spark-task-researchlisted

# Spark task research Build a factual repository brief that lets a planner remove design ambiguity before work reaches Qwen3-Coder-Next. Research only; do not implement or decompose the work into task packets. Here, Spark means an NVIDIA DGX Spark coding-agent target. Apache Spark data-processing work belongs to the normal research and backend skills unless the request also explicitly targets Qwen3-Coder-Next execution. ## Research contract 1. Read repository instructions and the named epic, issue, or task source. 2. Record the repository, branch, current commit, relevant package versions, and applicable local rules. 3. Verify every referenced file, symbol, behavior, dependency, and command against the pinned commit object plus any explicitly separated relevant local state. Read complete relevant files rather than treating snippets or search results as implementation truth. 4. Label each important item as `existing`, `proposed`, `missing`, or `conflicting`. Never present a proposed symbol as existing code. 5. Find reusable patterns and nearby tests. Cite repository evidence as `path:line`. 6. Identify public contracts, data changes, compatibility requirements, error semantics, security boundaries, concurrency constraints, and rollout concerns that the planner must resolve. 7. Confirm the cheapest focused checks and broader verification commands actually available in the repository. Do not invent commands merely to complete the brief. 8. Use external sources only for facts