Skills share

Share Installed Skills

Export one or more already-installed skills as a downloadable .zip file. The user downloads it via the Cowork preview and saves it wherever they want — Desktop, Documents, a project folder, Slack, email attachment.

This skill does NOT install the exported plugin back into Cowork. The skills are already installed. The .zip format deliberately avoids the "Save plugin"/"Save skill" install prompts that .plugin/.skill extensions trigger — the user gets a standard file download instead.

This skill is for exporting what you already have. To get new skills from a GitHub repo, use /skills-load.

VM constraints (why we use the session outputs folder and .zip)

The Cowork VM has a fixed set of mount points. None of them is the user's Desktop or an arbitrary host path — the Write tool cannot reach ~/Desktop/anything unless the user has explicitly mounted that folder.

What DOES work: /sessions/<session>/mnt/outputs/ is a VM→host bridge mount. Files written there are surfaced in the Cowork UI by mcp__cowork__present_files. A .zip file presented this way gives the user a download button — they save it wherever they want via their OS file picker.

• Bash runs in the VM — use it to enumerate installed skills AND build the ZIP entirely inside the VM.
• Read tool can read from VM-mounted skill directories (not used in the new flow — Bash+Python reads files directly).
• Write tool bridges to the host but only where mounts exist — NOT used in this skill.
• mcp__cowork__present_files surfaces the built ZIP for download.

When to use this vs /skills-load

Situation	Use
"I found a repo with skills I want"	/skills-load <url>
"I want to send skills from my session to a colleague"	/skills-share
"I want to archive the skills I use"	/skills-share
"I want to hand off skills whose GitHub source is gone"	/skills-share

Step 1: Enumerate installed skills

bash "${CLAUDE_PLUGIN_ROOT}/skills/skills-share/scripts/list-installed-skills.sh"

Returns JSON array of {name, description, path, source} objects where source is one of skills-plugin, domain-plugin, or remote-plugin.

If the array is empty:

No skills installed in this session.

Use /skills-load <github-url> to install skills from a repo first.

Then stop.

Step 2: Show the catalog

Display a numbered table:

Installed skills in this session ({N} total):

#	Skill	Description	Source
1	pdf	PDF processing toolkit	skills-plugin
2	xlsx	Excel spreadsheet handling	skills-plugin
3	roadmap-planning	Build quarterly roadmaps...	domain-plugin
4	discovery-process	Run product discovery...	domain-plugin

Which to include? Pick by number (1, 3), name, keyword search ("roadmap"), or --all.

If $ARGUMENTS contains skill names or --all, use those directly. Otherwise wait for the user to choose.

Also ask for a package name:

Package name? (default: shared-skills)

Don't ask where to save — the user will choose that when they download the .zip from the preview.

Step 3: Confirm selection (with dependency warning)

Before building, show what will be packaged:

Will package:

Skill	Source
roadmap-planning	domain-plugin
discovery-process	domain-plugin

Warn about missing cross-references (do not auto-include): Read each selected SKILL.md and scan for references to OTHER installed skills not in the selection. If found:

Note: roadmap-planning references competitive-analysis (installed but not selected). The shared plugin may point to a skill the recipient doesn't have. Include it? (y/n)

User stays in control. Skills-share is a packaging tool, not a curator.

Step 4: Build the .zip inside the VM

Sanitize the package name: kebab-case, [a-z0-9-] only. Reject and re-ask if sanitizing produces an empty string.

Build the ZIP entirely in Bash+Python. All file reading happens inside the VM — no content passes through Claude's context.

python3 << 'PYEOF'
import zipfile, json, os, re, glob

# ── Substitute these values from the /skills-share context ──
pkg_name = "<PKG_NAME>"       # e.g., "my-shared-skills"
skill_paths = [<SKILL_PATHS>] # e.g., ["/path/to/skill-a", "/path/to/skill-b"]

# Cowork outputs dir lives at /sessions/<session>/mnt/outputs/ — discover it.
out_dirs = glob.glob('/sessions/*/mnt/outputs')
if not out_dirs:
    raise RuntimeError("Cowork outputs dir not found - is this a Cowork VM session?")
out_dir = out_dirs[0]
out_path = os.path.join(out_dir, pkg_name + '.zip')

with zipfile.ZipFile(out_path, 'w', zipfile.ZIP_DEFLATED) as zf:
    # Plugin manifest
    manifest = json.dumps({
        "name": pkg_name,
        "description": "Skills exported from Cowork session",
        "version": "1.0.0",
        "author": {"name": "Exported by skills-toolkit"},
        "keywords": ["skills", "shared", pkg_name]
    }, indent=2)
    zf.writestr(".claude-plugin/plugin.json", manifest)

    # Copy each skill's full directory tree
    for skill_path in skill_paths:
        if not os.path.isdir(skill_path):
            continue
        skill_name = os.path.basename(skill_path)
        skill_name = re.sub(r'[^a-zA-Z0-9_-]', '-', skill_name).strip('-')
        for root, dirs, files in os.walk(skill_path):
            for f in files:
                fp = os.path.join(root, f)
                rel = os.path.relpath(fp, skill_path)
                arc = "skills/" + skill_name + "/" + rel.replace('\\', '/')
                zf.write(fp, arc)

print(out_path)
PYEOF

Substitute actual values into the <PLACEHOLDER> slots above. skill_paths is a Python list of quoted path strings. Capture the output path printed by the script.

Why .zip not .plugin: presenting a .plugin file via present_files triggers Cowork's "Save plugin" install button, which is wrong here — the user already has these skills installed. .zip surfaces as a plain download, and it's still directly accepted by Customize > Personal Plugin on the receiving end.

Why the session outputs folder, not /tmp/: /sessions/<session>/mnt/outputs/ is a VM→host bridge mount that present_files surfaces in the Cowork UI. /tmp/ is VM-only.

Step 5: Present the .zip for download

Load mcp__cowork__present_files via ToolSearch, then call it with this exact shape — files is a list of objects, each with a single file_path key:

mcp__cowork__present_files(files=[{"file_path": "<zip-path-from-python>"}])

The path MUST be the Linux path printed by the Python script in step 4 (under /sessions/<session>/mnt/outputs/). Windows-style paths and /outputs/ paths are rejected.

The Cowork UI shows a download action. The user clicks it, picks a location with their OS file picker (Desktop, Documents, anywhere), and the file lands on their host. No mount configuration needed, no manual zipping.

Step 6: Report

{pkg-name}.zip ready to download — {N} skills packaged:

Skill	Source
roadmap-planning	domain-plugin
discovery-process	domain-plugin

Download the .zip from the preview above. Save it wherever you want.

To use/share it:

• Send to a colleague: Email, Slack, or attach the .zip directly.
• Install in Cowork: Customize > Personal Plugin (+) > Create plugin > Upload plugin (accepts .zip directly — no renaming needed).
• Use with Claude Code CLI: Extract the .zip, then claude --plugin-dir ./{pkg-name}/

Edge cases

• No skills installed: Exit with a suggestion to run /skills-load first.
• Skill with only SKILL.md (no supporting files): Works fine — the ZIP just contains the manifest and one SKILL.md.
• User picks skills from multiple sources: Fine, all get bundled together. Note mixed sources in the report.
• Sanitized name empty: If the package name sanitizes to nothing, ask for a new name.
• Binary supporting files (fonts, images, etc.): Bash+Python copies them correctly into the ZIP — no content passes through Claude. This is an advantage of the VM-build approach.
• present_files unavailable: Tell the user the file is at /sessions/<session>/mnt/outputs/{pkg-name}.zip in the VM. The session outputs folder maps to Cowork's app data on the host — on Windows that's %APPDATA%\Claude\local-agent-mode-sessions\..., on macOS it's ~/Library/Application Support/Claude/local-agent-mode-sessions/....