eCode Reviewer

Package-style pre-push AI code reviewer for JavaScript and TypeScript diffs.

The best setup is to install the reviewer as a dev dependency and keep only project-specific config/rules in each repo. That way your team does not copy the tool source into every project.

Recommended Project Setup

In a project that should use the reviewer:

npm install --save-dev ecode-reviewer

Add scripts:

{
  "scripts": {
    "review": "ecode-reviewer review",
    "setup-hooks": "ecode-reviewer setup-hooks"
  }
}

Add a project config at reviewer.config.json:

{
  "llm": {
    "provider": "ollama",
    "endpoint": "http://localhost:11434/api/generate",
    "model": "qwen2.5-coder:14b",
    "timeout_ms": 30000,
    "runtime": "ollama"
  },
  "review": {
    "style_guide_paths": ["./style-guide", "./tools/style-guide.md"],
    "token_budget": 6000,
    "block_on": ["ERROR"],
    "base_ref": "origin/main",
    "ignore_patterns": ["*.min.js", "package-lock.json", "*.generated.ts"],
    "rule_files": []
  }
}

LLM Providers

The default provider is local Ollama:

{
  "llm": {
    "provider": "ollama",
    "endpoint": "http://localhost:11434/api/generate",
    "model": "qwen2.5-coder:14b",
    "timeout_ms": 30000
  }
}

OpenAI:

{
  "llm": {
    "provider": "openai",
    "endpoint": "https://api.openai.com/v1/chat/completions",
    "model": "gpt-4.1-mini",
    "api_key_env": "OPENAI_API_KEY",
    "timeout_ms": 50000
  }
}

OpenAI-compatible APIs such as vLLM, Groq, OpenRouter, or LM Studio:

{
  "llm": {
    "provider": "openai-compatible",
    "endpoint": "https://your-provider.example/v1/chat/completions",
    "model": "your-model-name",
    "api_key_env": "ECODE_REVIEWER_API_KEY",
    "timeout_ms": 50000
  }
}

Anthropic:

{
  "llm": {
    "provider": "anthropic",
    "endpoint": "https://api.anthropic.com/v1/messages",
    "model": "claude-3-5-sonnet-latest",
    "api_key_env": "ANTHROPIC_API_KEY",
    "timeout_ms": 50000
  }
}

API keys are read from environment variables only. Do not put secrets directly in reviewer.config.json.

Then each developer runs this once after cloning:

npm run setup-hooks

After that, the reviewer runs automatically on:

git push

For emergency pushes, Git’s normal bypass still works:

git push --no-verify

Local Package Testing

Before publishing, you can test this package from another project using a local file dependency:

npm install --save-dev /Users/sam/Desktop/eCode-rv

Then use the same scripts:

{
  "scripts": {
    "review": "ecode-reviewer review",
    "setup-hooks": "ecode-reviewer setup-hooks"
  }
}

Config Discovery

The CLI looks for config in this order:

1. reviewer.config.json
2. code-reviewer.config.json
3. tools/reviewer.config.json

You can also pass a config explicitly:

ecode-reviewer review --config ./config/reviewer.json

Use quiet mode to hide the loading indicator:

ecode-reviewer review --quiet

style_guide_paths can point to markdown files or folders of markdown files. Style-guide files are loaded fresh on every push.

The reviewer also applies a built-in baseline for Security, SOLID, DRY, TypeScript safety, error handling, and test coverage. Project rules are added on top of that baseline.

Behavior

• Blocks the push when the model returns an ERROR.
• Allows the push when there are warnings only.
• Fails open when the LLM server is down, times out, or returns malformed JSON.
• Reviews only JavaScript and TypeScript diffs.