Routemark

Routemark keeps the powerful model you already pay for in charge, and lets it hand execution work to a cheaper coding agent on your own machine. That model, the Controller (Claude Code or Codex), writes the task; a delegate (another coding-agent harness, such as OpenCode, with a cheaper model) does the work in a private copy; the Controller reviews the result, and only what was reviewed and passed is merged into your repository.

It is not a model, not a harness and not a service. It is a command-line tool and a set of skills your Controller reads, running on one machine. It calls no service of its own: your delegate's harness talks to its own provider under its own login. What Routemark keeps is in one directory, ~/.routemark, plus one routemark.toml file in each project you add.

What it guarantees

A project is open by default: the delegate runs as you, unsandboxed. A sandboxed project runs it inside a macOS sandbox; you make one with routemark project add DIR --containment sandboxed.

These hold because Routemark enforces them, not because a model was asked nicely:

• A delegate works in its own private copy and never writes to your real repository. On a sandboxed project it runs inside a macOS sandbox that lets it change only that copy and the private home and temporary folder Routemark makes beside it, and Routemark checks the sandbox holds on your machine before any work starts.
• What merges is exactly the tree that was judged. If your branch moved while the delegate worked, the judged change is applied on top of the branch as it is now, checked again, and only then merged.
• No credential's value is ever printed, and one the delegate brought in stops the merge. What could merge is scanned for credentials; a finding the delegate brought in holds the merge until you, in person, clear it with a reason. One that was already in your project is reported and does not block.
• Nothing merges without a review on record, unless you waive the review yourself. Waiving a review and clearing a credential finding work only when a person runs them at a terminal; a Controller is told to hand them to you.
• Checks the Controller keeps hidden stay hidden. The delegate never sees them.

On an open project the delegate is not sandboxed: it can see what you can see, your files and your network alike, and routemark project add says so. There the guarantees hold against mistakes and ordinary misbehaviour, not against a delegate deliberately attacking Routemark's own files; the sandbox is where they hold against that.

How it behaves

• A merge never overwrites work you have not committed. It stops before anything changes and names the files.
• Nothing stops a run except a person, or the budget your Controller gave it. A run that looks stuck is reported, with the facts.

What it does not do

• It does not pick models. The Controller recommends a harness, provider and model, you decide, and Routemark records what worked on this machine as advice for next time. It never permits or refuses a run because of a model.
• It does not judge providers. It ships no opinion on vendors, model families or countries.
• It does not get between you and your Controller. How far the Controller goes without asking you is agreed between the two of you; it is not a Routemark setting.

Install

The short way is to ask your Controller. Open Claude Code or Codex in the folder you want to use Routemark in, and paste this:

Help me install Routemark and use it in this folder. Install it with uv tool install routemark. Run routemark skills install and show me what it would link; link it when I say yes. Then read two of its skills, routemark-setup and routemark: routemark home prints Routemark's folder, and both are in skills/shipped/ under it. Follow the setup skill to set up a delegate and add this folder as a project. Run every command yourself and tell me what it printed. Ask me only the real choices, and ask before you install anything else.

What stays yours, in person:

• Logging in to the delegate's harness, in your own terminal (for OpenCode, opencode auth login). Routemark never sees a credential's value.
• On a Mac, if your Controller offers Routemark's launcher, typing yes to routemark launcher install yourself.
• Later, if they come up: waiving a review, and clearing a credential finding. Your Controller hands you those commands.

The session you paste this into was open before the skills were linked, so it may not list them: that is why the request names the files. A new session finds them. Codex may ask your approval for some steps, the install among them, because they reach outside its own sandbox; approving is correct.

By hand

You need macOS, Linux or Windows with WSL2, git, uv (or pipx), and Claude Code or Codex. uv brings its own Python; pipx needs Python 3.11 or later (on a Mac, the stock /usr/bin/python3 is 3.9).

uv tool install routemark

or pipx install routemark. Check it took:

routemark --version

A first delegation in ten minutes

The ten minutes leave out installing Routemark and a delegate, logging in to the delegate, the launcher if you want one, and the questions about your first project.

The one ordering rule: install Routemark, then routemark skills install, and only then open your Controller. The Controller reads a copy of the skills that skills install makes, so after installing or upgrading Routemark, run it again before you start a new Controller session. It links into a Controller it can see, so run Claude Code (or Codex) once beforehand: that is what creates ~/.claude (or ~/.codex).

Link Routemark's skills into your Controller. The first command refreshes Routemark's own copy of the skills, shows what it would link, and links nothing into your Controller; the second makes the links:

routemark skills install

routemark skills install --yes

See what is on this machine: the delegate harnesses, and the providers and models each one already has.

routemark discover

Now open Claude Code or Codex in a git project and ask it to delegate a task with Routemark, for example: "Using Routemark, delegate this to OpenCode: add a short Getting started section to README.md. Review it and merge it if it is good." If there is no delegate on the machine yet, the Controller offers to set one up with you, asks before installing anything, and has you log in to the harness yourself; Routemark never handles a credential's value.

What the Controller then runs, and what you will recognise on screen:

1. routemark delegate: prints Run 20261001T135234-a212c6. While it is going: ... and hands the task to the delegate in a private copy.
2. It waits for the delegate to finish: opencode finished in 41.2s.
3. It judges the result: The result of run ... is the tree 4461ffb3..., followed by every check it ran and the exact diff that would merge.
4. After its review, and your yes: routemark merge, which prints What merged is the tree that was judged: 4461ffb3....

Every command that starts or finishes a step prints the next one to run, filled in, except the reviewer's own harness and model, which only you know. routemark --help lists them all. The quickstart, docs/quickstart.md, is the full first run, with the output at each step.

The console

routemark console

opens a read-only page in your browser: what is waiting on you, what is running, what finished, and the track record. It changes nothing. When you open This machine it also asks your harnesses and Controllers what they are. It ends when its terminal does (Ctrl-C). routemark console --no-open opens no browser: of the two addresses it prints, paste the second, one-time address into the browser you want.

What it costs you

Keep your strongest model in charge of planning, writing the task, review and merging. Move suitable execution work to a cheaper delegate, paid for outside that model's plan. The aim is to get more useful work from the premium capacity you already pay for. Routemark reports the usage it can measure, and promises no saving: writing the task, reviewing the result and retrying after a failed review all cost your Controller something, and can use up the benefit.

routemark report and the console lead with the Controller's own usage per merged task, named for the Controller it was read from (Claude's, Codex's, or both): the tokens your Controller spent on each delegated task, from the moment it started on the task to the merge, in fresh tokens (input, output and cache writes, with cache reads shown beside and never in), by model where the session records it (Claude Code's do; Codex's are not split by model). It is worked out when you ask, from your Controller's own session records on this machine and Routemark's run records. It is never estimated: where a session is gone or cannot be read, the figure is absent and says so.

Beside it stands only something measured on this machine: the same figure over work your Controller did itself, recorded with routemark direct begin and routemark direct end, with how many tasks it rests on and the date of the latest. Nothing is netted and nothing is turned into money.

The delegate's own tokens are shown too, as the delegate's: what it spent on the merged tasks, set against nothing. The estimate your Controller gave of doing the work itself stays on the record and is not shown.

Where the data lives

Everything Routemark keeps is under one directory, ~/.routemark (set ROUTEMARK_HOME to put it elsewhere). routemark home prints where it is. In it are the run records (the track record your reports are read from), the private copies delegates work in, and the copy of the skills your Controller reads.

Outside it is only what you agree to: the routemark.toml in each project you add, the links routemark skills install --yes makes into your Controller (~/.claude/skills, ~/.codex/skills), and, if you install the launcher, its launchd entry in ~/Library/LaunchAgents and its log in ~/Library/Logs/routemark.

routemark disk

says how much each project is using.

routemark clean-up

takes back what merged and abandoned attempts left behind. A merge already reclaims its own working area; a run's record stays. routemark disk --sessions lists the harness sessions Routemark started, with the harness's own delete command for each: Routemark never deletes one.

The honest limits

• Open projects: macOS, Linux, and Windows through WSL2. On Windows the project, Routemark and the harness live on the Linux side; a Controller inside WSL2 (Claude Code, the Claude desktop app's WSL setting, Codex's command line) is measured, and one on the Windows side can drive Routemark through wsl with an approval per command but is not. Sandboxed projects: macOS today. Routemark's launcher is macOS's too: on Linux there is none yet.
• Two Controllers and two delegate harnesses have skills: Claude Code and Codex drive; OpenCode and Prime are the delegates. OpenCode runs on open and sandboxed projects, Prime on open projects only.
• A Codex Controller and a sandboxed project. Codex runs its commands in a sandbox of its own, and macOS will not start a second sandbox inside it. So a Codex Controller reaches a sandboxed project through Routemark's launcher (routemark launcher install, which shows what it creates and asks first), or by running routemark delegate outside its sandbox with your approval.
• These are beta releases, numbered 0.1.0b; routemark --version says which you have. They have been run on two Macs with Claude Code and Codex driving, OpenCode and Prime delegating, on real providers. 0.1.0 comes after three walk-throughs by a person: Claude Code on a Mac with no delegate at all, Codex on that Mac, and Claude Code on Linux.

Removing it

routemark skills remove --yes

uv tool uninstall routemark

Run routemark launcher remove first if you installed the launcher. skills remove takes away only the links; the state home is yours to delete, and routemark home says where it is.

Reporting a problem

Open an issue at https://github.com/abhayvc/routemark/issues, with what you ran and what it printed. A way round one of the guarantees above is reported privately instead: SECURITY.md says how. An id like MERGE-4 in Routemark's output names the rule it applied; quote it. Read what you paste before you send it: a terminal can hold more than Routemark printed.

Licence

What this package ships, the command line, the loop, the sandbox and the one-machine console, is free under the Apache License 2.0 (see LICENSE and NOTICE), copyright Saar Inc. The name Routemark is Saar Inc.'s mark, and the licence grants no right to it. Paid features, when there are any, will be a service for teams, reached from the console with a key; never in the command line, at install, or for safety.

From the repository

CHARTER.md says what Routemark does and what it guarantees.

uv sync

Build the wheel and prove it works when installed (the wheel and the sdist are left in dist/), and run the tests:

.venv/bin/python scripts/build_and_verify_package.py --out dist

.venv/bin/python -m pytest -q

The same suite on Linux, in containers that need only docker, on Debian and on Ubuntu 24.04: scripts/suite_linux.sh (add --platform linux/amd64 for x86).