Reliable by Design · 05.17
Worktree Isolation for Swarms & Converge
How parallel agent work gets its own git worktree, so a swarm or a converge cohort never steps on another branch's changes.
Parallel agents need genuinely separate working copies
When a task fans out to several worker sessions — a swarm, or the iterations of a converge/pursue loop — letting them all edit the same working directory concurrently is a recipe for one worker's half-finished edit corrupting another's. Each swarm worker gets its own git worktree; a converge cohort shares one isolated worktree across its iterations, so the cohort's incremental edits accumulate coherently instead of colliding with unrelated parallel work.
Continue
Restarting a run from a node, a failed branch, an edited node, or a checkpoint — deterministic partial replay.
router, merge, parallel, wait, checkpoint, phase_gate, subflow, loop, stop_error, return_output — every control-flow node explained.