jmb-sync
jmb-sync
Leave the board up to date after a merge: every live PR is back on top of main, the worktrees that landed disappear and the human sees in a table what is waiting for them. Orchestrate from the main checkout, on main, and nothing is committed here: every commit is made by a subagent in its worktree.
New tickets are launched by jmb-implement-frontier, not by this skill. The boundary between the two: here you touch what already has a branch; what has no branch is a launch.
Launch it with no subagents running. A live fixer and a dead one leave the same trace on the PR (draft with a pending review), and no lock file tells them apart: the one that dies does not delete it. Whoever launches the skill knows what they have running, and two subagents in the same worktree step on each other's commits.
It is idempotent: two runs in a row do not relaunch anything twice.
What is specific to the project comes from docs/agents/flow.md: each § is one of its headings, and each italic is the label of one of its fields. If the project does not have it, stop and ask for jmb-setup to be run. This skill belongs to parallel mode, which § Worktree turns on: if flow.md does not have it, stop and say that the project's flow is sequential only.