Shards are an evolution of Claude Skills and Skill Graphs, with the key difference being state. Shards run inside a Flint workspace, which means they can persist artifacts within the Flint's Mesh and allow workflows to build on top of each other and compound in capability.
Coding agents running shards can complete a wide range of complex knowledge work including agentic coding, conversing in notepads, writing specifications, managing tasks, producing reports. Each shard teaches the agent a specific domain, what to create and how to structure work.
Written in plain English, Shards make advanced agentic orchestration accessible to anyone. They are reliable through typed artifact templates that enforce consistency, while also seamlessly allowing for human-in-the-loop input through Flint.
A shard is composed of
Shards are composable. A single prompt can reference workflows from multiple shards and the agent orchestrates them in sequence, reading each shard's init, creating artifacts from templates, and logging progress.