Agent Context Isolation
日本語の概要は準備中です。原文の説明を表示しています。
Meta-skill for creating and managing Claude Code skills
インストール方法を見るインストールする前に、エージェントに与えられる指示の中身を確認できます。
Meta-skill for creating new Claude Code skills, including skills that wrap MCP pipelines.
Skills live in .claude/skills/<skill-name>/:
.claude/skills/my-skill/
├── SKILL.md # Required: Main skill definition
├── scripts/ # Optional: Supporting scripts
└── templates/ # Optional: Templates, examples
---
name: skill-name
description: Brief description (shown in skill list)
allowed-tools: [Bash, Read, Write] # Optional: restrict tools
---
# Skill Name
## When to Use
[When Claude should discover this skill]
## Instructions
[Step-by-step instructions for Claude to follow]
## Examples
[Usage examples]
To create a new MCP chain script and wrap it as a skill:
Copy the multi-tool-pipeline template:
cp $CLAUDE_PROJECT_DIR/scripts/multi_tool_pipeline.py $CLAUDE_PROJECT_DIR/scripts/my_pipeline.py
Reference the template pattern:
cat $CLAUDE_PROJECT_DIR/.claude/skills/multi-tool-pipeline/SKILL.md
cat $CLAUDE_PROJECT_DIR/scripts/multi_tool_pipeline.py
Edit your new script to chain the MCP tools you need:
async def main():
from runtime.mcp_client import call_mcp_tool
args = parse_args()
# Chain your MCP tools (serverName__toolName)
result1 = await call_mcp_tool("server1__tool1", {"param": args.arg1})
result2 = await call_mcp_tool("server2__tool2", {"input": result1})
print(result2)
Create .claude/skills/my-pipeline/SKILL.md:
---
name: my-pipeline
description: What the pipeline does
allowed-tools: [Bash, Read]
---
# My Pipeline Skill
## When to Use
- [Trigger conditions]
## Instructions
Run the pipeline:
\`\`\`bash
uv run python -m runtime.harness scripts/my_pipeline.py --arg1 "value"
\`\`\`
### Parameters
- `--arg1`: Description
## MCP Servers Required
- server1: For tool1
- server2: For tool2
Add to .claude/skills/skill-rules.json:
{
"skills": {
"my-pipeline": {
"type": "domain",
"enforcement": "suggest",
"priority": "medium",
"description": "What it does",
"promptTriggers": {
"keywords": ["keyword1", "keyword2"],
"intentPatterns": ["(pattern).*?(match)"]
}
}
}
}
For full details, read:
cat $CLAUDE_PROJECT_DIR/.claude/rules/skill-development.md
cat $CLAUDE_PROJECT_DIR/.claude/rules/mcp-scripts.md
Look at existing skills for patterns:
ls $CLAUDE_PROJECT_DIR/.claude/skills/
cat $CLAUDE_PROJECT_DIR/.claude/skills/commit/SKILL.md
cat $CLAUDE_PROJECT_DIR/.claude/skills/firecrawl-scrape/SKILL.md
まだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
Agent Context Isolation
日本語の概要は準備中です。原文の説明を表示しています。
Agent Orchestration Rules
日本語の概要は準備中です。原文の説明を表示しています。
Agentic Workflow Pattern
日本語の概要は準備中です。原文の説明を表示しています。
Guide for integrating Agentica SDK with Claude Code CLI proxy
日本語の概要は準備中です。原文の説明を表示しています。
Reference guide for Agentica multi-agent infrastructure APIs
日本語の概要は準備中です。原文の説明を表示しています。
Write reliable prompts for Agentica/REPL agents that avoid LLM instruction ambiguity
日本語の概要は準備中です。原文の説明を表示しています。