<skill_overview> <purpose>Turn a story into a small, verifiable task list that can be implemented without ambiguity.</purpose> <triggers> <trigger>You have a story with acceptance criteria but no implementation breakdown</trigger> <trigger>User asks "break it into tasks" / "need a decomposition"</trigger> <trigger>You need a safe order of work with dependencies</trigger> </triggers> </skill_overview>
<principles> <principle>Prefer vertical slices (end-to-end) over horizontal layers.</principle> <principle>Each task should be independently verifiable (yes/no).</principle> <principle>Happy path first, then validation/errors, then polish.</principle> <principle>Keep tasks small: 30-120 minutes each (guideline).</principle> </principles> <rules> <rule>Tasks describe intent and observable outcome, not micro-steps.</rule> <rule>Each task MUST include a done criteria (binary) and any dependencies.</rule> <rule>If a task spans multiple subsystems, split it or create explicit sub-tasks per boundary.</rule> <rule>Do not create "infrastructure-first" tasks unless required by the story.</rule> </rules><task_template> <template>
- •id: T1 title: ... intent: Why this task exists (1 sentence) inputs: What it needs (APIs, config, data, decisions) outputs: What it produces (code paths, docs, artifacts) done: A binary check (how we know it is done) depends_on: [] risks: [] </template>
</task_template>
<anti_patterns> <avoid>"Refactor everything" tasks (too broad)</avoid> <avoid>Tasks without a done check</avoid> <avoid>Tasks that only say "implement X" without specifying behavior</avoid> <avoid>Premature optimization tasks</avoid> </anti_patterns>