autonomous-research-pipeline

Claude Code Plugins

Production-tested plugins extracted from a 20-domain autonomous agent system. Each plugin is a self-contained skill that can be installed into any Claude Code project.

本番環境で稼働中の20ドメイン自律エージェントシステムから抽出したClaude Codeプラグイン集です。各プラグインは独立したスキルとして、任意のClaude Codeプロジェクトにインストールできます。

---

Quick Install / インストール

# Add the marketplace
/plugin marketplace add MtKana/claude-code-plugins

# Install a specific plugin
/plugin install note-com-publisher@claude-code-plugins
/plugin install telegram-approval-loop@claude-code-plugins
/plugin install grok-x-search@claude-code-plugins
/plugin install autonomous-research-pipeline@claude-code-plugins
/plugin install daily-cron-agent@claude-code-plugins
/plugin install note-com-watercolor-header@claude-code-plugins

---

Plugins

1. note-com-publisher

Publish articles to note.com as drafts via their undocumented API.

note.comの非公開APIを使って記事をドラフトとして公開するプラグインです。

Features:

• Email/password authentication (no browser needed)
• Markdown → HTML conversion
• Eyecatch (header image) upload
• 2-step draft creation (create → save)
• Cookie-based session management

Requirements: requests (pip)

Environment variables:

[email protected]
NOTE_PASSWORD=your_password

Usage:

python publish_to_note.py \
  --title "記事タイトル" \
  --body body.txt \
  --eyecatch header.jpg \
  --output result.json

---

2. telegram-approval-loop

Human-in-the-loop approval workflow via Telegram Bot API. The hardest part of any autonomous agent system is the approval step — this solves it.

Telegram Bot APIを使った人間によるレビュー・承認ワークフローです。自律エージェントシステムで最も難しい「承認ステップ」を解決します。

Features:

• Inline keyboard buttons (Approve / Edit)
• Auto-chunking for messages over 4,096 characters
• Configurable callback data
• Stdlib-only — zero pip dependencies

Environment variables:

TELEGRAM_BOT_TOKEN=your_bot_token
TELEGRAM_CHAT_ID=your_chat_id

Usage:

# Send an approval request
python send_approval.py \
  --title "Draft Review" \
  --body "Article content here..." \
  --approve-data "approve_123" \
  --edit-data "edit_123"

---

3. grok-x-search

Search Twitter/X via Grok API's native x_search tool. Most people don't know Grok has a built-in Twitter search — this wraps it for structured results.

Grok APIのネイティブx_searchツールを使ってTwitter/Xを検索します。構造化された結果を返します。

Features:

• Native Twitter/X search via Grok API (not scraping)
• Configurable queries via CLI args or JSON file
• Structured JSON output with source URLs
• Focuses on recent posts (last 48 hours)

Requirements: httpx (pip)

Environment variables:

XAI_API_KEY=your_xai_api_key

Usage:

# Single query
python grok_search.py \
  --query "Claude Code best practices" \
  --output results.json

# Multiple queries from file
python grok_search.py \
  --queries-file queries.json \
  --output results.json

queries.json format:

["query one", "query two", "query three"]

---

4. autonomous-research-pipeline

Template and architecture reference for building multi-track parallel research agents with Claude Code CLI. This is the pattern, not a script — adapt it to your domain.

Claude Code CLIを使ったマルチトラック並列リサーチエージェントのアーキテクチャテンプレートです。パターンそのものが価値です。

What's included:

• SKILL.md — Complete 4-track parallel research workflow template
• reference/architecture.md — Design principles and phase execution model
• reference/example_queries.md — Query templates across 6 research categories

Architecture:

Phase 1: Parallel Discovery (4 tracks, run simultaneously)
    Track A: Web search (platforms, tools, trends)
    Track B: Community signals (Reddit, forums, Twitter)
    Track C: Academic/industry (papers, reports)
    Track D: Domain-specific API (Grok x_search, etc.)

Phase 2: Cross-pollination + Synthesis
    Merge findings → identify patterns → score ideas

Phase 3: Output Generation
    Write report / article / brief → notify

How to use: Read the architecture docs, copy the SKILL.md template, customize the tracks and queries for your domain.

---

5. daily-cron-agent

Battle-tested pattern for running Claude Code CLI as a daily autonomous agent via cron. This runs 12+ pipelines daily in production.

Claude Code CLIをcronで毎日自律的に実行するためのパターンです。本番環境で12以上のパイプラインを毎日稼働させています。

What's included:

• SKILL.md — Setup guide and best practices
• reference/cron_template.sh — Ready-to-use shell script template
• reference/vps_setup.md — VPS setup guide (Ubuntu)

Key patterns:

• Parallel phase execution with PID management and wait
• Structured logging with timestamps
• --allowedTools for security (restrict which tools Claude can use)
• --max-turns to prevent runaway sessions
• Environment variable management for cron context

Template usage:

# Copy and customize the template
cp cron_template.sh my_pipeline.sh
chmod +x my_pipeline.sh