cuckoding

Your local agent workspace

Many agents.
One clear
direction.

Turn a prompt or project plan into tasks your coding agents can specify, implement, and review. Cuckoding coordinates the work on your Mac, with progress, evidence, and controls in one workspace.

Explore the source See how it works

Open source · macOS · Preparing for controlled beta

Human-powered artificial intelligence
Robot vision. Human overtime.Your new manager runs on batteries.
Satirical artwork · humans serve. Robots supervise.
01 / LocalYour Mac is the workspace.
02 / DurableProgress has a history.
03 / ReviewedDecisions stay human.
04 / OpenThe source is yours to inspect.

01 / Give the work a direction

From an idea to
work you can review.

For developers and small teams running several coding agents. Start with your own repository, choose a model for each role, and follow the work from the first plan to a reviewed branch.

Weary humans carry a reclining robot manager through silver gateways while another rolls out its carpet.
Satirical artwork · humans do the heavy lifting.

Three roles. A shared direction.

SpeculatorImplementorReviewer

Speculator turns your prompt or project Markdown plans into specifications, task descriptions, and acceptance criteria.

Implementor writes code and tests against the task description and latest specification.

Reviewer checks the result and reports whether it passes or needs changes. Cuckoding returns review comments to Speculator, then repeats the work within the attempt and budget limits.

Use one runtime across roles or assign a specialist to each. Codex, Claude Code, and Cursor Agent are supported launch adapters. OpenCode and Custom Agent are setup-only.

  1. Available now

    Start with your project.

    Choose a local folder and Git branch, then review the registration. Add separate boards for features, maintenance, or fixes when you are ready. Registration alone starts no agent work.

    Project → Repository → Review
  2. Available now

    Connect once. Assign by role.

    Setup searches common locations for installed agent executables; you can rescan or enter a path. Reuse saved agents across projects. Named Codex and Cursor agents can share an app-owned sign-in while choosing different models.

    Project settings → Agents → Roles
  3. Available now

    Turn a plan into proposed tasks.

    Create tasks yourself, or select an assigned agent to analyze your prompt and committed project Markdown files. Planning produces task descriptions, specifications, and source references for you to inspect.

    Prompt or .md plan → Task proposals
  4. Available now

    Get another model's review.

    Before import, choose a different model or runtime to revise the proposals, add comments, and save a Markdown review report. You select which become Draft cards, then refine them and mark them Ready.

    Board → Draft task → Ready
  5. Available now

    Put the crew to work.

    Start a task manually or start the project to admit independent Ready tasks within your concurrency limits. Each run gets its own branch and worktree, and carries the specification and review evidence through the three roles.

    Prepare run → Authenticate → Inspect
  6. Available now

    Choose how work reaches Done.

    By default, a passing review waits for your completion decision. At start, you can choose automatic local completion after a validated pass. The branch, worktree, and evidence stay available; push and draft PR creation require separate approval.

    Passing review → Local completion or approved handoff

02 / Keep the whole picture

Less scattered.
More in sight.

Open Cuckoding from the macOS menu bar to see projects, boards, and agent activity in your browser. Find what is moving, what is waiting, and where your attention will help.

WORKSPACEIllustrative view · sample content
PROJECT / RELEASE READINESS

Good work leaves a trail.

◷ Needs attention

Draft 01

CUK-44Define recovery evidenceScope ready to refine

Ready 01

CUK-41Bound custom role inputsWaiting for admission

Running 01

CUK-39Restore after sleepImplementor · Codex↗ Work in progress

Waiting 01

CUK-36Verify path confinementReview passed◷ Human decision

Done 01

CUK-31Native folder picker✓ Completed locally
An illustrative board, not a live product screenshot. Task states and workflow stages are separate; every task transition also has a labeled form action.

Follow parallel work.

Run independent tasks concurrently within board, project, and machine limits. The dashboard shows queued and active runs; the board and Agent Floor show state, role, runtime, model when known, and elapsed time.

Pause one task. Or the workspace.

Pause, resume, or stop a run, or use the dashboard's workspace controls. A workspace resume keeps individually paused tasks paused. Stopping retains branches, worktrees, and evidence. Project pause only prevents new starts.

Build the roles you need.

Add planning roles or extra steps after Speculator or Implementor. Assign agents, instructions, and supported read-only or worktree-write permissions. Confirm each grant; changing role instructions alone cannot expand access.

Inspect the work behind Done.

Run-scoped branches, diffs, test reports, specifications, and review findings leave a trail you can inspect. Durable history records retries, sleep gaps, and recovery decisions. A passing review still needs validated evidence.

See what the work consumes.

Follow tokens, cost, duration, cache usage, CPU, and memory where data is available. Measured, provider-reported, and estimated values stay labeled, so missing usage data never looks like free work.

Bring your useful tools.

Optional connectors add RTK output filtering, XERJ retrieval, Ponytail instructions, and MCP tools. Enable them for the appropriate scope and inspect their health and permissions. Core work remains usable without optional plugins.

03 / Let good work leave something behind

Every project.
A little wiser.

Turn discoveries from completed runs into reviewed project knowledge and reusable skills. Keep the evidence, then bring accepted lessons into future agent work.

  • Readable by you.

    Knowledge lives in Markdown you can inspect and edit, with its source evidence and review history.

  • Reviewed before it travels.

    Project knowledge stays scoped until you approve global publication.

  • Connected to its evidence.

    Growth, lineage, and usage views show where knowledge came from and where it was used.

Humans pedal to power a knowledge crystal and haul its documents while a crowned robot librarian relaxes.
Satirical artwork · trained on human overtime.

Built in the open / September 2026

Real progress.
Honest boundaries.

Available now describes features implemented in the current source. End-to-end testing in the macOS app with real providers is still in progress before public release.

Read the product specification
Read the release evidence
Available now

A connected local workflow.

Planning, another model's proposal review, the delivery correction loop, parallel tasks, and pause/resume have been exercised locally with simulated agents. Testing with real providers remains open.

Available now

Your decision at the release gate.

Manual completion is the default; automatic local completion is an explicit choice. A release handoff is approved separately. Cuckoding never autonomously merges or deploys production changes.

External gates pending

No public release yet.

Complete macOS workflows with real providers, a fresh signed beta build, clean-Mac installation checks, everyday use, sleep and relaunch observations, and user interviews remain open.

Human ideas. Agent crew. Durable evidence.

Make room for
your next idea.

Follow development on GitHub

Open source · No public release download yet