parallel-task-spark — Claude Code skill: what it does, how to install

What parallel-task-spark does

This orchestrator parses a markdown roadmap, analyzes depends_on entries, and builds an execution graph. Work proceeds in parallel waves using child agents assigned to the sparky role. The workflow requires manual invocation via the /parallel-task-spark slash command and never triggers passively during conversation.

Workers follow a test-driven approach by default. They generate failing tests through a tdd_test_writer role before touching application code. When an item cannot undergo automated tests, the worker must document an explicit reason alongside alternative verification like static checks. Each agent stages and commits only its relevant files locally. Git push operations are strictly prohibited inside the subagent cycle to avoid polluting remote branches.

Strict plan formatting is mandatory because broken task identifiers or malformed dependency lists halt the parser immediately. However, committing each wave locally before starting the next prevents merge clashes when multiple agents touch the tree.