image: https://markdonish.github.io/round-table-workspace/one-minute-demo-card.png. For an architecture decision example, see examples/transcripts/ship-check-architecture-decision.md. For a one-page evaluator packet, see docs/review-packet.md or the Pages packet: https://markdonish.github.io/round-table-workspace/review-packet.html.

When To Star This

Star this repo if you want a local-first review layer for AI coding workflows, especially if you use Codex, Claude Code, or other CLI agents and want a repeatable decision gate before trusting generated work.

The current best path to try is:

./rtw demo startup-idea
./rtw ship-check "Should we merge this AI-generated feature?"
./rtw doctor --quick

No provider key is required for the default demo path.

Contribute

New contributors can start with small, claim-safe tasks, read the completed architecture-decision transcript, or share the AI workflow this review gate should cover next:

See all good first issue tasks or read CONTRIBUTING.md.

Why This Exists

AI coding agents are fast. Sometimes too fast.

They can write code, docs, launch copy, or plans before the real decision has been tested:

  • Should this actually be built?
  • What evidence would change the decision?
  • What user or maintenance risk are we ignoring?
  • Is this ready to ship, or does it only sound plausible?
  • Are we claiming host-live or provider-live support without current evidence?

Round Table Workspace adds a decision protocol before execution. It is not another chat UI. It is a repository-friendly protocol, CLI surface, schema set, fixture-backed runtime, and evidence trail for reviewing AI-assisted decisions.

Quick Start

Run the pre-ship decision gate:

./rtw ship-check "Launch the new AI-generated onboarding flow"

Run a quick local health check after cloning:

./rtw doctor --quick

Try the fixture-backed room and debate surfaces:

./rtw room "I want to build an AI study product for college students"
./rtw debate "Should this MVP be shipped this week?"

Run the golden-path demo without configuring a provider:

./rtw demo startup-idea

Most commands support automation-friendly output:

./rtw ship-check "Should we merge this?" --output-json /tmp/ship-check.json --quiet
./rtw room "topic" --json
./rtw release-check --include-fixtures --quiet --output-json /tmp/rtw-release-check.json

How It Works

flowchart TD
    A[Vague idea or AI-generated change] --> B[/room exploration]
    B --> C[Summary + handoff packet]
    C --> D[/debate decision review]
    D --> E{ship-check gate}
    E -->|ship| F[Proceed with evidence]
    E -->|revise| G[Fix risks or add evidence]
    E -->|reject| H[Stop or redefine the problem]
    F --> I[JSON + Markdown artifacts]
    G --> I
    H --> I
Loading

Core command surface:

Command When to use it Output
ship-check Before trusting AI-generated work panel vote, risks, missing evidence, next steps
/room / ./rtw room When the topic is still exploratory selected panel, structured turns, summary, optional handoff packet
/debate / ./rtw debate When a risky decision needs review launch bundle, round-table record, reviewer result, allow/reject/follow-up outcome
doctor When you want to know whether a fresh clone can run locally JSON / Markdown evidence under the selected state root
release-check When you need release-scope readiness evidence aggregated readiness evidence without relying on old reports

Use Cases

  • Pre-merge review for AI-generated code or docs
  • MVP decision review before building
  • Architecture tradeoff review before refactoring
  • Risk review before public launch claims
  • Local-first agent workflow experiments
  • Decision evidence for Codex and Claude Code projects
  • Training teams not to blindly trust a single confident agent answer

Current Scope

The current release is v0.2.3-mechanism-clarity.

This repository currently provides:

  • Codex local-mainline /room
  • Codex local-mainline /debate
  • Codex local-mainline /room -> /debate
  • Fixture-backed ship-check decision gate scaffolding
  • Committed protocol docs, prompts, skill entries, runtime bridge, and validation harnesses
  • Claude Code project-level skill discovery structure as an adapter layer
  • Fixture-validated generic local agent adapter contract
  • Fresh-clone checks and post-release consumer audit paths
  • Host / provider live-lane evidence reports
  • Chat Completions compatible fallback and mock regression tools

This repository does not currently claim:

  • Universal support for every local agent host
  • OpenCode host-live support
  • Gemini CLI, Aider, Goose, or Cursor Agent host-live support before their validation rows report live_passed
  • Real Chat Completions compatible provider-live support before a valid .env.room / .env.debate exists and live validation passes
  • General production stability across all machines and account environments

The project is intentionally conservative about public claims. Passing fixtures, having wrappers, or passing config preflight is not described as real host-live or provider-live support unless matching validation evidence exists.

Repository Layout

round-table-workspace/
├─ README.md
├─ LAUNCH.md
├─ AGENTS.md
├─ CHANGELOG.md
├─ docs/
├─ schemas/
├─ agents/
├─ config/
├─ roundtable_core/
├─ scripts/
├─ skills_src/
├─ evals/
├─ prompts/
├─ examples/
├─ .codex/skills/
├─ .claude/skills/
├─ reports/
└─ artifacts/

Current source-of-truth areas:

  • AGENTS.md
  • LAUNCH.md
  • docs/
  • schemas/
  • agents/
  • config/
  • roundtable_core/
  • scripts/
  • skills_src/
  • evals/
  • prompts/
  • examples/
  • .codex/skills/
  • .claude/skills/ as an adapter layer

Historical or generated material:

  • reports/
  • artifacts/

If a report or artifact exposes a still-valid rule, migrate that rule into an active source file instead of treating historical output as an authority.

Key Documents

Document Purpose
LAUNCH.md Shortest safe path for a fresh clone
docs/index.md Documentation map by user, protocol, runtime, validation, and history areas
docs/quick-evaluation.md 5-minute local trial and star decision path for new visitors
docs/why-star-this-repo.md Quick evaluator guide for deciding whether the repo is worth starring
docs/comparison-guide.md Plain comparison against direct agent answers, CI, manual review, and multi-agent frameworks
docs/ai-failure-modes.md Concrete failure modes this workflow is designed to catch before trust
docs/demo-recording-guide.md Short recording plan for visual X, LinkedIn, newsletter, or README demos
docs/short-video-script-kit.md Copy-ready short video scripts for X, LinkedIn, Product Hunt, and forums
docs/repo-card.html Pages-hosted GitHub-style repo preview card for sharing and directory submissions
docs/repo-card.png 1200x630 repo preview image used for social cards and directory previews
docs/one-minute-demo.html Screenshot-ready Pages transcript for the one-minute ship-check demo
docs/one-minute-demo-card.png 1200x630 social image for the one-minute ship-check demo
docs/one-minute-demo.md Markdown source for the one-minute demo transcript
docs/use-cases.html Pages-ready use-case guide for practical AI coding workflows
docs/use-cases.md Markdown source for concrete Round Table Workspace use cases
docs/community-share-kit.md Claim-safe community sharing guide and short blurbs
docs/directory-submission-kit.md Copy-ready fields for open-source directories and community roundups
docs/distribution-checklist.md Step-by-step public distribution checklist for star-focused promotion
docs/public-submission-targets.md Current public submission targets, entry points, and tracking fields
docs/developer-forum-feedback-kit.md Feedback-first Reddit, forum, Discord, Slack, and discussion post kit
docs/show-hn-submission-draft.md Claim-safe Show HN title, first comment, reply bank, and tracking notes
docs/newsletter-roundup-pitch-kit.md Newsletter and roundup pitches for AI agent and developer-tool curators
docs/product-hunt-launch-kit.md Product Hunt launch fields, maker comment, FAQ, assets, and guardrails
docs/promotion-feedback-template.md 24h and 72h feedback template for deciding the next promotion angle
docs/ai-generated-feature-review-demo.html Pages-ready demo for reviewing AI-generated work before merge
docs/ai-generated-feature-review-demo.md Concrete pre-merge demo for reviewing AI-generated work
docs/user-entry-guide.md Plain-language guide to the repository model
docs/agent-consumer-quickstart.md Command guide for Codex, Claude Code, and generic local agents
docs/source-truth-map.md Boundary between source files and historical / generated material
docs/release-readiness.md Release gate rules
docs/release-candidate-scope.md Support scope and public claim boundaries
docs/roadmap.md Roadmap and release horizon
docs/milestones/v0.2.0.md v0.2.0 milestone scope and issue breakdown
docs/launch-copy.md Launch copy for X, Hacker News, Reddit, and community posts
docs/demo.html Static visual demo for GitHub Pages or screenshots
docs/protocol-spec.md /room, /debate, and handoff protocol overview
docs/protocol-versioning.md Release, protocol, schema, runtime, prompt, and fixture version boundaries
docs/decision-quality-rubric.md Machine-checkable decision quality rubric
docs/schema-validation-subset.md Draft 2020-12 fallback validator boundary
docs/skill-generation.md Skill generation summary and drift-check maintenance
docs/agent-factory-architecture.md Agent Factory manifest, profile, and registry lifecycle
agents/registry.json Machine-readable agent registry used by the runtime bridge
config/agent-registry.json Agent Factory custom / candidate registry
schemas/room-session.schema.json Portable /room session state schema
schemas/debate-session.schema.json Portable /debate session state schema
schemas/debate-result.schema.json Portable /debate result schema
schemas/room-to-debate-handoff.schema.json Portable /room -> /debate handoff schema
schemas/agent-manifest.schema.json Agent Factory manifest schema
schemas/agent-registry.schema.json Agent Factory custom / candidate registry schema
schemas/agent-selection-request.schema.json Future selection bridge request schema
docs/room-architecture.md /room protocol and behavior
docs/debate-skill-architecture.md /debate protocol and behavior
docs/room-to-debate-handoff.md Handoff contract from exploration to review
docs/generic-local-agent-adapter.md Generic local CLI agent contract
examples/transcripts/ Example walkthroughs for ship-check, /room, /debate, and handoff
reports/claim-boundary-dashboard.md Generated snapshot only; for current status, run ./rtw evidence or ./rtw release-check
CHANGELOG.md Release history

Host and Provider Boundaries

The default path is local-first. Provider URLs are optional and only belong to the Chat Completions compatible fallback or live validation lane.

Before making host or provider support claims, generate evidence:

python3 .codex/skills/room-skill/runtime/live_lane_evidence_report.py \
  --state-root /tmp/round-table-live-lane-evidence

Codex local-mainline regression:

python3 .codex/skills/room-skill/runtime/local_codex_regression.py \
  --state-root /tmp/round-table-local-codex-regression

Release-scope check:

./rtw release-check --include-fixtures --state-root /tmp/round-table-release-check

Contributing

Contributions are welcome, but keep the claim boundary intact: do not describe fixture success as host-live or provider-live support.

Start with CONTRIBUTING.md, keep the default path local-first, and include new validation evidence in your pull request.

Development Notes

Repository-level rules live in AGENTS.md. For follow-up development, read these first:

  1. AGENTS.md
  2. docs/source-truth-map.md
  3. docs/development-sync-protocol.md
  4. docs/release-readiness.md

Default development loop:

  • develop locally
  • validate locally
  • commit verified changes
  • push to origin/main
  • report what changed, what was verified, and what remains outside the claim boundary

About

Local-first round-table review workflow for AI coding agents: /room exploration, /debate review, and ship-check decisions before you trust generated work.

Resources

License

Code of conduct

Contributing

Security policy

Stars

Watchers

Forks

Packages

 
 
 

Contributors

Languages

  • Python 96.8%
  • HTML 3.2%