mine

name: mine description: Import Claude Code resources from GitHub repositories or local paths, and convert AI framework patterns to Claude Code format. Use when the user wants to integrate Claude artifacts (skills, commands, agents, hooks, MCP configs) from a repo, scan a repo for Claude Code assets, install skills from a GitHub repository, convert a repository into a Claude skill, convert Fabric patterns to Claude agents/commands, convert LangChain chains to agents, import workflow automation from existing codebases, or transform AI framework repositories into Claude-compatible resources. Automatically detects Claude Code resources and AI frameworks (Fabric, LangChain, AutoGen) with dry-run defaults and conflict protection.

MINE

Import and integrate Claude Code resources from GitHub repositories or local paths. This skill scans repositories for Claude artifacts (skills, commands, agents, hooks, MCP configs, CLAUDE.md), converts AI framework patterns (Fabric, LangChain, AutoGen) to Claude Code format, and generates new skill packs from repository workflows.

Quick Start

Import Claude Artifacts

# Dry-run scan (preview, no writes)
python3 ~/.claude/skills/mine/scripts/import_assets.py \
  --source https://github.com/user/repo \
  --scope user \
  --dry-run

# Apply import after reviewing
python3 ~/.claude/skills/mine/scripts/import_assets.py \
  --source https://github.com/user/repo \
  --scope user \
  --apply

Convert Fabric Patterns

# Auto-detect and convert Fabric patterns
python3 ~/.claude/skills/mine/scripts/import_assets.py \
  --source https://github.com/danielmiessler/fabric \
  --scope user \
  --mode convert \
  --apply

# Result: patterns/ → Claude agents/commands
# - patterns/extract_wisdom → ~/.claude/agents/extract_wisdom.md
# - patterns/summarize → ~/.claude/commands/summarize.md

Discover Agentic Content

# Find and convert prompts, agents, tools anywhere in a repo
python3 ~/.claude/skills/mine/scripts/import_assets.py \
  --source https://github.com/user/repo \
  --scope user \
  --discover-agentic \
  --min-confidence 0.65 \
  --apply

Generate Skill Pack

# Create skill from repository without Claude artifacts
python3 ~/.claude/skills/mine/scripts/import_assets.py \
  --source https://github.com/user/repo \
  --scope project \
  --mode generate \
  --target-repo ~/projects/myapp \
  --apply

Import Modes

Import (Default)

Copies detected Claude artifacts to the appropriate locations:

Convert

Converts AI framework patterns to Claude Code format:

Fabric Framework:

• Detects patterns/ directory with system.md files
• Simple patterns → Commands (extract, summarize, create)
• Complex patterns → Agents (analyze, review, evaluate)
• Multi-step patterns → Skills with workflows

LangChain Framework:

• Detects LangChain/LangGraph dependencies and imports
• Generates scaffold commands and assistant agents
• Includes conversion report with entry points

AutoGen Framework:

• Detects AutoGen dependencies and agent definitions
• Generates scaffold commands and coordinator agents
• Includes conversion report with agent mappings

Agentic Discovery

Heuristic scanner that finds agentic content anywhere in a repository:

• Enable: --discover-agentic flag
• Threshold: --min-confidence <0.0-1.0> (default: 0.65)
• Detects: System prompts, agent definitions, tool specs, workflows
• Converts to: Claude Code artifacts with provenance tracking
• Safety: Automatic secret redaction before conversion

Generate

Creates skill packs from repositories without Claude artifacts:

• Analyzes README, docs, package.json, pyproject.toml, Makefile
• Creates .claude/skills/<repo>-workflow/ with SKILL.md
• Optionally creates commands for common tasks
• Includes Context7 hints when MCP is configured

Skill-Creator Integration

For complex repositories where template-based generation produces suboptimal results, MINE can hand off to Anthropic's skill-creator skill for AI-assisted skill authoring.

Behavior:

• Auto-detect (default): Hands off when confidence is low AND skill-creator is installed
• Force handoff: Use --use-skill-creator to always use skill-creator
• Disable handoff: Use --no-skill-creator to use templates only

# Auto-detect (default) - uses skill-creator if installed and appropriate
python3 ~/.claude/skills/mine/scripts/import_assets.py \
  --source https://github.com/user/repo \
  --scope user \
  --mode generate

# Force skill-creator handoff
python3 ~/.claude/skills/mine/scripts/import_assets.py \
  --source https://github.com/user/repo \
  --scope user \
  --use-skill-creator

# Disable skill-creator (templates only)
python3 ~/.claude/skills/mine/scripts/import_assets.py \
  --source https://github.com/user/repo \
  --scope user \
  --no-skill-creator

When Handoff Occurs:

Requirements:

• Anthropic's skill-creator skill installed at ~/.claude/skills/skill-creator/
• skill-creator is recommended but not required — MINE works without it

Safety Features

Dry-Run Default

All operations preview changes without writing. Use --apply (or --dry-run=false) to apply.

Conflict Protection

Overlapping destinations are blocked unless --force-conflicting is specified.

Path Safety

• Path traversal (../) is blocked [claim: path_safety_traversal_blocked]
• All paths validated against scope roots [claim: path_safety_traversal_blocked]
• Symlinks are skipped to prevent traversal attacks [claim: symlink_safety_skipped]

Hook Security

• Hooks import to .claude/hooks.imported.<repo>/ (never auto-enabled) [claim: hooks_staged]
• Manual review and merge required [claim: hooks_staged]

Secret Protection

• Never stores credentials on disk [claim: secrets_redacted]
• Uses environment variables (GITHUB_TOKEN) when needed [claim: secrets_redacted]
• Automatic redaction during agentic conversion [claim: secrets_redacted]

Scripts

scan_repo.py

Scans a repository and outputs a JSON report of detected artifacts.

python3 ~/.claude/skills/mine/scripts/scan_repo.py \
  --source <url_or_path> \
  [--output report.json]

Output:

{
  "repo_id": "user/repo",
  "source": "https://github.com/user/repo",
  "framework_type": "fabric",
  "detected_artifacts": [...],
  "suggested_actions": ["convert"],
  "risks": []
}

import_assets.py

Main import/convert/generate workflow.

python3 ~/.claude/skills/mine/scripts/import_assets.py \
  --source <url_or_path> \
  --scope user|project \
  [--mode auto|import|convert|generate] \
  [--apply] \
  [--dry-run] \
  [--target-repo <path>] \
  [--overwrite-with-backup] \
  [--discover-agentic] \
  [--min-confidence <0.0-1.0>]

Arguments:

convert_framework.py

Standalone framework converter.

python3 ~/.claude/skills/mine/scripts/convert_framework.py \
  --framework fabric|langchain|autogen \
  --source <path> \
  --output <output_dir> \
  [--apply]

generate_skillpack.py

Standalone skill pack generator.

python3 ~/.claude/skills/mine/scripts/generate_skillpack.py \
  --source <url_or_path> \
  --target-dir <output_path> \
  [--repo-name <name>] \
  [--apply]

Detection Patterns

Claude Artifacts

Frameworks

Agentic Content

Context7 Integration

When Context7 MCP is configured (detected via .mcp.json), generated artifacts include Context7 usage hints:

• Commands include "Documentation" sections for API docs
• Agents include "Docs policy" sections recommending Context7
• Conversion reports note when Context7 is not configured

Note: The skill generates documentation that instructs Claude to use Context7; it does not directly invoke Context7 tools.

To enable Context7:

{
  "mcpServers": {
    "context7": {
      "command": "npx",
      "args": ["-y", "@upstash/context7-mcp"]
    }
  }
}

GitHub Authentication

Authentication methods (in order of preference):

1. GitHub CLI (gh) — Uses existing authentication
2. GITHUB_TOKEN — Environment variable (never stored on disk)
3. Unauthenticated — Falls back to git clone for public repos

# Optional: Set GitHub token
export GITHUB_TOKEN="ghp_your_token_here"

Platform Compatibility

• Cross-platform paths: Uses pathlib consistently
• Windows long paths: Handles paths >260 chars via \\?\ prefix
• Case-insensitive filesystems: Safe two-step renames
• Subprocess portability: Uses sys.executable for Python calls

Examples

Convert Fabric Framework

python3 ~/.claude/skills/mine/scripts/import_assets.py \
  --source https://github.com/danielmiessler/fabric \
  --scope user \
  --apply

Import from Public Repository

python3 ~/.claude/skills/mine/scripts/import_assets.py \
  --source https://github.com/example/claude-pdf-tools \
  --scope user \
  --apply

Generate Workflow Skill

python3 ~/.claude/skills/mine/scripts/import_assets.py \
  --source ~/code/my-python-app \
  --scope project \
  --mode generate \
  --target-repo ~/code/my-python-app \
  --apply

Scan Without Importing

python3 ~/.claude/skills/mine/scripts/scan_repo.py \
  --source https://github.com/example/repo \
  --output scan-results.json

Reference

For detailed technical information, see references/REFERENCE.md.