sdd-orchestrate-taskslisted

# Orchestrate SDD tasks If the caller limits execution to a batch, return after step 5 before starting another batch: `batch-completed` with pending IDs if the batch was approved, or `blocked` with evidence if the batch still has a failure or no task is eligible. When all tasks are complete, execute step 6 before returning. A batch return does not mean the feature is complete. 1. Resolve the feature and check `prd.md`, `techspec.md`, `tasks.md`, root, and `done/`. Read sources once per version, then task metadata. Confirm authorization to implement and preserve pre-existing changes. If there is evidence of an incorrect completion, reopen the task through the DAG owner: record reason, review, and previous handoff under `Problems and solutions`, move it from `done/` to root, and update link/state to pending. Preserve contract and IDs; revalidate affected dependents. This does not authorize regenerating completed tasks to change their scope. **Output:** IDs, links, states, and dependencies reconciled; divergences have evidence and block the affected unit. 2. Select pending tasks with completed dependencies. Use one executor subagent per task; parallelize only units without collisions in files, contracts, configuration, or build/test resources. Reserve available slots and keep the orchestrator as the manifest's sole writer. **Output:** batch with exclusive owners and scopes. Without available subagents, report the limitation and execute sequentially if the caller allow