spark-task-planninglisted

# Spark task planning Convert an approved design and repository-grounded evidence into work packets that Qwen3-Coder-Next can execute without making a new product or architecture decision. Here, Spark means an NVIDIA DGX Spark coding-agent target, not Apache Spark data processing. Use this skill only when the request explicitly targets Qwen, Qwen3-Coder-Next, or DGX Spark. Read [references/qwen-task-packet.md](references/qwen-task-packet.md) before drafting packets and use its schema. After drafting, run `scripts/validate-packets.py <packet.md>`. An explicit file must contain at least one visible packet and must be a regular non-symlink Markdown file no larger than 8 MiB. The validator accepts the standalone, embedded-ledger, and legacy-compatible bold-label renderings; requires the 15 fields in order and exactly one readiness outcome; and checks the execution-critical detail contract: nonblank revalidation trigger, repository inventory, absolute execution root, worktree/collision state, exact read and write sets, at least two line-separated change steps, four-case acceptance matrix, line-separated numbered verification with expected result and proof claim, and a revision-bound handoff. It also refuses branch-only evidence for `READY` or `NO_CHANGE_NEEDED`. A `BLOCKED_BY_SPEC` packet with no selected repository instead records `Not applicable - <reason>` for its immutable evidence and execution root. These are structural gates; the planning pass must still prove that ever