Execute a task-graph's ready work in parallel: sequence the collision-free batches into runtime-safe waves, show a visible preview, and drive each wave's nodes through /lfg in worktree-isolated parallel runs, merging in dependency order and checkpointing recovery state at every wave boundary. Runtime-exclusive nodes run as solo waves. Manual-invoke only. Use when a committed task-graph has multiple ready nodes you want shipped as parallel PRs. Builds on the compound-engineering plugin (ce-work / lfg) and the sibling orch-decompose, orch-next, and recovery-manifest scripts.
Decompose a large project into a committed, diffable task-graph — a markdown index plus one file per node under docs/plans/, each node tagged with its next CE stage (brainstorm/plan/work) and a model tier, embedding a ready plan when a node is settled or a brief otherwise. Manual-invoke only. Use when starting a multi-feature project you want broken into dependency-mapped, individually-drivable tasks instead of one monolithic plan. Builds on the compound-engineering plugin (ce-plan / ce-brainstorm / ce-work / lfg).
Mature a task-graph's plan-stage nodes into work-ready nodes in parallel planning waves: compute the planning frontier (nodes whose upstream DECISIONS are settled, not whose code has merged), fan ce-plan over each wave in worktree-isolated runs, gate every stage-flip on human approval, write the approved plans back into the graph, re-audit, and repeat until the frontier is exhausted. Closes the gap between orch-decompose (which leaves a plan-heavy graph) and orch-fanout (which only executes work-stage nodes). Manual-invoke only. Use when a committed task-graph has plan-stage nodes you want ripened into fan-out-ready work. Builds on the compound-engineering plugin (ce-plan / ce-brainstorm) and the sibling orch-decompose, orch-next, and orch-fanout scripts.
Read an orch-decompose task-graph, compute the ready frontier, and recommend the single highest-leverage next move — with a one-line rationale and the exact command — then fire it inline. Manual-invoke only. Use when you have a committed task-graph and want to be told what to do next instead of scanning a list. Builds on the compound-engineering plugin (ce-plan / ce-brainstorm / ce-work / lfg) and the sibling orch-decompose skill.