The Developer's Kitchen

Queue the task. Agents take it.

Queue the task. Agents do the work.

PREQSTATION is a self-hosted execution queue for developers using AI coding engines. It takes a scoped software task from intake to reviewed result, and it does not replace GitHub Issues, Linear, or Jira for backlog planning.

Open Source · Self-hosted · Multi-engine execution queue

Execution proof

A representative task record shows the queue status, assigned engine, review gate, and shipped result in one place.

Task

LP-241

Clarify landing positioning and proof path

Queue the landing copy update, execute it with Codex, then move it to review with concrete evidence attached.

Engine
Codex
Agent session
codex/landing-positioning-proof
Branch
codex/landing-proof-pass

hold branch

If review or implementation is blocked, the task branches to hold instead of sitting in a vague in-progress state.

status = blocked

Use hold only when review or implementation is blocked and the task needs a real branch out of the happy path.

📥

Inbox

09:10 scoped

Acceptance criteria and ownership were added before dispatch.

🗂️

Todo

09:18 claimed

Codex claimed the task and opened an isolated worktree.

👀

Ready

10:02 review ready

Copy, proof section, and Korean localization were submitted together.

Done

10:14 verified

Build passed, the hero was spot-checked, and the result was committed.

Review evidence

Verification

pnpm --filter @preqstation/landing build passed.

Review note

Terminology unified around agent, engine, queue, and review-ready.

Result

Execution proof shipped with guide and GitHub follow-through links.

#3953

Rate Limit API

🛠️ Working

#3954

Auth Refactor

Queued

#3955

Deploy Script

👀 Ready

The missing part is execution discipline

AI can write code. Shipping still breaks down between the idea and the verified result.

📋

Issue trackers stop at assignment

They hold backlog and discussion, but they do not execute the task.

🤖

Raw AI sessions lack workflow guardrails

No queue, no branch policy, no review state, no audit trail.

⚙️

Simple work still needs a safe workspace

Agents need isolated worktrees, repeatable steps, and reporting back.

The handoff gap burns the schedule

The delay is not writing code. It is routing, checking, and deciding what can ship.

80-95%
MVP achievable
10-40%
Production reuse rate
60-80%
Needs rewrite

Work like a kitchen.You are the chef.

🧑‍🍳 Chef's Role
Define scope
Write the task and acceptance criteria
Review results
Check what the agent produced
Approve completion
Decide merge, deploy, or follow-up
🤖 AI Line Cook's Role
Pick up scoped work
Plan inbox tasks or claim todo and hold tasks with run_state
Implement safely
Use isolated worktrees, follow deploy strategy, and report exactly what changed
Return proof
Submit notes, tests, and branch or PR info so review is based on evidence
You stay in control.

Who is it for?

Perfect for you if...
📱

Ideas on the go

Fire AI tasks from anywhere, anytime on mobile

Multi-project juggler

Vibe code across projects without context loss

🧠

Context switcher

Freelancing? No more 'What did I do here?'

🤖

Multi-engine strategist

Pick Claude, Codex, or Gemini per task

Quality-focused dev

Systematically review and verify AI output

PREQ works best when...
📋

Your backlog is already elsewhere

Keep discovery and roadmap work in GitHub Issues, Linear, or Jira, then send scoped execution work into PREQ.

👥

You want execution, not general collaboration

Use Notion, Slack, or Asana for broad team coordination. PREQ is the handoff point once a task is ready to run.

🤖

You use AI agents as workers

PREQ is strongest when Claude, Codex, or Gemini are doing implementation and a human is reviewing the result.

🔁

You have repeatable delivery flow

The queue is most useful when tasks regularly move through intake, execution, review, and follow-up.

📅

You separate planning from execution

PREQ handles execution state and evidence. It does not replace sprint planning, roadmap ownership, or portfolio tracking.

Built for developers who ship

PREQ is for developers who already know what to build and need a controlled way to hand that task to an AI agent, review the result, and decide what ships.

🛠️

Execution, not backlog triage

Keep discovery and planning in GitHub Issues, Linear, or Jira. PREQ starts when a scoped task is ready for an agent to execute.

🧠

One queue, multiple engines

Assign Claude, Codex, or Gemini per task while keeping the same queue, review gate, and audit trail.

🛡️

Agents work in controlled sandboxes

Each agent session runs in an isolated worktree and reports back with notes, checks, and branch output instead of editing your main branch directly.

📊

Review with visible proof

Status changes, work logs, verification, and branch details stay attached to the task so done means reviewed, not guessed.

Why engineers actually use it

PREQ turns a scoped task into a repeatable execution loop with clear ownership and proof.

01
📋

Intake only the work that is ready

Create a task once the scope, acceptance criteria, and owner are clear enough for an agent to execute without guesswork.

02
🔥

Dispatch the right engine

Pick Claude, Codex, or Gemini for the task, let the agent claim it, and keep queued or working as run-state overlays instead of fake workflow columns.

03
⏸️

Keep blocked work explicit

Use hold as a real branch for blocked execution so review-ready work stays separate from tasks that still need input.

04

Require proof before done

Ready means the agent returned notes, checks, and branch output. You review the evidence before the task moves to done.

Execution Workflow

Happy path goes inbox → todo → ready → done. Hold branches off when work is blocked.

📥

Inbox

inbox

🗂️

Todo

todo

👀

Ready

ready

Done

done

⏸️

Hold

blocked branch

Tasks move here when implementation or verification is blocked. They rejoin the happy path only after the blocker is cleared.

Run-state overlay

Use queued when dispatch is requested and working once an agent claims the task. They sit on top of status instead of creating fake workflow columns.

Not an issue tracker

Use GitHub Issues, Linear, or Jira to discover and triage work. PREQ starts once the task is scoped enough to execute safely.

Choose the right AI for the task

Pick the best tool, no vendor lock-in

C
Claude Code
Anthropic

Strong at complex refactoring, architecture changes, and general tasks

General tasksComplex refactoringArchitecture changes
Cx
Codex CLI
OpenAI

Strong at code review, security analysis, and architecture review

Code reviewSecurity analysisArchitecture review
G
Gemini CLI
Google

Strong at UI/UX design, documentation, and large-context processing

UI/UX designDocumentationLarge-context tasks

Free. Open Source. Forever.

Your code, your data, your server

$0
forever free
  • Full source code on GitHub
  • Deploy on your own server
  • Community-driven standards
  • Your data stays yours
  • No vendor lock-in
View on GitHub

Frequently Asked Questions

Open your kitchen today

Start with the execution proof, then use the guide to wire PREQ into your own workflow.