speckit-tasks — Claude Code skill: tasks from your spec

What speckit-tasks does

The skill reads the feature directory: plan.md and spec.md are required, while data-model.md, contracts/, research.md and quickstart.md are picked up when present. Tech stack and project structure come from the plan, user stories with their priorities come from the spec. The output follows a template: a setup phase, foundational blocking work, one phase per story in priority order, and a polish phase at the end. Every task carries a file path, and the document closes with a dependency graph and parallel execution examples.

Before generating anything, the skill looks at .specify/extensions.yml for entries under hooks.before_tasks. Hooks without an enabled field count as on; enabled: false skips them. If the YAML fails to parse, the skill reports the parser error and carries on without checking hooks. Mandatory hooks it runs itself and waits for the result, optional ones it only prints with a command you can call. The same pass happens for hooks.after_tasks after generation.

This only works inside a spec-kit project. You need the .specify/ directory with its scripts, including .specify/scripts/bash/setup-tasks.sh. Without it the setup step fails and the skill gets neither the document list nor the template. Missing data-model.md or contracts/ is fine — tasks are generated from whatever sits in the feature directory.