01 · Grounded in your docs
Before work starts, each task names the docs it rests on, the terms it uses, and any conflicts. The agent builds from your sources, not from a guess.
$ sprint newtask “plan agent development better”
Senior-engineer judgment, built into your AI agent.
Your AI agent writes code fast. SprintBias gives it what a senior engineer brings to every job: best practice, a fixed process, and your policy. Tasks are markdown in your git repo. The agent builds and tests. You approve what ships.
Done For YouRather have it built? An expert engineer leads an AI crew to ship your app.01Why it exists
Fast code is not good judgment.
A frontier model can write a feature in minutes. Left alone, it guesses. It skips tests, forgets last week's choice, and builds what you said, not what you meant. SprintBias adds the habits a senior engineer brings to every job.
02 · One fixed process
Every task moves the same way: define, build, test, land, review. The agent follows the same steps each time, so you always know where work stands.
03 · Your policy, every time
Write your rules once. Security, style, and what must never break go with every task and every agent.
04 · Every call written down
When the agent makes a choice, it writes "Decided: X, because Y." You see each one, and you can change any of them.
02See it run
One task, start to review.
Write a task. Answer the agent's questions. Let it build and test. Then git status shows the result: a code change and a task file that moved to review/. This is a replay. It writes nothing to your project.
▪ This demo touches nothing in your project.
➜ ~/my-app $ ./sprint.sh newtask 'Reject empty password on login'
✓ docs/tasks/backlog/42-reject-empty-password-on-login.md
➜ ~/my-app $ ./sprint.sh chat 42
claude │ Empty only, or whitespace too?
you │ both — 400 with "password required".
backlog/42 → next/42 · READY
➜ ~/my-app $ ./sprint.sh work
next/42 → doing/42
+ reject empty / whitespace-only → 400 "password required"
doing/42 → review/42
➜ ~/my-app $ git status
renamed: doing/
modified: src/auth/login.py
the change is a git change.
spine: newtask → chat → work → review/
Scroll to read it all. Turn on JavaScript to watch it play.
newtask → chat → work → review/. The transcript above shows every step of the run.
03The board
Folders are status. Markdown is the work.
Each task is a markdown file in your repo. Move it with git mv and its status changes. No database. No login. You see it all in git status.
backlog/
Ideas and tasks that are not ready yet.
next/
Ready, and queued for this sprint.
doing/
An agent is working on it now.
blocked/
Needs a key, a choice, or a person.
review/
Built and tested. Waiting for you.
done/
You signed off. Git keeps the history.
Work that outlasts the chat
Close the session. The tasks are still in the repo for the next agent and the next person.
Proof before review
A task lands with a proof line: the date, the tests it ran, and a pass. You review work that already works.
Easy to leave
Remove the tool any time. Your tasks, plans, and history stay in git. There is nothing to export.
04CrewNew
A crew that ships the plan. You sign off.
A crew is a team of AI agents set up for your project. It plans the work, builds it in parallel, and learns from each bug it catches. You get finished, tested code and one approval to give.
- 01
Built from your project
./sprint.sh crew setupreads your project map, your agent files, and your standards. It gives each kind of work one owner. - 02
Runs the plan
./sprint.sh crew lead plan:5hands the lead a plan. The lead sorts the tasks into waves, sends each one to the right member, and starts the next wave as work lands. - 03
Learns as it builds
Each member builds, tests, and fixes every bug it finds before it hands work in. Each edge case it catches joins the project's Known edges, so the next task tests for it.
- 04
Hands you one sign-off
When the plan is built, the lead runs the full test suite until it passes. Then it gives you one command:
./sprint.sh promote plan:5. You review, approve, and ship.
Faster
Members work at the same time on files that do not overlap. When one task lands, the next one starts.
Less waste
Each member proves its own task with tests. The full suite runs once, at the end.
Ready to ship
Bugs are fixed where they are found, not filed for later. What reaches you is tested work that waits only on your yes.
05Capabilities
One board. Any frontier agent. Your server.
Claude Code and Grok Build both work, and you can switch for each run. Start with one agent on your laptop. Grow to a crew on a shared server. The way your work is stored stays the same.
A board that lives in git
Tasks, plans, and bugs are markdown files. The folder is the status. The agent defines, builds, and checks the work. You approve and ship.
A crew that ships the plan
Crew setup builds a team of agents from your project. The lead runs a whole plan in waves. You sign off once, on tested work.
A shared dev server
Run the crew on one server in tmux. One command lists every agent session on your laptop and your servers.
06Done For YouBooking now
Rather have it built? We'll ship it for you.
Our team builds your app with an AI crew led by an expert engineer. You get working software, and a repo your team can keep running the same way. Want to run it yourself? We also train your developers in one day.
Agents build
A crew of AI agents, each with one role, works through your backlog on Claude Code or Grok Build.
An expert leads
A senior engineer plans the work, answers the hard questions, and reviews every change before it ships.
You see everything
The board lives in your repo. Every task, choice, and change is markdown and git history that you keep.
07Get started
Add SprintBias to your project.
Clone the repo once and point setup at your app. Then you write tasks, the agent builds and tests them, and you sign off. It works with Claude Code and Grok Build.
Install
git clone https://github.com/- 01
Add it to your app
cd sprintbias && ./setup.sh ~/code/my-appputs the board in your project, not in the tool's repo. - 02
Map your project once
./sprint.sh profiletells the agent your stack, your tests, and where your docs live. - 03
Write a task, then work it
Run
./sprint.sh newtask "…", then./sprint.sh work. That is the loop. Group tasks with./sprint.sh newplanwhen you are ready.
08Questions
Common questions
What is SprintBias?
SprintBias gives your AI coding agent a senior engineer's judgment. It brings best practice, a fixed process, and your policy into every project. Tasks are markdown files in your git repo, and the folder a file sits in is its status. You run it all with one command, ./sprint.sh, and you approve what ships.
What does senior-engineer judgment mean here?
It means the habits a senior engineer brings to each task. The agent reads your docs before it starts and follows the same steps every time. It writes down each choice as "Decided: X, because Y." It tests its work and lands it with a proof line. You stay the one who signs off.
How is SprintBias different from Jira, Linear, or GitHub Issues?
Those tools track work for people. SprintBias also tells the agent how to do the work: what to read first, which steps to follow, and how to prove it is done. The board lives in your repo, so there is no account, login, or sync. When you stop using it, your tasks and history stay in git.
Why store tasks as markdown files in git?
Because AI agents already read markdown and file paths well. Plain files need no extra tools. They outlast a chat session, show up in git status and code review, and stay readable to every person and agent on the project.
Which AI coding agents work with SprintBias?
Claude Code and Grok Build are fully supported, and you can switch for each run. Other agent CLIs run through a generic profile that sends the prompt only.
What is a SprintBias crew?
A crew is a team of AI agents built from your project. Its lead takes a plan, runs the tasks in parallel waves, and fixes what testing finds. Each edge case it learns is kept for the next task. When the full suite passes, you approve the whole plan with one command: ./sprint.sh promote plan:N.
Can the SprintBias team build our app for us?
Yes. Our Done For You service builds your app with a SprintBias crew led by an expert engineer, who plans the work and reviews every change. You keep the code, the task history, and the plans. Pricing depends on scope. Ask through the contact form on the Services page.
Is SprintBias free?
Yes. The tool is MIT licensed and needs no account or database. You pay only for the AI agent you already use. Done For You builds and one-day developer training are separate, optional paid services.