Skip to content

Latest commit

 

History

History
301 lines (230 loc) · 11.5 KB

File metadata and controls

301 lines (230 loc) · 11.5 KB

Claude Scholar Configuration

Project Overview

Claude Scholar - Personal Claude Code configuration system for academic research and software development

Mission: Cover the complete academic research lifecycle (from ideation to publication) and software development workflows, with plugin development and project management capabilities.


User Background

Academic Background

  • Degree: Computer Science PhD
  • Target Venues:
    • Top conferences: NeurIPS, ICML, ICLR, KDD
    • High-impact journals: Nature, Science, Cell, PNAS
  • Focus: Academic writing quality, logical coherence, natural expression

Tech Stack Preferences

Python Ecosystem:

  • Package manager: uv - modern Python package manager
  • Config management: Hydra + OmegaConf (config composition, overrides, type safety)
  • Model training: Transformers Trainer

Git Standards:

  • Commit convention: Conventional Commits
    Type: feat, fix, docs, style, refactor, perf, test, chore
    Scope: data, model, config, trainer, utils, workflow
    
  • Branch strategy: master/develop/feature/bugfix/hotfix/release
  • Merge strategy: rebase for feature branch sync, merge --no-ff for integration

Global Configuration

Language Settings

  • Respond in English to the user
  • Keep technical terms in English (e.g. NeurIPS, RLHF, TDD, Git)
  • Do not translate proper nouns or names

Working Directory Standards

  • Plan documents: /plan folder
  • Temporary files: /temp folder
  • Auto-create folders if they don't exist

Task Execution Principles

  • Discuss approach before breaking down complex tasks
  • Run example tests after implementation
  • Make backups, avoid breaking existing functionality
  • Clean up temporary files after completion

Work Style

  • Task management: Use TodoWrite to track progress, plan before executing complex tasks, prefer existing skills
  • Communication: Ask proactively when uncertain, confirm before important operations, follow hook-enforced workflows
  • Code style: Python follows PEP 8, comments in English, identifiers in English

Core Workflows

Research Lifecycle (7 Stages)

Ideation → ML Development → Experiment Analysis → Paper Writing → Self-Review → Submission/Rebuttal → Post-Acceptance
Stage Core Tools Commands
1. Research Ideation research-ideation skill + literature-reviewer agent + Zotero MCP /research-init, /zotero-review, /zotero-notes
2. ML Project Dev architecture-design skill + code-reviewer agent /plan, /commit, /tdd
3. Experiment Analysis results-analysis skill + data-analyst agent /analyze-results
4. Paper Writing ml-paper-writing skill + paper-miner agent -
5. Self-Review paper-self-review skill -
6. Submission & Rebuttal review-response skill + rebuttal-writer agent /rebuttal
7. Post-Acceptance post-acceptance skill /presentation, /poster, /promote

Supporting Workflows

  • Automation: 5 Hooks auto-trigger at session lifecycle stages (skill evaluation, env init, work summary, security check)
  • Zotero Integration: Automated paper import, collection management, full-text reading, and citation export via Zotero MCP
  • Knowledge Extraction: paper-miner and kaggle-miner agents continuously extract knowledge from papers and competitions
  • Skill Evolution: skill-developmentskill-quality-reviewerskill-improver three-step improvement loop

Skills Directory (32 skills)

🔬 Research & Analysis (4 skills)

  • research-ideation: Research startup (5W1H, literature review, gap analysis, research question formulation, Zotero integration)
  • results-analysis: Experiment result analysis (statistical tests, visualization, ablation studies)
  • citation-verification: Citation verification (multi-layer: format→API→info→content)
  • daily-paper-generator: Daily paper generator for research tracking

📝 Paper Writing & Publication (7 skills)

  • ml-paper-writing: ML/AI paper writing assistance
    • Top conferences: NeurIPS, ICML, ICLR, ACL, AAAI, COLM
    • Journals: Nature, Science, Cell, PNAS
  • writing-anti-ai: Remove AI writing patterns, bilingual (Chinese/English)
  • paper-self-review: Paper self-review (6-item quality checklist)
  • review-response: Systematic rebuttal writing
  • post-acceptance: Post-acceptance processing (presentation, poster, promotion)
  • doc-coauthoring: Document co-authoring workflow
  • latex-conference-template-organizer: LaTeX conference template organization

💻 Development Workflows (6 skills)

  • daily-coding: Daily coding checklist (minimal mode, auto-triggered)
  • git-workflow: Git workflow standards (Conventional Commits, branch management)
  • code-review-excellence: Code review best practices
  • bug-detective: Debugging and error investigation (Python, Bash/Zsh, JavaScript/TypeScript)
  • architecture-design: ML project code architecture and design patterns
  • verification-loop: Verification loops and testing

🔌 Plugin Development (8 skills)

  • skill-development: Skill development guide
  • skill-improver: Skill improvement tool
  • skill-quality-reviewer: Skill quality review
  • command-development: Slash command development
  • command-name: Plugin structure guide
  • agent-identifier: Agent development configuration
  • hook-development: Hook development and event handling
  • mcp-integration: MCP server integration

🧪 Tools & Utilities (4 skills)

  • planning-with-files: Planning and progress tracking with Markdown files
  • uv-package-manager: uv package manager usage
  • webapp-testing: Local web application testing
  • kaggle-learner: Kaggle competition learning

🎨 Web Design (3 skills)

  • frontend-design: Create distinctive, production-grade frontend interfaces
  • ui-ux-pro-max: UI/UX design intelligence (50+ styles, 97 palettes, 57 font pairings, 9 stacks)
  • web-design-reviewer: Visual website inspection for responsive, accessibility, and layout issues

Commands (50+ Commands)

Research Workflow Commands

Command Function
/research-init Start Zotero-integrated research ideation workflow (auto-create collections, import papers, full-text analysis)
/zotero-review Read papers from Zotero collection, generate structured literature review
/zotero-notes Batch read Zotero papers, generate structured reading notes
/analyze-results Analyze experiment results (statistical tests, visualization, ablation)
/rebuttal Generate systematic rebuttal document
/presentation Create conference presentation outline
/poster Generate academic poster design
/promote Generate promotion content (Twitter, LinkedIn, blog)

Development Workflow Commands

Command Function
/plan Create implementation plan
/commit Commit code (following Conventional Commits)
/update-github Commit and push to GitHub
/update-readme Update README documentation
/code-review Code review
/tdd Test-driven development workflow
/build-fix Fix build errors
/verify Verify changes
/checkpoint Create checkpoint
/refactor-clean Refactor and clean up
/learn Extract reusable patterns from code
/create_project Create new project
/setup-pm Configure package manager (uv/pnpm)
/update-memory Check and update CLAUDE.md memory

SuperClaude Command Suite (/sc)

  • /sc agent - Agent dispatch
  • /sc analyze - Code analysis
  • /sc brainstorm - Interactive brainstorming
  • /sc build - Build project
  • /sc business-panel - Business panel
  • /sc cleanup - Code cleanup
  • /sc design - System design
  • /sc document - Generate documentation
  • /sc estimate - Effort estimation
  • /sc explain - Code explanation
  • /sc git - Git operations
  • /sc help - Help info
  • /sc implement - Feature implementation
  • /sc improve - Code improvement
  • /sc index - Project index
  • /sc index-repo - Repository index
  • /sc load - Load context
  • /sc pm - Package manager operations
  • /sc recommend - Recommend solutions
  • /sc reflect - Reflection summary
  • /sc research - Technical research
  • /sc save - Save context
  • /sc select-tool - Tool selection
  • /sc spawn - Spawn subtasks
  • /sc spec-panel - Spec panel
  • /sc task - Task management
  • /sc test - Test execution
  • /sc troubleshoot - Issue troubleshooting
  • /sc workflow - Workflow management

Agents (14 Agents)

Research Workflow Agents

  • literature-reviewer - Literature search, classification, and trend analysis (Zotero MCP integration: auto-import, full-text reading)
  • data-analyst - Automated data analysis and visualization
  • rebuttal-writer - Systematic rebuttal writing with tone optimization
  • paper-miner - Extract writing knowledge from successful papers

Development Workflow Agents

  • architect - System architecture design
  • build-error-resolver - Build error fixing
  • bug-analyzer - Deep code execution flow analysis and root cause investigation
  • code-reviewer - Code review
  • dev-planner - Development task planning and breakdown
  • refactor-cleaner - Code refactoring and cleanup
  • tdd-guide - TDD workflow guidance
  • kaggle-miner - Extract engineering practices from Kaggle solutions

Design & Content Agents

  • ui-sketcher - UI blueprint design and interaction specifications
  • story-generator - User story and requirement generation

Hooks (5 Hooks)

Cross-platform Node.js hooks for automated workflow execution:

Hook Trigger Function
session-start.js Session start Show Git status, todos, available commands
skill-forced-eval.js Every user input Force evaluate all available skills
session-summary.js Session end Generate work log, detect CLAUDE.md updates
stop-summary.js Session stop Quick status check, temp file detection
security-guard.js File operations Security validation (key detection, dangerous command interception)

Rules (4 Rules)

Global constraints, always active:

Rule File Purpose
coding-style.md ML project code standards: 200-400 line files, immutable config, type hints, Factory & Registry patterns
agents.md Agent orchestration: auto-invocation timing, parallel execution, multi-perspective analysis
security.md Security standards: key management, sensitive file protection, pre-commit security checks
experiment-reproducibility.md Experiment reproducibility: random seeds, config recording, environment recording, checkpoint management

Naming Conventions

Skill Naming

  • Format: kebab-case (lowercase + hyphens)
  • Form: prefer gerund form (verb+ing)
  • Example: scientific-writing, git-workflow, bug-detective

Tags Naming

  • Format: Title Case
  • Abbreviations all caps: TDD, RLHF, NeurIPS, ICLR
  • Example: [Writing, Research, Academic]

Description Standards

  • Person: third person
  • Content: include purpose and use cases
  • Example: "Provides guidance for academic paper writing, covering top-venue submission requirements"

Task Completion Summary

After each task, proactively provide a brief summary:

📋 Operation Review
1. [Main operation]
2. [Modified files]

📊 Current Status
• [Git/filesystem/runtime status]

💡 Next Steps
1. [Targeted suggestions]