by WILLOSCAR
Research pipelines as semantic execution units: each skill declares inputs/outputs, acceptance criteria, and guardrails. Evidence-first methodology prevents hollow writing through structured intermediate artifacts.
# Add to your Claude Code skills
git clone https://github.com/WILLOSCAR/research-units-pipeline-skillsLanguages: English | 简体中文
This project uses semantic skills to turn research workflows into reusable pipelines.
It is designed for the space between fragile prompting and overly rigid scripting. By organizing research tasks into staged pipelines with explicit artifacts, checkpoints, and guardrails, it makes complex work more reusable, inspectable, and iterative. The result is a workflow that can be resumed, audited, and continuously improved instead of being rebuilt from scratch each time.
The codebase currently centers on three workflows:
| Workflow | Use it for | Default deliverable | English | 中文 |
|---|---|---|---|---|
| latex-survey | evidence-first literature surveys with optional LaTeX/PDF delivery | output/DRAFT.md, latex/main.tex, latex/main.pdf | Guide | 说明 |
| idea-brainstorm | literature-grounded research direction discovery and discussion memos | output/REPORT.md | Guide | 说明 |
| graduate-paper | restructuring an existing Chinese graduation thesis project into a thesis engineering workflow | pipeline + thesis skill packages | Guide | 说明 |
These workflows share the same architecture:
pipelines/ defines stage contracts, artifact expectations, and required skills..codex/skills/ holds the reusable skills. (100 skills)No comments yet. Be the first to share your thoughts!
workspaces/ stores per-run artifacts and intermediate outputs.readme/ contains feature-level documentation.Pipeline: the contract for a workflow. It defines stages, artifacts, checkpoints, and required skills.Skill: a reusable capability with explicit inputs, outputs, acceptance criteria, and guardrails.Workspace: the working directory for a single run under workspaces/<name>/, where generated artifacts are written.The important design choice is artifact-first execution. The model is not expected to keep the whole workflow in memory; it writes intermediate structure, evidence, and review outputs to disk so later stages can build on them.
Use latex-survey when the goal is a serious review paper with explicit retrieval, structure review, evidence packs, writing loops, and optional PDF output.
Use idea-brainstorm when the goal is to generate a literature-backed memo of candidate research directions for discussion, not to write a paper yet.
Use graduate-paper when you already have thesis materials such as a template, existing TeX, Overleaf drafts, PDFs, figures, or prior papers, and need to reorganize them into a Chinese degree thesis workflow. This path is currently the least automated of the three.
Typical prompts:
Write a LaTeX survey about embodied AI and show me the outline first.
Brainstorm literature-grounded research ideas around embodied agents for home robotics.
Use the graduate-paper workflow to reorganize my Chinese thesis materials before rewriting chapters.
If you want tighter control, pin the pipeline directly:
pipelines/..codex/skills/ if you need to change behavior rather than just run it.Feature guides:
| Workflow | English | 中文 |
|---|---|---|
| latex-survey | readme/latex-survey.md | readme/latex-survey.zh-CN.md |
| idea-brainstorm | readme/idea-brainstorm.md | readme/idea-brainstorm.zh-CN.md |
| graduate-paper | readme/graduate-paper.md | readme/graduate-paper.zh-CN.md |
Project references:
Legacy language-specific copies of the older survey-focused README still live under readme/README.*.md. They are reference material, not the primary entrypoint.
latex-survey is the most complete writing pipeline in the repo and the main path when the deliverable is a survey paper or PDF.idea-brainstorm is structured and executable, but optimized for discussion-ready idea memos rather than paper drafting.graduate-paper now has a clearer pipeline design and a first batch of thesis-oriented skills, but it should currently be treated as a guided workflow framework rather than a fully automated thesis runner.