[FEATURE] Cowork: Auto-register workspace skills as slash commands in Cowork / Claude Desktop

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

In my workspace I have a .claude/skills folder with some workspace specific skills. The skills in this folder are available to Claude and the agent can execute them. However, they are not directly exposed in the agent's skills list, or in the ui. Using a (example) /review command Claude Desktop gives an Unknown skill: review error.

If I manually zip and upload them to Claude Desktop they are available, but as its a growing project, the skills are subject to change and require manual refreshing after an update.

Proposed Solution

Cowork (and possible also Claude Desktop in Code mode) should automatically discover and register any skills found in the active workspace's .claude/skills/ folder at session start, surfacing them identically to manually installed skills:

They appear in the available skills list injected into context.
They are triggerable via /skill-name slash commands.
Workspace skills and globally installed skills coexist without conflict.
Workspace skills could be scoped to the session (no persistent installation needed).

Alternative Solutions

Current workaround is to refresh the skills after update. But according to the docs (I haven't tried this), you can only upload 1 skill at a time. So for my (currently 7) skills, i have to zip 7 times then upload 7 times. A slightly better 'band-aid' would be to allow multiple skills to exist in a zip (as said: that might be supported, havent tested that scenario)

Priority

Medium - Would be very helpful

Feature Category

Configuration and settings

Use Case Example

I am using Cowork as an assistent to for writing a book. In the ongoing process I am developing an vs-code extension and mcp server to better assist me. I (with Claude :)) generated skills for the agent to help me. The skills are 'Work in progress', and are updated regularly.

They live in the workspaces .claude/skills folder. I write my book in vs code, but have a cowork session active where I ask to review changes, discuss characters, story arc, etc.

Now i can 'ask': can you review my changes. And 'sometimes' it will pick up the skill ('I can see it in ... and run it'). But the agent cant see it directly in its skills list.

Also i can't use / commands to activate them (which all the docs state, so its consistent).

I'd use /review, /memory , /brainstorm probably multiple times per session.

Additional Context

Why this matters

• Version-controlled skills: Teams can ship skills alongside their codebase. Everyone who opens the project gets the same slash commands — no per-user setup.
• Iterative skill development: Skill authors can edit SKILL.md in their repo and test immediately, without re-uploading to the desktop app on every change.
• MCP-generated skills: Tools like my MCP server already generate skills into .claude/skills/ via a setup_ai_files tool. Today those skills are only useful if the user also manually installs them — defeating the purpose of auto-generation.
• Consistency with Claude Code: Claude Code already reads skills from ~/.claude/skills/. Cowork/Desktop should extend that to workspace-level skills, consistent with how CLAUDE.md project instructions already work (workspace-scoped, no installation required).

Suggested precedent

CLAUDE.md project instructions are a good model: they live in the repo, are picked up automatically at session start, and require no separate "installation" step. Workspace skills should behave the same way.

Environment

Claude Desktop (Cowork mode)
Workspace with skills in .claude/skills/ (generated by Bindery MCP setup_ai_files)
Skills are valid SKILL.md files with frontmatter (name, description)