メニュー
This skill should be used when the user asks to 'create an agent', 'add tool calling', 'build workflows', or 'orchestrate AI tasks'. Guides agentic AI patterns and multi-step reasoning.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/aaronmaturen/rumpleskill --skill agentsコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
SKILL.md を表示中
This skill should be used when the user asks to 'integrate an API', 'handle streaming responses', 'add authentication', or 'manage API calls'. Covers Claude API integration and streaming patterns.
This skill should be used when the user asks to 'add a command', 'parse arguments', 'improve CLI output', or 'add interactive features'. Covers command routing, argument parsing, and stdio handling.
This skill should be used when the user asks to 'scan directories', 'read files', 'parse JSON or Markdown', or 'traverse the file system'. Guides safe file I/O patterns and recursive directory operations.
SOC 職業分類に基づく
| name | agents |
| description | This skill should be used when the user asks to 'create an agent', 'add tool calling', 'build workflows', or 'orchestrate AI tasks'. Guides agentic AI patterns and multi-step reasoning. |
| version | 1.0.0 |
| metadata | {"internal":false} |
Patterns for building AI agents with tool calling, streaming responses, and multi-step reasoning workflows.
import Anthropic from "@anthropic-ai/sdk";
client = ({
: process..,
});
stream = client..({
: ,
: ,
: systemPrompt,
: [{ : , : userContext }],
: ,
});
Use Sonnet 4.5 as default. Adjust max_tokens based on expected output length.
// src/utils/claude.ts pattern
export async function callClaude(
systemPrompt: string,
userContext: string,
verbose = false
): Promise<string> {
const stream = await client.messages.create({
model: "claude-sonnet-4-5-20250929",
max_tokens: 16000,
system: systemPrompt,
messages: [{ role: "user", content: userContext }],
stream: true,
});
let fullResponse = "";
for await (const event of stream) {
if (event.type === "content_block_delta" && event.delta.type === "text_delta") {
const text = event.delta.text;
fullResponse += text;
if (verbose) {
process.stderr.write(text); // Real-time feedback
}
}
}
return fullResponse;
}
Verbose mode writes to stderr so stdout remains clean for piping.
// src/prompts/claude-md.ts pattern
export const CLAUDE_MD_PROMPT = `
You are a **Senior Technical Writer** generating a focused skill file.
## OUTPUT FORMAT
\`\`\`markdown
---
name: {skill-name}
description: This skill should be used when the user asks to {actions}.
version: 1.0.0
---
# {Title}
...
\`\`\`
## GUIDELINES
- Be concise, Claude is already smart
- Use trigger phrases in description
- Include copy-paste ready code examples
`;
Structure prompts with:
// src/generators/my-skill.ts
import { callClaude } from "../utils/claude.js";
import { MY_SKILL_PROMPT } from "../prompts/my-skill.js";
import { scanDirectory, readFileContent } from "../utils/file-system.js";
export async function generateMySkill(verbose = false): Promise<string> {
// 1. Gather context from codebase
const files = await scanDirectory(".");
const packageJson = await readFileContent("package.json");
const context = `
## Project Files
${files.map((f) => `- ${f}`).join("\n")}
## Dependencies
${packageJson}
`;
// 2. Call Claude with prompt + context
return await callClaude(MY_SKILL_PROMPT, context, verbose);
}
Generators are pure async functions that return markdown strings.
// src/index.ts - Adding new agent command
const command = process.argv[2];
const isVerbose = process.argv.includes("--verbose");
if (command === "my-skill") {
const result = await generateMySkill(isVerbose);
console.log(result);
process.exit(0);
}
Use direct process.argv parsing. Keep CLI logic minimal.
// src/utils/file-system.ts patterns
import { readdir, readFile } from "fs/promises";
import { join } from "path";
export async function scanDirectory(
dir: string,
ignored = ["node_modules", ".git", "dist"]
): Promise<string[]> {
const entries = await readdir(dir, { withFileTypes: true });
const files: string[] = [];
for (const entry of entries) {
if (ignored.includes(entry.name) || entry.name.startsWith(".")) {
continue;
}
const fullPath = join(dir, entry.name);
if (entry.isDirectory()) {
files.push(...(await scanDirectory(fullPath, ignored)));
} else {
files.push(fullPath);
}
}
return files;
}
Always exclude build artifacts and hidden files.
src/prompts/ and src/generators/ distinct for versioningstream: true for all user-facing outputs to show progress.js extensions in imports, top-level await, and native Node.js APIsThis project doesn't yet implement Claude's tool use API, but the pattern would be:
const response = await client.messages.create({
model: "claude-sonnet-4-5-20250929",
max_tokens: 4096,
tools: [
{
name: "scan_directory",
description: "Recursively scans a directory",
input_schema: {
type: "object",
properties: {
path: { type: "string" },
},
required: ["path"],
},
},
],
messages: [{ role: "user", content: "Analyze the src/ directory" }],
});
// Check for tool_use blocks in response.content
Consider adding tool calling for interactive agents that need file system access.