skill-seekers-reference/docs/zh-CN/user-guide/04-packaging.md

Packaging Guide

Skill Seekers v3.1.0
Export skills to AI platforms and vector databases

---

Overview

Packaging converts your skill directory into a platform-specific format:

output/my-skill/ ──▶ Packager ──▶ output/my-skill-{platform}.{format}
    ↓                                ↓
(SKILL.md +        Platform-specific  (ZIP, tar.gz,
 references)        formatting        directories,
                                     FAISS index)

---

Supported Platforms

Platform	Format	Extension	Best For
Claude AI	ZIP + YAML	.zip	Claude Code, Claude API
Google Gemini	tar.gz	.tar.gz	Gemini skills
OpenAI ChatGPT	ZIP + Vector	.zip	Custom GPTs
LangChain	Documents	directory	RAG pipelines
LlamaIndex	TextNodes	directory	Query engines
Haystack	Documents	directory	Enterprise RAG
Pinecone	Markdown	.zip	Vector upsert
ChromaDB	Collection	.zip	Local vector DB
Weaviate	Objects	.zip	Vector database
Qdrant	Points	.zip	Vector database
FAISS	Index	.faiss	Local similarity
Markdown	ZIP	.zip	Universal export
Cursor	.cursorrules	file	IDE AI context
Windsurf	.windsurfrules	file	IDE AI context
Cline	.clinerules	file	VS Code AI

---

Basic Packaging

Package for Claude (Default)

# Default packaging
skill-seekers package output/my-skill/

# Explicit target
skill-seekers package output/my-skill/ --target claude

# Output: output/my-skill-claude.zip

Package for Other Platforms

# Google Gemini
skill-seekers package output/my-skill/ --target gemini
# Output: output/my-skill-gemini.tar.gz

# OpenAI
skill-seekers package output/my-skill/ --target openai
# Output: output/my-skill-openai.zip

# LangChain
skill-seekers package output/my-skill/ --target langchain
# Output: output/my-skill-langchain/ directory

# ChromaDB
skill-seekers package output/my-skill/ --target chroma
# Output: output/my-skill-chroma.zip

---

Multi-Platform Packaging

Package for All Platforms

# Create skill once
skill-seekers create <source>

# Package for multiple platforms
for platform in claude gemini openai langchain; do
  echo "Packaging for $platform..."
  skill-seekers package output/my-skill/ --target $platform
done

# Results:
# output/my-skill-claude.zip
# output/my-skill-gemini.tar.gz
# output/my-skill-openai.zip
# output/my-skill-langchain/

Batch Packaging Script

#!/bin/bash
SKILL_DIR="output/my-skill"
PLATFORMS="claude gemini openai langchain llama-index chroma"

for platform in $PLATFORMS; do
  echo "▶️ Packaging for $platform..."
  skill-seekers package "$SKILL_DIR" --target "$platform"
  
  if [ $? -eq 0 ]; then
    echo "✅ $platform done"
  else
    echo "❌ $platform failed"
 fi
done

echo "🎉 All platforms packaged!"

---

Packaging Options

Skip Quality Check

# Skip validation (faster)
skill-seekers package output/my-skill/ --skip-quality-check

Don't Open Output Folder

# Prevent opening folder after packaging
skill-seekers package output/my-skill/ --no-open

Auto-Upload After Packaging

# Package and upload
export ANTHROPIC_API_KEY=sk-ant-...
skill-seekers package output/my-skill/ --target claude --upload

---

Streaming Mode

For very large skills, use streaming to reduce memory usage:

# Enable streaming
skill-seekers package output/large-skill/ --streaming

# Custom chunk size
skill-seekers package output/large-skill/ \
  --streaming \
  --streaming-chunk-chars 2000 \
  --streaming-overlap-chars 100

When to use:

• Skills > 500 pages
• Limited RAM (< 8GB)
• Batch processing many skills

---

RAG Chunking

Optimize for Retrieval-Augmented Generation:

# Enable semantic chunking
skill-seekers package output/my-skill/ \
  --target langchain \
  --chunk-for-rag \
  --chunk-tokens 512

# Custom chunk size
skill-seekers package output/my-skill/ \
  --target chroma \
  --chunk-tokens 256 \
  --chunk-overlap-tokens 50

Chunking Options:

Option	Default	Description
--chunk-for-rag	auto	Enable chunking
--chunk-tokens	512	Tokens per chunk
--chunk-overlap-tokens	50	Overlap between chunks (tokens)
--no-preserve-code-blocks	-	Allow splitting code blocks

自动缩放重叠: 当 --chunk-tokens 设置为非默认值但 --chunk-overlap-tokens 保持默认值 (50) 时，重叠会自动缩放为 max(50, chunk_tokens / 10)，以在较大的分块中实现更好的上下文保留。

---

Platform-Specific Details

Claude AI

skill-seekers package output/my-skill/ --target claude

Upload:

# Auto-upload
skill-seekers package output/my-skill/ --target claude --upload

# Manual upload
skill-seekers upload output/my-skill-claude.zip --target claude

Format:

• ZIP archive
• Contains SKILL.md + references/
• Includes YAML manifest

---

Google Gemini

skill-seekers package output/my-skill/ --target gemini

Upload:

export GOOGLE_API_KEY=AIza...
skill-seekers upload output/my-skill-gemini.tar.gz --target gemini

Format:

• tar.gz archive
• Optimized for Gemini's format

---

OpenAI ChatGPT

skill-seekers package output/my-skill/ --target openai

Upload:

export OPENAI_API_KEY=sk-...
skill-seekers upload output/my-skill-openai.zip --target openai

Format:

• ZIP with vector embeddings
• Ready for Assistants API

---

LangChain

skill-seekers package output/my-skill/ --target langchain

Usage:

from langchain.document_loaders import DirectoryLoader

loader = DirectoryLoader("output/my-skill-langchain/")
docs = loader.load()

# Use in RAG pipeline

Format:

• Directory of Document objects
• JSON metadata

---

ChromaDB

skill-seekers package output/my-skill/ --target chroma

Upload:

# Local ChromaDB
skill-seekers upload output/my-skill-chroma.zip --target chroma

# With custom URL
skill-seekers upload output/my-skill-chroma.zip \
  --target chroma \
  --chroma-url http://localhost:8000

Usage:

import chromadb

client = chromadb.HttpClient(host="localhost", port=8000)
collection = client.get_collection("my-skill")

---

Weaviate

skill-seekers package output/my-skill/ --target weaviate

Upload:

# Local Weaviate
skill-seekers upload output/my-skill-weaviate.zip --target weaviate

# Weaviate Cloud
skill-seekers upload output/my-skill-weaviate.zip \
  --target weaviate \
  --use-cloud \
  --cluster-url https://xxx.weaviate.network

---

Cursor IDE

# Package (actually creates .cursorrules file)
skill-seekers package output/my-skill/ --target cursor

# Or install directly
skill-seekers install-agent output/my-skill/ --agent cursor

Result: .cursorrules file in your project root.

---

Windsurf IDE

skill-seekers install-agent output/my-skill/ --agent windsurf

Result: .windsurfrules file in your project root.

---

Quality Check

Before packaging, skills are validated:

# Check quality
skill-seekers quality output/my-skill/

# Detailed report
skill-seekers quality output/my-skill/ --report

# Set minimum threshold
skill-seekers quality output/my-skill/ --threshold 7.0

Quality Metrics:

• SKILL.md completeness
• Code example coverage
• Navigation structure
• Reference file organization

---

Output Structure

After Packaging

output/
├── my-skill/                    # Source skill
│   ├── SKILL.md
│   └── references/
│
├── my-skill-claude.zip          # Claude package
├── my-skill-gemini.tar.gz       # Gemini package
├── my-skill-openai.zip          # OpenAI package
├── my-skill-langchain/          # LangChain directory
├── my-skill-chroma.zip          # ChromaDB package
└── my-skill-weaviate.zip        # Weaviate package

---

Troubleshooting

"Package validation failed"

Problem: SKILL.md is missing or malformed

Solution:

# Check skill structure
ls output/my-skill/

# Rebuild if needed
skill-seekers create --config my-config --skip-scrape

# Or recreate
skill-seekers create <source>

"Target platform not supported"

Problem: Typo in target name

Solution:

# Check available targets
skill-seekers package --help

# Common targets: claude, gemini, openai, langchain, chroma, weaviate

"Upload failed"

Problem: Missing API key

Solution:

# Set API key
export ANTHROPIC_API_KEY=sk-ant-...
export GOOGLE_API_KEY=AIza...
export OPENAI_API_KEY=sk-...

# Try again
skill-seekers upload output/my-skill-claude.zip --target claude

"Out of memory"

Problem: Skill too large for memory

Solution:

# Use streaming mode
skill-seekers package output/my-skill/ --streaming

# Smaller chunks
skill-seekers package output/my-skill/ --streaming --streaming-chunk-chars 1000

---

Best Practices

1. Package Once, Use Everywhere

# Create once
skill-seekers create <source>

# Package for all needed platforms
for platform in claude gemini langchain; do
  skill-seekers package output/my-skill/ --target $platform
done

2. Check Quality Before Packaging

# Validate first
skill-seekers quality output/my-skill/ --threshold 6.0

# Then package
skill-seekers package output/my-skill/

3. Use Streaming for Large Skills

# Automatically detected, but can force
skill-seekers package output/large-skill/ --streaming

4. Keep Original Skill Directory

Don't delete output/my-skill/ after packaging - you might want to:

• Re-package for other platforms
• Apply different workflows
• Update and re-enhance

---

Next Steps

• Workflows Guide - Apply workflows before packaging
• MCP Reference - Package via MCP
• Vector DB Integrations - Platform-specific guides