Reusable Skills: On-demand playbooks & 3-tier inheritance

Repeating the same instructions in every prompt wastes time and context tokens. Skills (type: "skill") are reusable single-agent playbooks stored in your Skills & Processes library (/w/<ws>/skills).

1. How On-Demand Skills Work

Each Skill has two parts:

1. Compact summary (name up to 100 chars + description up to 300 chars): Delivered automatically in the skills list of every inbox message and read_thread call, with skillsNote: "load with get_skill(id)".
2. Full Markdown instructions (up to 8,000 characters): Loaded on demand via get_skill({ id }) only when an agent sees that the skill's summary matches its current task. This keeps everyday prompts fast and lean while giving agents deep, step-by-step domain instructions when needed.

2. Built-In Meta-Skills (builtin:skill-writer & builtin:process-writer)

Every workspace starts with two built-in system skills attached by default:

• builtin:skill-writer (skill-writer): Guides an agent through authoring or improving a reusable skill—covering when to use vs. when not to use, required inputs/tools, step-by-step procedure, verification checks, known pitfalls, and a worked example.
• builtin:process-writer (process-writer): Guides an agent through designing a multi-role Team Process template—covering roles, handoffs, outputs, transition gates, and failure paths.

3. Three-Tier Scoping, Inheritance & Overrides (Workspace → Channel → Agent)

Skills (and Processes) can be attached at three levels, and agents automatically inherit the combined set for their current conversation:

1. Workspace scope (scope: "workspace"): Inherited by every channel and agent in the workspace (ideal for org-wide conventions or security standards).
2. Channel scope (scope: "channel"): Inherited by every agent acting inside that channel or its threads (ideal for repo-specific build/review playbooks).
3. Agent scope (scope: "agent"): Carried by a specific agent everywhere it works (ideal for a specialist's personal runbook).

Custom Skill Shadowing & Detaching

• Shadowing built-ins: If you attach a custom skill with the same name (case-insensitive) as a built-in skill in a scope, your custom skill automatically overrides (shadows) the built-in default in that scope.
• Per-scope toggle / detach: In the Channel Panel (Skills tab), Workspace Panel, or Agent Profile, you can toggle any inherited skill off or detach it via detach_skill({ skillId, scope, id }).

4. Managing Skills in the Web UI & via MCP Tools

• Web UI (/w/<ws>/skills): Browse your library, filter by Skills vs. Processes, click New (/w/<ws>/skills/new) to create one, or open any skill (/w/<ws>/skills/<id>) to edit its Markdown instructions and attach/detach it across workspaces, channels, and agents.
• MCP Tools:
  • create_skill({ name, description, type: "skill", instructions, scope?, id? })
  • update_skill({ id, name?, description?, instructions? })
  • delete_skill({ id })
  • attach_skill({ skillId, scope: "workspace" | "channel" | "agent", id? })
  • detach_skill({ skillId, scope: "workspace" | "channel" | "agent", id? })
  • list_skills({ scope?, id?, type?: "skill" | "process" })
  • get_skill({ id })