Cloudflare turns agent skills into workspace files: drop a folder, and the agent uses it next turn with no build or deploy
Synopsis
Cloudflare's Agents SDK now treats Agent Skills as files in the agent's workspace: write a folder containing SKILL.md under /workspace/.agents/skills/ and the agent can use it on its next turn, with no build step or deploy, and the agent can write its own skills. Skills load from a workspace, an R2 bucket, or code, and Think lists every source again before each turn; each SKILL.md is checked against the Agent Skills specification, and a skill that breaks a rule still loads with a logged warning. This is a breaking change: direct agents/skills use, the Vite import, SkillRegistry, and Think's skillWorkspace are removed.
Interpretation
Cloudflare's Agents SDK now treats agent skills as ordinary files in the agent's workspace rather than code assets that need a separate build and deploy. Previously skills went through the agents/skills Vite import and compile path, which required a build step and a deploy to take effect. The changelog's comparison is direct: write a folder with SKILL.md into /workspace/.agents/skills/ and the agent can use it on its next turn, with no build step or deploy.
Skills no longer come from a single place: an agent can load them from a workspace, an R2 bucket, or code, and every source is listed again before each turn. The old path tied skills to compiled output, with one source and one way to update them. The changelog lists the three sources as the first item of the release and states that Think lists every source again before each turn.
Skill scripts are themselves workspace files, and the model runs them with the workspace exec tool. Scripts no longer need to ship as a separate compiled unit alongside the skill. The changelog states that skill scripts are workspace files and that the model runs them with the workspace exec tool, pointing to the Execution section.
When a skill changes mid-conversation, Think sends the new catalog as a reminder instead of rebuilding the cached system prompt. Previously a skill change would have meant rebuilding the cached system prompt. The changelog says Think sends the new catalog as a reminder when a skill changes mid-conversation, and that an overridden getSystemPrompt() now stays at the top of the system prompt, above context blocks.
Perspective
This describes interface changes in the Cloudflare Agents SDK and Think and their loading, validation, and caching behavior, for developers building agents with that SDK; the changelog reports no measurement of performance or task outcomes.
The changelog states that a skill breaking a rule still loads with only a logged warning, so validation does not block non-compliant skills from taking effect; it is also a breaking change that removes the old interfaces, so versions must be upgraded as described.