ref

Adds, analyzes, and synthesizes reference papers and their code repositories into a structured refs/ library.

Updated Aug 26, 2026
One-click install
npx skills add https://github.com/coollx/stable-harness --skill ref-coollx
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: ref
Source: https://github.com/coollx/stable-harness/tree/main/.claude/skills/ref
Command: npx skills add https://github.com/coollx/stable-harness --skill ref-coollx

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Managing research references manually—downloading papers, writing structured analyses, cloning and dissecting code repos, and keeping a cross-paper index consistent—is slow and error-prone. This Skill turns every reference operation into a repeatable workflow with a single entry point. ## Core Features & Use Cases - Add a reference: Given a PDF path, arXiv ID/URL, any URL, or just a title, it downloads the paper, classifies it as method or empirical, writes a structured paper_analysis.md, clones and analyzes the code repo, and updates refs/index.md. - Literature review: Synthesizes findings across multiple papers into refs/lit-review/ documents covering agreements, conflicts, and implications for the project. - Compare and quick-find: Compares how cloned repos implement a component, or pinpoints where a specific mechanism lives in a repo with file:line references. - Use Case: A researcher says "add the Adaptive Planner paper"—the Skill fetches the PDF, produces a rigorous analysis following a PhD-level prompt, clones the GitHub repo, maps its 10 key components (reward function, RL loop, token masking, etc.), and indexes everything. ## Quick Start Ask the assistant to add a reference by saying "add ref" followed by a paper title, arXiv ID, URL, or local PDF path.

Frequently Asked Questions about ref

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I add a research paper to my reference library?

Invoke the add mode with a local PDF path, arXiv ID or URL, any web URL, or just a paper title. The skill resolves the input to a PDF, creates a folder under refs/related-work/, writes a structured analysis, clones the code repo if one exists, and updates refs/index.md.

How do I analyze a paper's GitHub repository?

The repo-analysis protocol orients on the repo structure, then examines 10 key components including entry points, special token handling, reward functions, RL training loops, and evaluation. Output is a repo_analysis.md with file:line locations, formulas, and reuse recommendations.

What is the difference between method and empirical paper analysis?

Method papers get a prompt focused on pipeline design, architecture, formulas, and reproduction steps. Empirical papers get a prompt focused on research questions, study design, findings with evidence, and limitations. The skill auto-classifies and announces its choice.

Can I compare how two repos implement the same component?

Yes, compare mode builds a table across cloned repos in refs/related-work/*/repo/ with rows for aspect, file:line location, method, and key differences. It reuses existing repo_analysis.md files first and only re-greps code for uncovered details.

What happens if a paper has no public code?

The skill searches the paper text and the web for a repository URL. If none is found, it skips the repo analysis step, notes "no public code" in the paper analysis and the index, and finishes with the paper analysis alone.

Does the skill check OpenReview for paper reviews?

Yes, both analysis prompts end with a Reviewer Reception section that searches OpenReview by title and venue. If a public page exists it summarizes the decision, criticisms, and rebuttal; otherwise it records that no page was found without fabricating reviews.