ds-agent-rules

繁體中文版 README

A portable, composable rules system for AI coding agents — one source of truth for Data Science, Machine Learning, and AI Engineering projects.

Write rules once. Sync to Claude Code · GitHub Copilot · OpenAI Codex · Gemini Code · Cursor · Windsurf — all at once.

The Problem

Without explicit rules, AI agents silently introduce bad habits:

ds-agent-rules solves this with a layered, composable rule system that keeps every AI tool aligned.

How It Works

 ┌────────────────────┐
 │   base/core.md     │  ← always loaded
 │   base/ds-ml.md    │  ← project-type overlay
 │   snippets/rag.md  │  ← domain-specific rules
 │   team/*.md        │  ← team overrides (optional)
 └────────┬───────────┘
          │  sync.sh
          ▼
 ┌────────────────────────────────────┐
 │  CLAUDE.md                        │
 │  AGENTS.md                        │
 │  .github/copilot-instructions.md  │
 │  .gemini/styleguide.md            │
 │  .cursorrules                     │
 │  .windsurfrules                   │
 └────────────────────────────────────┘

Layer model: core (always) → overlay (project type) → snippets (domains) → team (overrides)

Quickstart

1. Install

Choose your preferred method:

# npm (zero-install via npx)
npx ds-agent-rules init

# pip
pip install ds-agent-rules
ds-agent-rules init

# git clone (full control)
git clone https://github.com/Edwarddev0723/ds-agent-rules ~/.ai-rules
cd ~/.ai-rules && chmod +x sync.sh new-project.sh

2. Pick your path

cd /path/to/your/project
npx ds-agent-rules preset llm-project    # npm
ds-agent-rules preset llm-project        # pip

# or interactive
npx ds-agent-rules new-project

cd /path/to/your/project
~/.ai-rules/new-project.sh

cd /path/to/your/project
~/.ai-rules/sync.sh --preset llm-project

cd /path/to/your/project
~/.ai-rules/sync.sh --init          # creates .ai-rules.yaml template
vim .ai-rules.yaml                   # edit to match your project
~/.ai-rules/sync.sh                  # sync (auto-reads config)

3. Useful flags

./sync.sh --list                     # show all overlays, snippets, presets
./sync.sh --dry-run ds-ml rag        # preview without writing files
./sync.sh --diff                     # show unified diff before applying changes
./sync.sh --validate                 # check project structure against rules
./sync.sh --output-dir /other/proj   # write to a different project
./sync.sh --team ./team-rules        # include team-specific rules

4. Make targets

make help                            # show all available targets
make lint                            # run ShellCheck on all scripts
make test                            # run bats test suite
make validate                        # validate current project
make ci                              # lint + test (same as CI)

Project Structure

ds-agent-rules/
├── base/                    # Project-type overlays
│   ├── core.md              # Universal rules (always included)
│   ├── ds-ml.md             # Data Science / ML
│   ├── llm-eng.md           # LLM / GenAI Engineering
│   ├── data-eng.md          # Data Engineering
│   ├── software-eng.md      # Traditional Software Engineering
│   └── research.md          # Research / Academic
│
├── snippets/                # Domain-specific rule modules (mix & match)
│   ├── agentic-ai.md        # AI Agents & tool use
│   ├── audio-speech.md      # ASR / TTS / Audio
│   ├── chinese-nlp.md       # Traditional Chinese NLP
│   ├── ctr-prediction.md    # CTR / Recommendation Systems
│   ├── cv.md                # Computer Vision
│   ├── data-labeling.md     # Annotation & Active Learning
│   ├── distributed-training.md  # Multi-GPU/Node (DeepSpeed, FSDP)
│   ├── edge-inference.md    # Mobile / Edge Deployment
│   ├── evaluation-framework.md  # Systematic Evaluation
│   ├── graph-ml.md          # Graph Neural Networks
│   ├── jax.md               # JAX / Flax
│   ├── llm-finetuning.md    # LLM Fine-Tuning (LoRA, RLHF)
│   ├── mlops.md             # MLOps & Deployment
│   ├── nlp-general.md       # General NLP
│   ├── prompt-engineering.md    # Prompt Design & Versioning
│   ├── pytorch.md           # PyTorch
│   ├── rag.md               # RAG Pipeline
│   ├── responsible-ai.md    # Responsible AI & Safety
│   ├── streaming-ml.md      # Online Learning & Streaming
│   ├── synthetic-data.md    # Synthetic Data & Privacy
│   ├── tabular-ml.md        # Tabular ML
│   ├── time-series.md       # Time Series Forecasting
│   └── vlm.md               # Vision-Language Models
│
├── presets/                  # Named combos for one-command setup (15 presets)
├── templates/                # Directory scaffolds per project type (5 templates)
├── tests/                    # bats test suite
│   └── sync.bats
├── .github/
│   ├── workflows/ci.yml      # CI (ShellCheck + bats on ubuntu & macos)
│   ├── PULL_REQUEST_TEMPLATE.md
│   └── ISSUE_TEMPLATE/       # Issue templates (new snippet, bug report)
├── sync.sh                   # Main sync script
├── new-project.sh            # Interactive project initializer
├── Makefile                  # make lint / test / validate / ci
├── CONTRIBUTING.md           # Contributor guide & snippet format spec
├── CHANGELOG.md              # Release history
└── README.md

Presets

Run ./sync.sh --list to see your local presets.

Configuration

.ai-rules.yaml (per-project)

Drop this in your project root. sync.sh auto-detects it.

profile: ds-ml
snippets:
  - llm-finetuning
  - rag
  - pytorch
  - mlops

# team_dir: ./team-rules     # optional: team-specific rules
# preset: llm-project        # optional: use a preset instead

Team Rules

Append company/team-specific .md rules after all snippets:

mkdir team-rules && vim team-rules/our-standards.md

# Via CLI
./sync.sh --team ./team-rules ds-ml rag

# Or in .ai-rules.yaml
# team_dir: ./team-rules

Extending

Installation & Git Strategy

# Option 1: npm (recommended for JS/TS developers)
npm install -g ds-agent-rules        # global install
npx ds-agent-rules sync ds-ml rag    # or run directly via npx

# Option 2: pip (recommended for Python developers)
pip install ds-agent-rules
ds-agent-rules sync ds-ml rag

# Option 3: Standalone (git clone)
git clone https://github.com/Edwarddev0723/ds-agent-rules ~/.ai-rules

# Option 4: Git submodule in dotfiles
cd ~/.dotfiles && git submodule add https://github.com/Edwarddev0723/ds-agent-rules

Committing generated files?

Recommended Workflow

# 1. Start a new project
mkdir my-project && cd my-project && git init

# 2. Initialize (pick one)
~/.ai-rules/new-project.sh              # interactive
~/.ai-rules/sync.sh --preset llm-project # one-liner
~/.ai-rules/sync.sh --init              # config file

# 3. Work with your AI tools — they auto-read the generated files

# 4. Validate project structure
~/.ai-rules/sync.sh --validate

# 5. Evolve your rules
vim ~/.ai-rules/snippets/rag.md
~/.ai-rules/sync.sh
cd ~/.ai-rules && git add -A && git commit -m "rule: ..."

AI Tool → File Mapping

Contributing

We welcome contributions! See CONTRIBUTING.md for:

• Snippet format specification & quality criteria
• Preset & overlay format
• Commit conventions
• PR checklist

Changelog

See CHANGELOG.md for release history.

Who Uses This

Using ds-agent-rules in your project or team? We'd love to hear about it! Open an issue or PR to add your name here.

License

MIT