← ClaudeAtlas

dispatch-ladderlisted

Walk ONE tactic node the whole way up the dispatch phase ladder — align-tactics through main-qa, then the node-scoped reconcile pass (merge, absorb, review-stall route), to phase done — as a detached shell driver with no model turn between phases. For when the dispatch tick structurally cannot reach the node.
natb1/commons.systems · ★ 3 · AI & Automation · score 69
Install: claude install-skill natb1/commons.systems
# Dispatch Ladder Drive one node up the real dispatch ladder until it reaches phase `done` or halts. Each phase runs as a real spawned session running the real dispatch phase skill — nothing here re-implements a phase. The dispatch router normally owns this. `/dispatch-ladder` is for when it cannot reach the node — dispatch is paused and the fix is what would unpause it, or any other bootstrap deadlock the tick has no route into — or when the author wants one node driven under supervision. **Detached execution, attended judgment.** The driver runs as a transient systemd unit and outlives the session that launched it. What detaches is the **waiting**, not the judgment: every halt is unconditional — no retry, no auto-park, no resume — and stays halted until a person reads it and acts. The session's job is to launch, poll, engage a halt if there is one, and run the closing cross-phase synthesis. **Each phase is already evaluated by then** — the driver spawns `/rsi` for every phase at its own boundary, and for the phase a halted run owes — so what is left for this session is only what no single phase's evaluator can see. `/dispatch-ladder` inherits attended and pace-exempt status from whichever attended thread invokes it — an author directly. It has none of its own. ## What this is not Not a scheduler, not a selector, not a second orchestration surface. **The driver may sequence; it may never gate.** Every eligibility question belongs to a script that already owns it: - `d