What problem does it solve? Executing a multi-node campaign plan by hand is error-prone: workers drift into the wrong worktree, merges happen out of order, and resumed runs rely on memory instead of the git record. This Skill turns a validated campaign DAG into a runnable state machine that a composer or cheap-tier runner can drive mechanically, with every transition decided by gate exit codes rather than judgment. ## Core Features & Use Cases - Deterministic driver state machine: Lifts the orchestration protocol's pseudocode (DERIVE, RUN_LAYER, DISPATCH, AWAIT, RECONCILE, MERGE, BOUNDARY, CHECKPOINT, CLOSE) into a table where each state names the exact command and each exit code routes the next transition. - Log-first resume: Reconstructs campaign state from the flight recorder's git log and the single in-flight sketch, then boot-gates the reconstruction against contract exports and adherence audits before any dispatch. - Wired quality gates: Runs surface-honesty checks, coherence-impact over cumulative diffs, internal-ID leak sweeps, premise-freshness re-verification, merge-consent, and recorder-close gates at the exact protocol steps. - Use Case: A campaign with a validated DAG and worker IBCs needs execution — invoke the driver to derive the Kahn layer schedule, dispatch workers into isolated git worktrees per layer, reconcile and merge in node-id order, and halt at human seams for final acceptance. ## Quick Start Ask the agent to run the orchestration driver over the validated campaign DAG at .ledger/state/dag.yaml and drive dispatch and reconcile through to the merged integration branch.
