Skip to content

04-multi-platform.md

Latest commit

 

History

History
207 lines (142 loc) · 4.19 KB

04-multi-platform.md

File metadata and controls

207 lines (142 loc) · 4.19 KB
Raw

Tutorial 4: Multi-Platform Publishing

Deploy your skills to Claude, OpenAI, and LangChain from a single source.

Overview

SkillForge can publish skills to multiple AI platforms:

Platform Modes Use Case
Claude code, api, project Claude Code, Claude API, claude.ai
OpenAI gpt, assistant, api Custom GPTs, Assistants API
LangChain hub, module, json LangChain Hub, Python code

Publishing to Claude

Claude Code (Default)

# Install directly to Claude Code
skillforge publish ./my-skill --platform claude

# Same as: skillforge install ./my-skill

Claude API

# Generate API-ready system prompt
skillforge publish ./my-skill --platform claude --mode api

# Output: claude_system_prompt.txt

claude.ai Project

# Generate project knowledge format
skillforge publish ./my-skill --platform claude --mode project

# Output: project_knowledge.md

Publishing to OpenAI

Custom GPT

skillforge publish ./my-skill --platform openai --mode gpt

# Output:
# gpt_config.json - GPT configuration
# gpt_instructions.txt - System instructions
# conversation_starters.json - Suggested prompts

Upload these to chat.openai.com/gpts/editor.

Assistants API

skillforge publish ./my-skill --platform openai --mode assistant

# Output: assistant_config.json

Use with the OpenAI Assistants API:

from openai import OpenAI
import json

client = OpenAI()

with open("assistant_config.json") as f:
    config = json.load(f)

assistant = client.beta.assistants.create(**config)

Publishing to LangChain

LangChain Hub

# Push to LangChain Hub (requires langchain-hub)
skillforge publish ./my-skill --platform langchain --mode hub

# Requires: pip install langchain-hub
# And: export LANGCHAIN_API_KEY=...

Python Module

skillforge publish ./my-skill --platform langchain --mode module

# Output: my_skill_prompt.py

Generated module:

from langchain_core.prompts import ChatPromptTemplate

my_skill_prompt = ChatPromptTemplate.from_messages([
    ("system", """You are an expert..."""),
    ("human", "{input}")
])

# Usage:
# chain = my_skill_prompt | llm
# result = chain.invoke({"input": "user message"})

JSON Config

skillforge publish ./my-skill --platform langchain --mode json

# Output: langchain_prompt.json

Dry Run Mode

Preview publishing without making changes:

skillforge publish ./my-skill --platform openai --mode gpt --dry-run

# Shows what would be generated without writing files

Publish to All Platforms

skillforge publish ./my-skill --all

# Generates output for all platforms:
# - claude/
# - openai/
# - langchain/

Platform-Specific Considerations

Claude

  • Supports rich markdown formatting
  • Can include reference documents
  • Best for complex, nuanced instructions

OpenAI GPTs

  • Conversation starters are auto-generated from examples
  • Consider enabling Code Interpreter for technical skills
  • 8,000 character limit for instructions

LangChain

  • Template variables use {variable} syntax
  • Consider adding input/output parsers
  • Works with any LangChain-compatible LLM

Example Workflow

# 1. Create and test your skill
skillforge generate "Review pull requests for code quality"
skillforge test ./skills/pr-reviewer

# 2. Security scan
skillforge security scan ./skills/pr-reviewer

# 3. Preview for each platform
skillforge publish ./skills/pr-reviewer --platform claude --dry-run
skillforge publish ./skills/pr-reviewer --platform openai --mode gpt --dry-run

# 4. Publish
skillforge publish ./skills/pr-reviewer --platform claude
skillforge publish ./skills/pr-reviewer --platform openai --mode gpt

Viewing Available Platforms

skillforge platforms

# Output:
# Available Platforms:
#
# claude
#   Modes: code, api, project
#   Description: Anthropic Claude integration
#
# openai
#   Modes: gpt, assistant, api
#   Description: OpenAI GPT and Assistants
#
# langchain
#   Modes: hub, module, json
#   Description: LangChain prompt templates

Next Steps