langchain-skills-adapter 0.1.0

Anthropic Agent Skills implementation for LangChain

LangChain Skills Adapter

This package provides an easy way to work with Agent Skills in your LangChain/LangGraph project by treating Skills as any other tool in LangChain, allowing you to easily integrate skills with existing workflows and agents. For developers, Skills work like any other LangChain tool.

Create SkillTool

SkillTool accepts 2 arguments: directories (required) and description_template (optional).

Supported Directory Structures:

Scenario 1 - Direct skill directory:

skills/
└── pdf/
    └── SKILL.md          # Skill name: "pdf"

Usage: SkillTool(directories="skills/pdf")

Scenario 2 - Parent directory with multiple skills:

.copilot/skills/
├── pdf/
│   └── SKILL.md          # Skill name: "pdf"
├── pptx/
│   └── SKILL.md          # Skill name: "pptx"
└── excel/
    └── SKILL.md          # Skill name: "excel"

Usage: SkillTool(directories=".copilot/skills") # Discovers all three skills

Scenario 3 - Nested skill directories:

skills/
├── documents/
│   ├── pdf/
│   │   └── SKILL.md      # Skill name: "pdf"
│   └── word/
│       └── SKILL.md      # Skill name: "word"
└── data/
    └── csv/
        └── SKILL.md      # Skill name: "csv"

Usage: SkillTool(directories="skills") # Recursively discovers all nested skills

Scenario 4 - Multiple parent directories:

~/.agents/skills/
└── pdf/
    └── SKILL.md          # Skill name: "pdf"

./custom-skills/
└── excel/
    └── SKILL.md          # Skill name: "excel"

Usage: SkillTool(directories=["~/.agents/skills", "./custom-skills"])

from langchain_skills import SkillTool

# Load skills from a directory
skill_tool = SkillTool(directories=".agent/skills/")

Example Usage

from langchain_community.agent_toolkits import FileManagementToolkit
from langchain_community.tools import ShellTool

from langchain_skills import SkillTool

# Load skills from a directory
skill_tool = SkillTool(directories=".agent/skills/")

# Create tools
file_management_toolkit = FileManagementToolkit(root_dir='..').get_tools()
bash_tool = ShellTool()

# Combine all tools
tools = file_management_toolkit + [bash_tool, skill_tool]

agent = create_agent(model, tools)

It is highly recommended to provide FileManagementToolkit and ShellTool, which act as supporting tools in case your skill is not just a SKILL.md file. Even if the skill is a simple single SKILL.md file, it depends on what is actually written in that SKILL.md file. It depends on your use case.

Important:

Since this setup makes use of FileSystemTools and ShellTool, ensure proper security and safety measures. This is not the responsibility of this library.

Inspiration

This work is inspired by Spring AI's SkillTool implementation:
https://github.com/spring-ai-community/spring-ai-agent-utils/blob/main/spring-ai-agent-utils/src/main/java/org/springaicommunity/agent/tools/SkillsTool.java