LMDHQ-0420/ResearchPilot-Skills

ResearchPilot‑Skills – An AI‑driven research assistant

What it is – A collection of Claude/Code‑buddy “skills” that guide a researcher through every step of a typical ML/AI paper project, from idea generation to the final manuscript. The repo ships a set of command‑style prompts (e.g. /research[A]-exploration) that the Claude‑style LLM interprets and then creates the required artefacts (markdown reports, design docs, code skeletons, figure files, LaTeX drafts, etc.). The workflow is split into seven numbered phases (A‑G) with sub‑skills for each sub‑task.

Why it matters – Academic AI research often stalls because the hand‑off between literature review, experiment design, implementation and writing is manual and fragmented. By codifying the process into reproducible “skills” the tool promises:

• Consistent documentation (design docs must be updated before code changes).
• Automatic generation of publication‑ready figures (/research[fig]).
• Structured, AI‑assisted drafting of every paper section, including bilingual translation.
• Version‑controlled logs of design, code, and manuscript revisions.

How to get started – Clone the repo and run the platform‑specific install script (install-zh.sh for Claude Code, install‑zh.sh codex for OpenAI Codex, or install‑zh.bat on Windows). The script copies 17 skill directories into the user’s ~/.claude/skills/ folder. After installation you can invoke a skill from the LLM console, for example:

/research[A]-exploration 我想做电池 SOH 预测，现有 Transformer 方法没有利用局部时序特征

The assistant will then:

1. Search at least 15 papers, build a three‑level research‑question hierarchy, and output docs/idea_report.md Part 1.
2. Prompt you to confirm before moving to the next phase.

Key components

Component	Role
docs/idea_report.md	Central research plan, built incrementally across phases A‑C.
implementation.md	Function‑level coding guide produced in phase D.
dev_log.md	Immutable development log for phases E & F.
code/fig/	Scripts and generated PDF/SVG/500 dpi PNG figures.
code/src/	Skeleton for models, data pipelines, training/evaluation scripts.
latex‑zh‑paper‑template	Optional Chinese LaTeX template used in phase G.

Supported platforms – The skills are written for Claude‑Code, OpenAI Codex CLI, and Tencent CodeBuddy. They run on macOS/Linux or Windows (via batch scripts).

License – MIT, so you can adapt the skill set or integrate it into your own research pipelines.

---

All details above are taken directly from the repository’s README; no additional features are inferred.