Commands, agents, hooks, and skills for Python-based deep learning research workflows
npx claudepluginhub jaaaackielai/deep-learning-claude-codeInteractive brainstorming skills that guide collaborative idea exploration through structured dialogue — one question at a time, multiple approaches, incremental validation
Language: English | 繁體中文
This repository includes commands, agents, and skills, which I commonly use.
I hope that this repo can help software engineers who use Python in deep learning work, and other settings can make your work flow more smoothly.
Follow the plugin-marketplaces and skills instructions provided by Anthropic.
You can install this plugin marketplace in two ways:
In Claude Code CLI:
/plugin add jaaaackieLai/deep-learning-claude-code
In Terminal:
claude plugin add jaaaackieLai/deep-learning-claude-code
Specialized agents that change how Claude works on your tasks:
| Agent | Purpose |
|---|---|
| planner | Creates phased implementation plans with dependencies and risk assessment |
| tdd-guide | Enforces test-driven development (Red/Green/Refactor) with 80%+ coverage |
| analyzer | Systematic analysis using one-variable-at-a-time methodology |
See Agent Guide for details.
Slash commands for common workflows:
| Command | Purpose |
|---|---|
/plan | Create an implementation plan before coding |
/tdd | Start a TDD session with test-first methodology |
/test-coverage | Analyze coverage and generate missing tests |
/update-codemaps | Scan codebase and update architecture docs |
/update-docs | Sync documentation from source-of-truth |
/commit | Analyze uncommitted changes and create small, focused commits grouped by functionality |
/merge-worktree | Merge the current worktree branch into main and clean up |
See Command Guide for details.
Extend Claude's capabilities with specialized knowledge:
| Skill | Purpose |
|---|---|
| scientific-brainstorming | Structured brainstorming for research hypotheses and experimental design |
| software-brainstorming | Structured brainstorming for software architecture and feature planning |
| paper | Transform research papers and code repos into reusable skill references via multi-agent council |
| continuous-learning | Real-time misunderstanding detection and communication optimization |
| autoresearch | Autonomous experiment loop: modify code, train, evaluate, keep/discard, repeat indefinitely |
See Skill Guide for details.
Reusable rule files for consistent code style. Copy them to your project's rules/ directory or ~/.claude/rules/ for global use.
| Rule | Purpose |
|---|---|
| coding-style | Python coding conventions for deep learning projects: GPU setup, type hints, match statements, tqdm progress bars, Ray Tune subprocess pattern, etc. |
| comment-style | Comment guidelines based on "comment why, not what": function docs, design comments, why comments, teacher comments, and anti-patterns to avoid |
Setup:
# Project-level (applies to one project)
cp rules/*.md /path/to/your-project/rules/
# User-level (applies to all projects)
cp rules/*.md ~/.claude/rules/
Template CLAUDE.md files and a status line script. Copy and customize for your own setup.
| File | Purpose |
|---|---|
| user-CLAUDE.md | User-level ~/.claude/CLAUDE.md template: TDD cycle, Tidy First, commit rules, agent auto-dispatch, code quality standards |
| project-CLAUDE.md | Project-level CLAUDE.md template: project structure, code style, testing, security, git workflow |
| statusline.sh | Custom status line showing model, context usage, and rate limit bars |
CLAUDE.md setup:
# User-level (global instructions for all projects)
cp claude-settings/user-CLAUDE.md ~/.claude/CLAUDE.md
# Project-level (edit to fit your project)
cp claude-settings/project-CLAUDE.md /path/to/your-project/CLAUDE.md
Status line setup:
Copy the script:
cp claude-settings/statusline.sh ~/.claude/statusline.sh
chmod +x ~/.claude/statusline.sh
Add to ~/.claude/settings.json:
{
"statusLine": {
"type": "command",
"command": "bash ~/.claude/statusline.sh"
}
}
Requirements: bash, jq
This repository focuses on custom skills for deep learning researchers. We also recommend adding official Anthropic skills (like document processing, MCP builders, etc.) using the Claude Code plugin system:
# Add all Anthropic official skills
claude plugin add anthropic/skills
# Or add specific skill categories
claude plugin add anthropic/skills/document-skills
Available official Anthropic skill categories: