luke-skills

Luke Skills Catalog

The Luke Skills Catalog provides a framework for allowing Luke to learn and read public pages while ensuring strict boundaries around access and security. Each skill defines when it applies, what public host it can read from, and outlines its intent and return parameters. Importantly, skills cannot alter external systems or install executable code, thus maintaining a level of safety and predictability.

Key Features

• Defined Skill Boundaries: Skills are designed with specific boundaries that restrict unauthorized access to private pages. For instance, a skill may read a project's public release page and summarize details for a specified version without gaining additional access rights.
• Authentication Handling: User authentication is managed through Luke's isolated browser profile, with no credentials stored or transmitted through the skill itself. Skills are put in place solely for read-only operations.
• Curation and Trust: The curation of skills doesn’t equate to a guarantee of quality or safety. Each submission goes through a detailed review process that documents evidence and checks, but ultimately, users must independently verify content hashes on their machines.
• Proposal and Release Lifecycle: The repository facilitates a step-by-step process for contributors to propose new skills, undergo review, and release them to the public while ensuring that they meet the stringent requirements set out for V1 eligibility.

Eligibility and Constraints

Only skills that adhere to specific eligibility criteria are accepted in the catalog:

• Skills must have a single inert UTF-8 SKILL.md file.
• Read-only behaviors are required against defined public hosts or through web search without authentication.
• All actions fall within a scope defined by the skill's intent, with no capabilities to send, purchase, or alter content externally.

Lifecycle Terms

The project employs a set of precise terms that outline the state of skills throughout their lifecycle:

Term	Definition
Curated	The skill has passed a review process but does not imply trust or operational quality.
Validated	The code has undergone mechanical checks but does not ensure factual accuracy or future behavior.
Proposed	An indication of user needs with boundaries and evidence, awaiting review.
Reviewed	Maintainers have evaluated the skill and recorded a decision in the public issue.
Release-eligible	Skills that are approved and can be selected for public release.
Published	Confirmed skills selected by a governed release process are made available for use.

Public Interaction

Contributors can interact with the repository through filed proposals that document user needs and evidence. Furthermore, a designated public issue provides a transparent view of skill status, governance, and ongoing updates. The repository also outlines procedures for security and conduct reporting, ensuring a structured approach to maintaining project integrity.

For anyone interested in contributing to or understanding the future development of these skills, key documents such as CONTRIBUTING.md and SECURITY.md are essential resources.