SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsresearchclaw-cn

researchclaw-cn

Writing & research

View repository

Updated 10-15 23:39

About this skill

ResearchClaw Skill — Autonomous Research Pipeline (Chinese Version)

Important note: This skill requires installing AutoResearchClaw。 This skill is a wrapper layer and cannot run independently. Please install the upstream project first, then install this skill.

This skill wraps AutoResearchClaw, a 23-stage autonomous research pipeline. Enter a research topic and it automatically completes the entire process from literature review to paper generation, including real citation retrieval, sandbox experiment execution, multi-agent peer review, and citation verification.

Honesty principle: This skill will not fabricate functionality. Every command maps to a real upstream feature. If some features fail, the skill will report the actual error and suggest specific fixes — never pretending the problem does not exist.

Command list

Command Function
/researchclaw Show help information and available subcommands
/researchclaw:setup Check and install all prerequisite dependencies (Python, Docker, LaTeX, pip packages)
/researchclaw:config Interactive configuration wizard — generate a usable config.yaml
/researchclaw:run Start the research pipeline
/researchclaw:status View pipeline run status
/researchclaw:resume Resume the run from the last successful stage
/researchclaw:diagnose Automatically detect and explain common faults
/researchclaw:validate Validate configuration, dependencies, and connectivity before running

/researchclaw — help

When called without a subcommand, display the command list and status summary:

  1. Check whether the researchclaw CLI is installed:which researchclaw
  2. Check whether the current directory has config.yaml
  3. Print the command table above
  4. Suggest next steps based on missing components

/researchclaw:setup — prerequisite installation

Required: obtain user consent before installing anything. Show missing items and obtain explicit approval.

Run the prerequisite check script:

bash "${CLAUDE_SKILL_DIR}/scripts/check-prereqs.sh"

The script checks each dependency and outputs a JSON report. Based on the report:

  1. Python 3.11+: Check python3 --version. If missing or the version is too old, suggest pyenv install 3.11 or the system package manager.
  2. pip / uv: Check pip3 --version or uv --version. If there is no uv, suggest installing it (faster).
  3. Docker: Check docker info. If the Docker daemon is not running, truthfully inform the user — this skill cannot start Docker on most systems.
  4. LaTeX: Check pdflatex --version. If missing, suggest sudo apt-get install texlive-full(Linux) or brew install --cask mactex（macOS）。State truthfully: this is a large download (2-4 GB).
  5. AutoResearchClaw: Check pip3 show researchclaw. If not installed: 
    pip3 install researchclaw
    or install from source: 
    git clone https://github.com/aiming-lab/AutoResearchClaw.git
cd AutoResearchClaw
pip3 install -e ".[all]"

Users in mainland China: Use domestic mirrors to speed up installation:

pip install researchclaw[all] -i https://pypi.tuna.tsinghua.edu.cn/simple

Things this skill cannot do:

  • Start the Docker daemon (requires system-level privileges)
  • Install LaTeX on Linux without sudo
  • Fix network/firewall issues that block API access
  • Provide LLM API keys — users must provide them themselves

/researchclaw:config — interactive configuration wizard

Generate a usable configuration by asking the user in batches config.yaml. Use for each batch AskUserQuestion。

First batch — Essential settings (must ask):

  1. Research topic: What do you want to research? (free text)
  2. LLM provider: Which LLM API to use? Options:openai、anthropic、azure、deepseek、local
  3. API key: Provide the API key, or the name of the environment variable that stores the key (e.g., OPENAI_API_KEY）
  4. Model: Which model to use? Suggested defaults by provider:
    • openai：gpt-4o
    • anthropic：claude-sonnet-4-20250514
    • deepseek：deepseek-chat

Second batch — Experiment settings (ask with smart defaults):

  1. Experiment mode：simulated(no code execution, fastest),sandbox(local execution), or ssh_remote(GPU server). Default:simulated
  2. Automatic approval gating: Skip manual approval for stages 5, 9, 20? Default: choose true
  3. Output directory: Where to save artifacts. Default:artifacts/

Third batch — Optional advanced settings (provided but not required):

  1. Paper template：neurips、icml、iclr or generic. Default:neurips
  2. Maximum iterations: Iterative pipeline mode. Default:3
  3. Literature source：arxiv、semantic_scholar or both. Default:both

After collecting answers, use assets/config-template.yaml to generate config.yamlfrom the template. Write to the current directory and show it to the user.

/researchclaw:run — Execute pipeline

Pre-flight checks (must run before launch):

  1. Silently run /researchclaw:validate logic
  2. If any check fails, report it and ask the user whether to continue or fix it first

Launch pipeline:

researchclaw run --topic "$ARGUMENTS" --config config.yaml --auto-approve 2>&1 | tee researchclaw-run.log

If $ARGUMENTS is empty, read the topic from config.yaml .

During execution:

  • The pipeline runs 23 stages. Each stage produces output in artifacts/<run-id>/stage-N/ produces output in
  • Monitor progress by checking whether the stage directory exists
  • If the pipeline fails, capture the error output and automatically run /researchclaw:diagnose logic

After completion:

  • Report which stages succeeded and which failed
  • Show the path to the generated paper
  • Show total execution time

/researchclaw:status — Pipeline status

Check the current status of the pipeline run:

  1. Find the latest artifacts/rc-* directory
  2. Count the number of completed stages
  3. Check pipeline_summary.json— if it exists, the run is complete
  4. Report:阶段 X/23 已完成。当前阶段：[阶段名称]。状态：[运行中/失败/完成]

Stage name mapping:

Stage Name
1 Topic initialization
2 Problem decomposition
3 Literature search
4 Literature analysis
5 Research direction (gated)
6 Hypothesis generation
7 Experiment design
8 Experiment review
9 Experiment approval (gated)
10 Code generation
11 Code review
12 Experiment execution
13 Result collection
14 Result analysis
15 Paper outline
16 Section writing
17 Paper first draft
18 Peer review
19 Revision
20 Final review (gating)
21 Citation verification
22 Visualization
23 Final export

/researchclaw:resume — resume a failed run

Resume from the last successful stage:

  1. Find the latest run directory
  2. Find the last completed stage
  3. Run: 
    researchclaw run --config config.yaml --from-stage STAGE_NAME --output <run-dir> --auto-approve

Known issues (upstream): --from-stage flag may not work correctly in all versions. If resume fails, tell the user honestly and suggest starting over.

/researchclaw:diagnose — automatically diagnose failures

Read the most recent logs and error output to identify the problem:

tail -100 researchclaw-run.log 2>/dev/null || tail -100 researchclaw-resume.log 2>/dev/null

Common failure patterns and fixes:

Error pattern Cause Fix
HTTP 401 API key invalid or expired Check config.yaml → llm.api_key or environment variable
HTTP 429 API rate limit Wait 60 seconds and resume, or switch models
Stage 10 failed Code generation produced invalid Python Check the generated code, use a stronger model
Docker error Docker not running or insufficient permissions Run docker info to verify
pdflatex not found LaTeX not installed Install texlive-full
quality_score < threshold Quality gate too strict Lower quality.min_score(default 2.0 is strict)
MemoryError Insufficient memory (32GB+ required) Use simulated mode

/researchclaw:validate — validate before running

Run all checks without starting the pipeline:

  1. Config syntax: verify YAML is parseable
  2. Config completeness: check that API key and research topic are set
  3. API connectivity: test the LLM endpoint
  4. Docker health：docker info(if in sandbox mode)
  5. Disk space: warn when below 10 GB
  6. Write permissions: test that artifacts/ is writable

Report results as a pass/fail checklist.

Special notes for users in mainland China

DeepSeek is recommended

DeepSeek is more stable to access in mainland China:

llm:
  provider: deepseek
  model: deepseek-chat
  api_key_env: DEEPSEEK_API_KEY

Network proxy

If arXiv or Semantic Scholar is inaccessible:

export HTTP_PROXY=http://your-proxy:port
export HTTPS_PROXY=http://your-proxy:port

Mirror acceleration

pip install researchclaw[all] -i https://pypi.tuna.tsinghua.edu.cn/simple

Principles

  1. Never lie. If a feature is broken, say so honestly. If a feature does not exist upstream, do not pretend it does.
  2. Always test. Run validation before every pipeline execution. Check results after every operation.
  3. Ask before acting. Never install packages, modify configuration, or start long-running processes without the user's explicit approval.
  4. Report honestly. Show actual error messages, not beautified summaries.
  5. Stay updated. This skill targets AutoResearchClaw v0.3.x.