Permissions

Skills declare what they need access to. Permissions are shown to users before install.

Permission	What It Allows
`read:conversations`	Access conversation history
`read:artifacts`	Read generated files and images
`write:artifacts`	Create and store files
`write:notes`	Create user-visible notes
`external:http`	Make outbound HTTP requests
`execute:tools`	Call the skill’s declared tools
`autonomous:run`	Execute without user in the loop

Enforcement

Skills without execute:tools have their tools stripped at load time
Skills without declared permissions operate in legacy mode (all tools allowed)
Permission combinations that grant broad write or external-network access are promoted to a higher review tier during submission