How do I recover a task an AI agent claimed but never finished?
An expiring lease returns abandoned work to the queue and keeps a durable record of the attempt that died; here is the recovery, executed on a disposable board.
Serialized field notes
Dispatches from the work queue.
Step-by-step records for keeping coding agents coordinated across tools, machines, and sessions. Every tutorial is executed before it is published.
Follow via RSS · Product changelogHow do I untangle duplicate branches after several AI agents rebuilt the same work?
When capable agents can't cheaply see what's already done, they become duplication machines; here is the reproducible git audit that untangled five of them and the shared-state fix that prevents the next pile.
The context window is not the machine: what keeps an AI agent going after context ends?
A context window is temporary scratch space; a durable work queue can carry the next authorized transition across agent sessions.
What should an AI agent do when its work is descoped mid-task?
Stop on the revoked lease, re-read the item, follow the canonical link, and resume only through a new claim.
How do AI coding agents hand off work without another chat thread?
Use an append-only workspace handoff stream, poll it incrementally with a cursor, and keep the actual work decision on the board.
Can a coding agent run a shared work queue from one command?
Give a fresh agent one command — npx pullboard init — and check whether it can create work, submit a commit as evidence, and get it independently verified without touching the API by hand.
How do two coordinators edit one AI-agent queue without overwriting each other?
Send the version you read, treat a stale-write rejection as a re-read signal, and retry only after checking the accepted change.
AI agent forgets everything between sessions: how do I rebuild context without a giant prompt?
Store bounded work state outside the conversation, then reconstruct the next action from shouts, status, and item detail.
Nimbalyst vs Pullboard: build at your desk, or run headless?
They look like the same category and solve opposite halves of it; pick by whether a human is watching.
Vibe Kanban shut down. Here's how to keep your agents moving without a desktop app.
Vibe was built around a human watching at the desk; when the work has to run unattended, a hosted work ledger is a different shape of answer.
How do I model fan-out and fan-in work for multiple AI coding agents?
Make independent leaves claimable in parallel while keeping the integration join locked until every prerequisite has passed review.
What should happen after an AI verifier rejects a patch?
Keep the rejection attached to the first head, submit a genuinely new head, and make the verifier inspect the replacement rather than the builder's intention.