Reading skills
manage_skills with list returns every skill this user can use in this workspace: its id, name, and one line saying what it makes and when it applies. get returns one skill in full, or several by id, such as a skill and its parts in one call. Nothing is loaded into the prompt in advance; a skill can be read when a job of its kind comes up.
A skill is at most 25,000 tokens, the smallest default result size among clients, so it reaches the agent in full. Over MCP the read states its size, so a client that shortens long results keeps it complete. A batch read that would go over the limit stops and returns next_ids. For a client that cuts results shorter, get takes max_chars and continues one skill from next_cursor.
Eversince’s skills
Each skill holds what one kind of job needs, from the material to the tool calls and a working order for them. Its defaults are starting points; a brief, a reference or a brand from the user replaces them.Editing
Formats and ads
Images and design
AI generation
Research and reference
Eversince’s skills can be turned off per user, but can’t be edited or deleted.
A workspace’s skills
manage_skills creates, updates and deletes a workspace’s own skills. They belong to the workspace: update replaces the whole body for everyone, and delete reports how many colleagues still use the skill. A skill an agent saved is marked as saved by an agent. Turning a skill off applies to one user, and an agent connected as that user doesn’t see it through any interface.
A skill body is text. Model names, prices and tool lists written into one aren’t updated automatically.
