01 · Lean into agent bias
Shape work around what an AI agent already does well — read context, reason, converse, decide. Prefer paths that walk with those strengths instead of fighting them.
$ whoami
Build for how AI already thinks.
SprintBias is agent-friendly, file-based project management. Every feature, task, and bug is plain markdown in your repo — folders are status, markdown is the work, history is free. One board your agents and your team both see.
01The bias
Stop forcing AI to work your way.
Models already treat docs/, markdown, and git paths as high-signal context. SprintBias leans into that training distribution instead of inventing a private schema the agent has to relearn every session. Session tools help inside a chat. SprintBias is how the work stays when the chat ends.
02 · Minimize context cost
Every file, command, and help page costs context when an agent loads it. Fewer, sharper commands beat many overlapping ones. Pruning is a feature.
03 · Name in common language
Plain words that read the same to people and agents — task, chat, work, plan. Lifecycle folders backlog → next → doing → review are the affordance.
04 · Instruct positively
State the desired path as the rule. Prohibition lists hand the model a map of failure modes and no map of the work. Bias toward action.
02Live theater
Planning your coding agents already know.
Capture a stalled field claim. Sharpen it into finishable jobs. Bind a plan. Unblock. Ship. Pure theater — nothing writes to your project.
▪ Theater only — nothing writes to your project.
01 · capture
attachment never lands. file it before the claim dies.
────────────────────────────────
➜ ~/MyProject$ sprint newbug 'Field WO attachment never reaches office'
✓ docs/bugs/open/22-field-wo-attachment-never-reaches-office.md
02 · sharpen
chat the bug. ask until it's three jobs.
────────────────────────────────
➜ ~/MyProject $ sprint chat bugs
you │ Tech texts a WO photo. Office never sees it. Claim sits.
ai │ Three finishable jobs: parse, file to DMS, SMS shop when blocked.
→ 61-parse-field-wo-attachment
→ 62-file-claim-package-to-dms
→ 63-sms-shop-when-claim-blocked · vendor SMS
03 · plan
bind them. promote to next/.
────────────────────────────────
➜ ~/MyProject$ sprint newplan 'Unstick stalled warranty claim from the field' 61 62 63
✓ docs/plans/11-unstick-stalled-warranty-claim.md
➜ ~/MyProject $ sprint plan start
backlog/* → next/61 62 63 · READY
04 · unblock
notify needs a key. set once, run forever.
────────────────────────────────
➜ ~/MyProject $ sprint work 63
⏸ blocked — missing SHOP_SMS_API_KEY
.env local · Secrets Manager in prod
➜ ~/MyProject $ export SHOP_SMS_API_KEY=… && sprint work 63
blocked/63 → review/63
05 · work it
pick a member. land it in review.
────────────────────────────────
➜ ~/MyProject $ sprint plan
next/14 → doing/14
+ WO PDF validated · claim queued · shop SMS on block
doing/14 → review/14
════════════════════════════════
shipped. folders are status. you review the diff.
spine: newbug → chat → plan start → work → review
Scroll for the full spine · enable JavaScript for the live theater
Field WO stuck → three jobs → plan → review. Hit Replay if you blinked.
03The board
Folders are status. Markdown is the work.
Move a file with git mv and you change state. No database. No login. Status you can see in git status.
backlog/
Identified, not ready. Park it here until it can be finished.
next/
Queued for the current sprint. The folder is the sprint.
doing/
In progress. Keep WIP tight so agents stay focused.
blocked/
Needs a key, a decision, or a human. Unblock and resume.
review/
Done — awaiting your eyes on the diff.
done/
Shipped. History lives in git.
Work that lasts past one chat
Plain files in the repo. Close the session — the board is still there for the next agent and the next human.
One board, two audiences
Same folders, same markdown. Agents already speak paths and prose. Humans get a board they can open in any editor.
Easy exit
Remove the tool anytime. Tasks, plans, and history stay in git. No SaaS lock-in, no database export dance.
04Get started
Drop the board into your project.
Clone once. Point setup at your app path. From then on you capture work, see it, and ship it — with Claude Code or Grok Build as first-class hosts.
Install
git clone git@github.com:jnun/sprintbias.git- 01
Wire it into your app
cd sprintbias && ./setup.sh ~/code/my-app— the board lands in your project, not in the tool repo. - 02
Teach the stack once
./sprint.sh profile— so the agent knows your languages, tests, and conventions. - 03
Capture → work → review
./sprint.sh newtask "…"then./sprint.sh work. That's the loop. Group withnewplanwhen you're ready.