Plugins

Plugins are custom code scripts that extend Kaizen's capabilities. Use them for tasks that require deterministic code execution: PDF generation, image processing, data transformation, API calls, and more.

Skills vs Plugins

	Skills	Plugins
Written in	Natural language	Code (Python, Node.js, etc.)
Best for	Agentic workflows, multi-step tasks	Deterministic operations, heavy computation
Self-improving	Yes, via memory	No, code is fixed until edited
Quality checked	Via guardrails	Via code pipeline (test + review)

Skills and plugins work together. A skill can reference a plugin for its code-heavy steps (e.g., a "Weather Dashboard" skill that calls a chart-generation plugin).

Supported Languages

Language	Runtime	File Extension
Python	python3 (or python on Windows)	.py
Node.js	node	.js
TypeScript	npx tsx	.ts
Bash	bash	.sh
Ruby	ruby	.rb

Creating a Plugin

From the UI

1. Go to Plugins (/plugins)

2. Click Create Plugin

3. Choose a language, name, and description

4. Write your code in the Monaco editor

5. Define input parameters

6. Add dependencies if needed

From Chat

Ask the agent:

"Create a plugin that takes a CSV file and generates a bar chart as a PNG."

The agent will write the code, test it through the code pipeline (developer, execute, review), and register it as a tool.

Input / Output Contract

Plugins receive input as JSON and must output JSON to stdout.

Python

import json, sys

input_data = json.loads(sys.argv[1])
name = input_data.get("name", "World")

result = {"message": f"Hello, {name}!"}
print(json.dumps(result))

Node.js / TypeScript

const input = JSON.parse(process.argv[2] || '{}');
const name = input.name || 'World';

const result = { message: `Hello, ${name}!` };
console.log(JSON.stringify(result));

File Output

If your plugin generates files, include a files array in the output. Kaizen will automatically track them as artifacts.

result = {
    "message": "Generated report",
    "files": ["report.pdf", "chart.png"]
}
print(json.dumps(result))

Input Parameters

Define input parameters in the plugin detail page. Each parameter has:

• Name: Parameter key

• Type: string, number, boolean, etc.

• Description: What the parameter does

• Required: Whether the parameter is mandatory

These generate a JSON Schema that the agent uses to validate inputs when calling the plugin.

Dependencies

Add package dependencies in the plugin detail page:

• Python: Installed via pip install

• Node.js / TypeScript: Installed via npm install

• Ruby: Installed via gem install

Dependencies are installed in the plugin's directory, keeping them isolated from other plugins and the main project.

Code Pipeline

When a plugin is created or edited through chat, it goes through a quality pipeline:

1. Developer writes or patches the code

2. Syntax validation checks for errors before execution

3. Execute runs the plugin with test inputs

4. Review inspects the output (especially for visual content like images or HTML)

The pipeline runs up to 3 attempts. On attempt 2+, the developer uses patch mode for surgical edits rather than full rewrites.

Configuration

Setting	Default	Description
Timeout	60 seconds	Maximum execution time before the plugin is killed
Enabled	true	Whether the plugin is available as a tool

Storage

Plugins are stored on disk at workspace/plugins/{plugin-name}/ with the main script file and any dependencies. Plugin metadata (name, description, schema, config) is stored in the database.