skill-import

name	skill-import
description	Imports skills from the dwarvesf/claude-skills GitHub repo into Claude.ai. Use when the user says "import skill", "fetch skill from repo", "update skill from GitHub", "pull latest skill", "/skill-import", or wants to bring a skill from the GitHub repo into their Claude.ai /mnt/skills/user/ directory. This is the reverse of skill-sync (which pushes Claude.ai → GitHub).
updated	"2026-03-28T10:00:00.000Z"

Skill Import

Import skills from the dwarvesf/claude-skills GitHub repo into Claude.ai's /mnt/skills/user/ directory. This is the reverse of skill-sync (which pushes from Claude.ai to GitHub).

Prerequisites

• Running on Claude.ai (needs write access to /mnt/skills/user/)
• GitHub repo must be accessible (public repo or authenticated access)
• For MCP method: Github MCP Worker must be connected

Trigger

User says: "import skill", "fetch skill", "pull skill from repo", "update skill from GitHub", "get latest skills", "/skill-import"

Skip conditions:

• If the user wants to push to GitHub, use skill-sync instead
• If the user wants a zip download, use skill-export instead
• If running in Claude Code (not Claude.ai), this skill doesn't apply. Just git pull.

Workflow

Step 1: Identify skills to import

If the user specifies a skill name, use that. Otherwise, fetch the repo's skill index:

Option A (preferred, if read_skill MCP tool exists):

list_notes() or read the repo README

Option B (fallback, via GitHub raw URL):

WebFetch: https://raw.githubusercontent.com/dwarvesf/claude-skills/master/README.md

Parse the skills table from README.md to show available skills:

## Available skills in repo

| Skill | Description |
|-------|-------------|
| prompt-improver | Sharpens prompts before executing |
| content-spec | Enforce spec phase before content work |
| knowledge-capture | Capture learning moments to Capacities |
| skill-sync | Push skills from Claude.ai to GitHub |
| skill-export | Export skills to repo-ready format |
| skill-import | Import skills from GitHub to Claude.ai |

Import which? [name / several / all]

Step 2: Fetch skill content from GitHub

For each skill to import, fetch the SKILL.md content:

Option A (preferred, if read_skill MCP tool exists):

read_note(topic: "skills/<skill-name>", title: "SKILL")

Option B (fallback, via GitHub raw URL):

WebFetch: https://raw.githubusercontent.com/dwarvesf/claude-skills/master/skills/<skill-name>/SKILL.md

Also check for supporting files by fetching the directory listing or known paths:

• skills/<skill-name>/references/
• skills/<skill-name>/scripts/
• skills/<skill-name>/templates/

Step 3: Diff check (conflict detection)

Before overwriting, check if a local version already exists.

Read local version:

cat /mnt/skills/user/<skill-name>/SKILL.md

Compare:

1. No local version exists → New skill. Proceed to install.
2. Local matches repo → Already in sync. Skip unless user forces.
3. Versions differ → Show a diff summary:

## Conflict: knowledge-capture

Local version:  updated 2026-03-25T09:15:00Z
Repo version:   updated 2026-03-28T14:30:00Z (newer)

Changes from repo:
- Added visual pipeline section
- Updated push destination format
- Fixed Capacities type mapping

Action: [import repo version] [keep local] [show full diff]

Conflict resolution rules:

• If repo updated is newer than local updated: recommend importing (repo is newer)
• If local updated is newer than repo updated: warn that local has newer changes, recommend pushing to repo first via skill-sync
• If neither has updated field: show full diff, ask user to decide
• If content differs but timestamps are the same: show diff, ask user (concurrent edits)

User can always force import by saying "import anyway" or "overwrite local".

Step 4: Install to Claude.ai

For each approved skill:

# Create skill directory if it doesn't exist
mkdir -p /mnt/skills/user/<skill-name>

# Write the SKILL.md
cat > /mnt/skills/user/<skill-name>/SKILL.md << 'SKILLEOF'
<full SKILL.md content from repo>
SKILLEOF

For supporting files (references, scripts, templates):

mkdir -p /mnt/skills/user/<skill-name>/references
# Write each supporting file

Step 5: Report results

After all imports complete:

## Import Results

Imported:
- knowledge-capture (updated: 2026-03-28T14:30:00Z)
- content-spec (updated: 2026-03-28T10:00:00Z)

Skipped:
- prompt-improver (already in sync)
- skill-sync (local is newer, push to repo first)

Not in repo:
- my-custom-skill (local only)

Skills are now available. Start a new conversation to pick up changes.

Step 6: Remind about conversation reload

Skills in /mnt/skills/user/ are loaded at conversation start. After importing:

NOTE: Imported skills will be available in your next conversation.
Start a new chat to use the updated skills.

Batch import

When the user says "import all" or "fetch all skills":

1. Fetch the skill list from the repo README
2. For each skill, fetch content from GitHub
3. Diff check each against local versions
4. Show combined report:

## Import Report

| Skill | Local Status | Repo Version | Action |
|-------|-------------|--------------|--------|
| knowledge-capture | Outdated (local 03-25) | 03-28 (newer) | Import |
| content-spec | Missing | 03-28 | Import (new) |
| prompt-improver | In sync | 03-28 | Skip |
| skill-sync | Local newer (03-29) | 03-28 | Skip (push first) |

Ready to import: knowledge-capture, content-spec

5. Import all approved skills after user confirms
6. Report results

Important rules

1. Always diff check before overwriting. Never silently replace a local version that might have unpushed changes.
2. Warn when local is newer. If the local version is newer than repo, the user should skill-sync first to avoid losing work.
3. Show the import report before writing. User must confirm what will be installed.
4. This skill only runs on Claude.ai. In Claude Code, skills live in the git repo directly; just git pull.
5. Respect the updated field. Use it as the source of truth for version comparison.
6. One skill at a time for writes. Don't batch file writes; if one fails, the rest should still succeed.
7. Always remind about conversation reload. Skills are loaded at conversation start, not mid-conversation.