Notion 3.7 turns team workflows into portable agent skills
Notion 3.7 adds a shared skills library and exports workflows as SKILL.md packages for Claude Code, Codex, Cursor, Gemini, and Grok.
Notion 3.7 adds a shared skills library for teams and lets users export those workflows as SKILL.md packages for Claude Code, Codex, Cursor, Gemini, and Grok. The change moves repeatable operating instructions out of private prompts and scattered documents into a maintained workspace artifact.
Notion 3.7 makes team skills a managed workspace asset
Notion describes skills as reusable instructions that teach an AI how a team works. Examples on the release page include producing a monthly business review in a house format, reviewing a document against internal standards, and generating a new-hire checklist from company policies.
The library gives teams one place to create, share, update, and discover these workflows. Skills can also be run automatically, which makes the feature closer to lightweight process automation than a prompt gallery.
The export is a SKILL.md package, not a vendor lock-in format
The important interoperability detail is the export surface. Notion says it writes a SKILL.md file together with approved supporting files. When a skill changes, Notion marks the local copy so users can identify that a newer workspace version is available.
That format lines up with documented receiving surfaces from more than one AI vendor. OpenAI’s Codex documentation describes skills as packages containing instructions, resources, and optional scripts. Claude Code documents skills as reusable extensions that can be created, managed, and shared. Those independent docs support the practical claim that the exported artifact maps to real agent workflows rather than remaining a Notion-only automation.
What teams should check before exporting workflows
Portability does not remove governance work. A skill can carry internal terminology, approval rules, file references, or scripts into a local agent environment. Teams should review the supporting files, define which repositories or folders an agent may access, and decide who can publish updates to a shared skill.
The update indicator is useful for change control, but it is not a security review. A modified skill still needs the same inspection as any other instruction or executable helper brought into an agent workspace. Keep secrets out of skill packages and make external actions explicit.
For builders, the immediate opportunity is to standardize one narrow workflow—such as release-note drafting or PR triage—then export it to the agents the team already uses. The unresolved operational question is how consistently each target agent handles the same supporting files and optional scripts; the package format is shared, but runtime behavior remains product-specific.
Sources
Try the related loot
Run Low-Cost, Deterministic Computer Use with TypeSafe