BlogDocsGitHubDownload

An unattended engineering Shift for your Mac

Wake up to pull requests for work nobody wrote down.

Your agents stop when you stop, and nothing decides what they do next. Yellowhammer picks the next Feature from your spec each Night, runs your own agent CLIs on it, and leaves one pull request per repository and a verdict on your Linear board by morning.

Download for MacView on GitHub
  • Requires macOS 26
  • Your code, credentials and models stay on your Mac.
The Yellowhammer window the morning after a Night: the header reads Night done, Feature ORB-212 verified, all 18 clauses met. Nothing needs you, and the Feature has one pull request open in each of its three repositories.The Yellowhammer window the morning after a Night: the header reads Night done, Feature ORB-212 verified, all 18 clauses met. Nothing needs you, and the Feature has one pull request open in each of its three repositories.
Example Project.

Works with

  • Orca ADE
  • Linear
  • GitHub
  • Claude Code
  • Codex
  • Antigravity

How a Night runs

One Night, start to finish

A Night is one overnight run for one of your products, from the first step to the morning report. Here is one, from the Feature it picked to the verdict it left.

The Night's account in the app: the Author, Build and Land phases of one Night on a timeline from 22:00 to 06:00, with the playhead on the Night's one problem, a failed Attempt whose Route was then excluded.The Night's account in the app: the Author, Build and Land phases of one Night on a timeline from 22:00 to 06:00, with the playhead on the Night's one problem, a failed Attempt whose Route was then excluded.
The app's account of one Night: what ran in each phase, and when. Example Project.

  1. 22:00 Pick

    It picks the work.

    Each Night it reads your specification and picks the next Feature, plus the repos it touches, even on a Night that started with an empty board. Then it writes the Work Cards: one Linear issue per piece of the Feature, per repo. The Feature carries a Definition of Done: what “finished” means, written before any code exists, every line pointing to your spec.

    A Feature Card in Linear: its status line, then the Work Cards Yellowhammer wrote for the Feature, each marked Done.A Feature Card in Linear: its status line, then the Work Cards Yellowhammer wrote for the Feature, each marked Done.

  2. 22:16 Build

    Your agents write the code.

    A failed try goes to a different model, then it stops.

    Each Work Card runs in its own Worktree on a Route: which agent CLI, model and effort level runs it. Several repos build side by side. A retry always uses a Route not yet tried on that Work Card. Retries are capped by limits you set, and a Work Card that still cannot be finished comes back to you as Blocked or Waiting on You, with the reason.

    The Pulse during a Night: three Attempts running in three repositories, on Claude Code, Codex and Antigravity Routes, and the inspector showing one Attempt's Route and Round.The Pulse during a Night: three Attempts running in three repositories, on Claude Code, Codex and Antigravity Routes, and the inspector showing one Attempt's Route and Round.
    Example Project.

  3. 05:30 Land

    One Feature, one pull request per repository.

    “Done” cites your spec, clause by clause.

    Multi-repo work comes back as one pull request per repo instead of disconnected tickets, and a pull request with work missing says so in its first lines. Verification reports on each Definition of Done clause, never a bare “passed”. The merge stays yours.

    A GitHub pull request opened by Yellowhammer, showing its Verification section: each Definition of Done clause with its spec citation, its verdict and the evidence behind it.A GitHub pull request opened by Yellowhammer, showing its Verification section: each Definition of Done clause with its spec citation, its verdict and the evidence behind it.

  4. 08:00 Morning

    A verdict on the board every morning.

    Every Night gets one Night Card, the Linear issue that carries each Night's report. A crashed Night, an empty Night and a productive Night each say what they were, so you never open a terminal to find out.

    Examples of the verdict a Night Summary leads with.

    • Night done. Nothing was ready.
    • Blocked: check failure. Answer it in Linear, not here.
    The Night Card in Linear: the Night's verdict, the Feature it selected and each Work Card with its Attempt, Route and Check.The Night Card in Linear: the Night's verdict, the Feature it selected and each Work Card with its Attempt, Route and Check.

Times are an example.

What it never does

It never writes a line of code itself, and it never merges to main.

The agent CLIs write the code. Yellowhammer judges between them: it routes each Work Card, runs the checks, and verifies the result against your specification.

  • No code of its own

    Your agent CLIs write every line. Yellowhammer composes their instructions and reads their results.

  • No merge to main

    It hands off at the pull request, one per touched repository. The merge is yours.

  • No cloud execution

    Your code, credentials and models stay on your Mac. The Night runs on your existing subscriptions.

What you need

It sits on top of the tools you already use

Yellowhammer replaces none of them. You need a Mac on macOS 26 that stays awake through the Night, the Orca ADE, a Linear workspace, GitHub, and at least one supported agent CLI.

  • Worktrees

    Orca ADE

    Creates, places and cleans up a Worktree for each Feature and repository. Yellowhammer only holds its id and path.

  • The board

    Linear

    Intent comes from your Linear board, and every report goes back to it. You don't have to be a workspace admin: setup can send an admin a link to approve the app.

  • Pull requests

    GitHub

    Receives each Feature Branch and one pull request per touched repository. Yellowhammer writes to it and never merges.

Agent CLIs

The coding agents you already use in the terminal. They write the code. Each Work Card runs on a route: which agent CLI, model and effort level runs a piece of work.

  • Claude Code

    claude

    Probed for unattended runs.

  • Codex

    codex

    Probed for unattended runs.

  • Antigravity

    agy

    Probed for unattended runs.

Adding another agent CLI takes an adapter and a probe, not a configuration edit.

Questions

The questions worth asking first

Where does my code go?

Nowhere. Your code, credentials and models stay on your Mac, and the Night runs on your existing subscriptions.

Will it push something broken to main?

It never merges. Your agents write the code, Yellowhammer opens one pull request per repo, and the merge is yours.

What if it builds the wrong thing well?

It can. Every Definition of Done clause cites your spec, which makes that auditable, not impossible.

Will it burn through my usage overnight?

It does not track cost. It stops by counting: attempts per Work Card, review rounds and unanswered Nights, all set per Project by you.

What if it needs me at 3am?

The Work Card goes to Waiting on You and you answer it in Linear whenever you like. Nothing depends on an answer during the Night.

Does it replace my tools?

No. It sits on top of Orca, Linear, GitHub and your agent CLIs, and replaces none of them.

What do I need?

macOS 26, a Mac that stays awake, the Orca ADE, Linear (a workspace admin approves the app, at your Mac or from a link you send them), GitHub, and at least one agent CLI.

Is it open source?

Yes, under the MIT licence. The code is on GitHub.