What problem does it solve? Writing a tasks.md for a spec-driven work unit often produces wrong granularity: bloated tasks mixing multiple flows, purely structural setup tasks, or catch-all validation tasks at the end. This Skill defines the standard for decomposing a spec and plan into an ordered, dependency-driven task checklist. ## Core Features & Use Cases - Granularity rules: Defines when to group artifacts by direct dependency and when to split tasks (multiple user flows, more than 3 distinct actions, mixed technical layers). - Ordering by dependency graph: Ensures artifacts are born in the task that first uses them, forbidding upfront "technical base" tasks and final "integrated validation" tasks. - Forward-dependency tracking: Specifies the forward-deps.md record format for temporary stubs that a later task must replace. - Boundary rules: Distinguishes executable tasks from preconditions, open questions, and cross-domain dependencies recorded in spec.md and plan.md. - Use Case: After producing spec.md and plan.md for a new feature unit, generate a tasks.md where each task delivers one testable behavior with its tests included, ordered by real dependencies. ## Quick Start Use the tasks-authoring skill to generate the tasks.md for the work unit in .agents/specs/012-reporting from its spec.md and plan.md.
