othmanadi/researchclaw-skill

Turn your coding agent into a one-command autonomous research paper generator.

中文文档 Chinese Documentation

---

This skill requires AutoResearchClaw to be installed. It is a wrapper that simplifies setup, configuration, execution, and troubleshooting — it does not replace the upstream project. Install AutoResearchClaw first, then install this skill on top.

---

Quick Install

npx skills add OthmanAdi/researchclaw-skill --skill researchclaw -g

中文版 / Chinese:

npx skills add OthmanAdi/researchclaw-skill --skill researchclaw-cn -g

Works with Claude Code, Cursor, Codex, Gemini CLI, OpenClaw, and any agent supporting the Agent Skills spec.

Prerequisites

You must install AutoResearchClaw before using this skill:

pip install researchclaw

Or from source (recommended for latest features):

git clone https://github.com/aiming-lab/AutoResearchClaw.git
cd AutoResearchClaw
pip install -e ".[all]"

You also need:

• Python 3.11+
• An LLM API key (OpenAI, Anthropic, DeepSeek, or any OpenAI-compatible provider)
• Docker (optional — only for sandbox experiment mode)
• LaTeX (optional — only for PDF compilation)

Commands

How It Works

You: /researchclaw:config
     ↓ answers 3 batches of questions (topic, LLM, experiment mode)
     ↓ generates config.yaml

You: /researchclaw:run "Attention mechanisms for time series forecasting"
     ↓ pre-flight validation (config, API key, Docker, disk space)
     ↓ launches AutoResearchClaw pipeline
     ↓ monitors 23 stages automatically
     ↓ auto-diagnoses failures via PostToolUse hook
     ↓ reports results with paper location

You: artifacts/rc-20260320-141523-a7f3/deliverables/paper.pdf

Custom Hooks

This skill includes 4 Claude Code hooks that run automatically — no configuration needed:

What This Skill Does NOT Do

Honesty is a core principle. This skill:

• Does not replace AutoResearchClaw — It wraps the official CLI. You must install the upstream project first.
• Does not start Docker — It checks if Docker is running, but cannot start the daemon for you
• Does not provide API keys — You must supply your own LLM API keys
• Does not fix network issues — If your firewall blocks arXiv or Semantic Scholar, the skill tells you but cannot fix it
• Does not guarantee paper quality — Output depends on the LLM model, topic complexity, and experiment mode
• Does not modify upstream code — Zero changes to AutoResearchClaw’s codebase

Project Structure

skills/                                 # Publishing source (skills.sh + npx skills add)
├── researchclaw/
│   ├── SKILL.md                        # Main skill definition (English)
│   ├── assets/
│   │   └── config-template.yaml        # Config generation template
│   ├── references/
│   │   ├── pipeline-stages.md          # All 23 stages documented
│   │   ├── config-reference.md         # Every config field explained
│   │   ├── troubleshooting.md          # 10 error patterns with fixes
│   │   └── README-CN.md               # Chinese reference
│   └── scripts/
│       ├── check-prereqs.sh            # JSON prerequisite report
│       ├── post-run-check.sh           # PostToolUse error scanner
│       ├── pre-config-write.sh         # PreToolUse config backup
│       ├── pre-delete-guard.sh         # PreToolUse artifact guard
│       └── notify-completion.sh        # Completion notification
└── researchclaw-cn/
    ├── SKILL.md                        # Chinese skill definition
    └── scripts/                        # Shared hook scripts
.claude/
├── hooks.json                          # 4 Claude Code hooks (local dev)
└── skills/                             # Local install mirror
    ├── researchclaw/
    └── researchclaw-cn/
tests/
└── test-skill.sh                       # 58-test self-validation suite
docs/
└── zh-CN/
    └── README.md                       # Full Chinese documentation
media/
├── banner.png                          # Repo banner
├── logo.png                            # Logo (researcher lobster)
└── logo_wizard.png                     # Logo variant (wizard lobster)

Validation

Evaluated using SkillCheck and a manual security audit following the same methodology used for planning-with-files.

SkillCheck (Free Tier)

Strengths detected (6/6): Example section, error handling, trigger phrases, output format, structured instructions, prerequisites documented.

Security Audit

Testing

bash tests/test-skill.sh

Validates file structure (21 checks), content quality (20 checks), script syntax (5 checks), hooks configuration (4 checks), and config template (8 checks).

58/58 tests passing.

System Requirements

Upstream

This skill wraps AutoResearchClaw v0.3.x by aiming-lab. All research pipeline functionality comes from the upstream project. This skill adds setup automation, interactive configuration, error diagnosis, and hooks — it does not fork or modify the upstream codebase.

Inspired by Karpathy’s autoresearch philosophy and the wave of autonomous research tools that followed.

Contributing

1. Run bash tests/test-skill.sh before submitting
2. Ensure all 58 tests pass
3. Add tests for new functionality
4. Keep the honesty policy — never fabricate capabilities

License

MIT — same as AutoResearchClaw upstream.