project-referee — Claude Code skill: ML paper review
A skill that reviews machine learning papers as a conference referee would: identifies weaknesses, gaps in evidence, missing citations, and delivers structured feedback tailored to the manuscript stage—early draft, in progress.
no license 1 installs ★ 152 updated 2026-07-15
npx skills add sundial-org/skills --skill project-referee -a claude-codeWhat project-referee does
The skill adjusts its approach to three stages of a manuscript. For an early draft, it focuses on novelty and feasibility of the core idea. For work in progress, it tracks development trajectory and preliminary results against claims. For final submission, it applies the full reviewer checklist: soundness, presentation, and contribution.
Internally, the skill works in layers. One layer asks stage-specific questions. Another layer searches the web for related work—recent papers in the domain, similar methods, relevant benchmarks. The third generates structured feedback: summary, strengths, weaknesses, open questions, and numerical scores (soundness, presentation, contribution on 1–4 scales, overall score 1–10, confidence 1–5).
For papers that combine large language models with symbolic reasoning, the skill includes deeper criteria: how well the symbolic formulation is constructed, which solver is chosen and why, how faithfulness is verified, and how the work compares to methods like Logic-LM.
What it helps with
- anticipate reviewer concerns before submission
- spot gaps in evidence and logical support
- identify missing or overlooked citations
- assess manuscript completeness at different stages
- get detailed structured feedback with scores
Example prompts for the agent
- Review this transformer paper as a referee—it's an early draft
- Check my paper for missing citations and weaknesses before I submit it
- Give feedback on this work in progress and flag what needs more development
How to install project-referee
1Skills CLI (Vercel)
npx skills add sundial-org/skills --skill project-referee -a claude-code2Manually: put the files in place
~/.claude/skills/project-referee/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/project-referee/
1Skills CLI (Vercel)
npx skills add sundial-org/skills --skill project-referee -a codex2Manually: put the files in place
.agents/skills/project-referee/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add sundial-org/skills --skill project-referee -a cursor2Manually: put the files in place
.cursor/skills/project-referee/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install sundial-org/skills skills/project-referee2Manually: put the files in place
.github/skills/project-referee/SKILL.md1Skills CLI (Vercel)
npx skills add sundial-org/skills --skill project-referee -a gemini-cli2Platform CLI
gemini skills install https://github.com/sundial-org/skills --path skills/project-refereeManually: put the files in place
.devin/skills/project-referee/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add sundial-org/skills --skill project-referee -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about project-referee
How do I install project-referee in Claude Code?
Run npx skills add sundial-org/skills --skill project-referee -a claude-code, then open ~/.claude/skills/project-referee/SKILL.md.
What does project-referee do?
It reviews ML papers as a conference reviewer would: finds weaknesses, verifies evidence quality, surfaces missing references, and returns a structured critique with scores.
Why does manuscript stage matter?
The skill adjusts its criteria: an early draft is judged on novelty and feasibility, a final version gets full review rigor, and in-progress work is assessed on its trajectory and completeness so far.