execute-dag-nodelisted

# Execute one DAGrail node Operate only the stable Role and Node named by the work package. If neither is named, do not guess a Role or acquire a lease; ask the DAG controller for an assignment. First verify that the DAGrail MCP tools are callable in this process. A visible skill can outlive an older harness process and does not prove MCP activation. If tools are absent, run `dagrail doctor install`, start a fresh session, or use `dagrail context`, `dagrail inspect`, `dagrail action apply`, `dagrail reconcile`, and `dagrail pre-wait` as appropriate. Never replace a missing tool with a hand-authored lifecycle envelope. 1. Bind that Role to this harness session. Never reuse a session bound to another formal Role. Use takeover only after the former lease expires. 2. Call `dag_context` with `view: worker`, `role_id`, and `node_id`. Treat its cursor, checkpoint, resource refs, and `allowedActions` as authority. Follow an opaque ref with `dag_inspect` only when the bounded package is insufficient. 3. Apply only the current `allowedActions[].ref` with a stable idempotency key. Never construct a transition, outcome envelope, hash, or resource ID yourself. 4. Start with `node.start` or continue the current Attempt. After material progress and before yielding, apply `attempt.checkpoint` with a replacement-ready summary and digest-only evidence refs. Exclude prompts, secrets, transcripts, and artifact bodies. 5. Publish material execution through `evidence.publish`.