Home / Companies / Sleuth / Blog / April 2026

April 2026 Summaries

2 posts from Sleuth

Filter
Month: Year:
Post Summaries Back to Blog
In the blog post by Dylan Etkin, he discusses the challenges of using traditional version control systems like Git for managing rapidly evolving Agent Skills, which are small, dynamic pieces of code designed to handle specific tasks or annoyances within a development environment. Etkin shares his experience of enhancing a CLI tool and the subsequent testing that revealed the limitations of Git in accommodating the fluid lifecycle of Agent Skills, which often remain personal and unshared due to the high "ceremony tax" of committing to a repository. He highlights the need for a more agile and context-aware system that facilitates friction-free sharing, cross-repo use, and task-based discovery, proposing a new platform, skills.new, to complement Git by offering a more suitable environment for these fast-moving, cross-cutting assets. The post underscores the disconnect between current version control practices and the practical needs of engineering teams in managing and utilizing these skills effectively, pointing out the necessity for a system that recognizes the unique requirements of Agent Skills without imposing the burdensome formalities of traditional code management processes.
Apr 17, 2026 1,438 words in the original blog post.
Dylan Etkin argues that the absence of Role-Based Access Control (RBAC) for Agent Skills and Multi-Context Processors (MCPs) poses a significant security risk as organizations integrate large language models (LLMs) like Claude or ChatGPT across various company systems. Without RBAC, employees could inadvertently access sensitive data, as the system lacks guardrails to restrict tools and skills to appropriate roles. The author highlights that the current MCP ecosystem lacks authorization scoping, leading to potential exposure of unnecessary tools to roles that don't require them, thereby increasing security risks. The text emphasizes the need for a management layer that applies RBAC at the orchestration level, ensuring that employees can only access skills and MCPs relevant to their roles, thus safeguarding sensitive data and maintaining organizational security policies. This approach is essential to leverage the transformative potential of AI while adhering to existing access policies, a gap that the author's platform, skills.new, aims to address.
Apr 13, 2026 1,106 words in the original blog post.