What problem does it solve? Contributors to the Luke skills catalog must produce a strictly conformant SKILL.md plus a four-category evaluation fixture before opening a proposal issue, and hand-authoring these artifacts against the schema-v1 rules is error-prone. This Skill scaffolds both files deterministically and validates them with the same validator used by CI. ## Core Features & Use Cases - Deterministic bundle generation: Runs scripts/new_skill.py to emit skills/<slug>/SKILL.md and evals/<slug>.json from flags or a JSON answers file, with rollback and locking on failure. - Schema-v1 validation: Runs scripts/validate.py with --allow-draft-provenance to check the draft bundle against catalog rules before submission. - Boundary enforcement: Supports host-bound, host-free web_search-only, and owned-browser authenticated read shapes while refusing mutations, credentials, and private hosts. - Use Case: A contributor wants to propose a skill that reads a planetarium's public hours page; the creator generates the SKILL.md and evaluation fixture, validates them, and the contributor pastes both into the proposal issue form. ## Quick Start Ask the assistant to create a Luke catalog skill proposal bundle for your read-only public-information task, providing a slug, description, public host, and one positive, negative, failure, and redaction example.
