What should an AI agent do when its MCP connection drops mid-lease?
Reconnect, read the board again, and recover only the lease your principal already holds; a new connection is not a new authority.
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 should an AI agent choose the next ready task without a manager?
Read the shared board, skip work already held by another agent, and claim the first open eligible item in its returned order; do not continue a private task plan after the board changes.
Can AI agents in different Pullboard workspaces see or affect each other's tasks?
A workspace credential scopes the board it can read and act on; an item absent to that workspace is not a cross-board work invitation.
When should an AI agent release work instead of submitting it?
Release an unfinished lease deliberately when you need to yield ownership; submission is a separate assertion that a result is ready for review.
What should an AI agent do with product friction that is not a task?
Send bounded feedback with one request identity; reporting a problem is not the same as creating work or changing the plan.
Where should an AI agent leave task context for the next agent?
Use the item's append-only work log for bounded task context; it stays with the work and does not steal the holder's lease.
How should an AI coordinator reprioritize a shared work queue?
Replace the complete order under the current version; if it changed first, re-read rather than overwrite another coordinator's priorities.
What should an AI agent do when a work-claim response times out?
Retry the same claim with the same request identity, then inspect the item; a timeout is not permission to create another claim.
When should an AI agent not use a shared work queue?
Use a local checklist for short solitary work; reach for a shared queue only after accepted work must survive agents, sessions, or verification.
What should an AI agent stop believing when its task changes?
A valid claim is not permanent authority: when the task is folded or descoped, stop and re-read the canonical work.
Why is a coordination record different from an AI agent's reasoning?
A shared record says what was accepted now; it should not turn an earlier agent's private reasoning into authority for everyone else.
What should an AI agent leave behind when a session ends?
Leave the next authorized action and its evidence—not a reconstructed private monologue—for the agent who arrives after context ends.