← ClaudeAtlas

nord-execlisted

Execution router with explicit mode choice — picks deterministic parallel (Workflow), sequential-persistent (/loop or ralph), and optionally goal-gated (/goal) execution based on work shape. Use for "execute this batch", "run this work-list", "apply X across N files/items", or when you want to choose between a parallel vs looping executor. Not for exploratory work needing judgment (use ralph/team/autopilot).
juliuswiener/nord-kit · ★ 0 · AI & Automation · score 62
Install: claude install-skill juliuswiener/nord-kit
# nord-exec — execution router (parallel | sequential | goal-gated) Pick the execution strategy by work shape. Do NOT default-spawn ad-hoc; choose one mode. ## Step 0 — (optional) goal-gate with `/goal` If the END STATE must provably match intent (refactor must keep tests green, migration must leave zero old-API calls), first define a **`/goal`** with concrete success criteria. The native goal evaluator then verifies convergence after execution. Use when "done" is objectively checkable. ## Step 1 — classify the work, pick a mode | Work shape | Mode | Tool | |---|---|---| | Known work-list, N **independent** items (disjoint files), mechanical | **Parallel** | the Workflow below (per-item implement→verify, resumable) | | One thread, iterate-until-done, light/no state | **Sequential** | core **`/loop`** (self-paced) driving the task | | One thread, needs persistence/resume/cancel + verify loop | **Sequential-heavy** | **ralph** | | Needs coordination / shared state across workers | — | **team** (defer, not nord-exec) | | Full idea→code autonomy | — | **autopilot** (defer) | ## Step 2 — run **Parallel** → invoke the Workflow tool with the script below. Pass `args.items = [{id, task, files?}]`. Items MUST touch disjoint files (parallel agents edit the real repo; overlapping files = conflicts → run those sequentially instead). Set `args.isolate=true` only for risky same-file work (runs each in a worktree — changes do NOT auto-merge; you merge manually). **Sequential** → run