Micro Agent

To update to the latest version, you can use the built-in update command or npm:

micro-agent update
# OR
npm update -g @builder.io/micro-agent

Run Micro Agent on a specific file and iterate until a provided test script passes.

Default behavior: Micro Agent looks for a test file with the same name as the target file but with .test.ts appended (e.g., ./file.ts looks for ./file.test.ts).

Flags:

• -t, --test <string>: The test script to run (e.g., npm test).
• -f, --test-file <string>: Specify a custom test file path.
• -p, --prompt <string>: Specify a custom prompt file path (defaults to <filename>.prompt.md).

Micro Agent requires Node.js v18 or later. Install the CLI globally using npm:

npm install -g @builder.io/micro-agent

To start Micro Agent in interactive mode, where it will ask you questions about the code it generates and use your feedback to improve it, run the command without arguments:

micro-agent

Micro Agent supports Claude, OpenAI, Ollama, or any OpenAI compatible provider (like Groq). Use the config set command to configure your environment.

For OpenAI/Compatible providers:

• Set API key: OPENAI_KEY
• Set model: MODEL (defaults to gpt-4o)
• Set custom endpoint: OPENAI_API_ENDPOINT

For Claude:

• Set API key: ANTHROPIC_KEY
• Set model: MODEL (use claude as an alias for claude-3-5-sonnet-20241022)

# OpenAI configuration
micro-agent config set OPENAI_KEY=<your token>
micro-agent config set MODEL=gpt-4o
micro-agent config set OPENAI_API_ENDPOINT=https://api.groq.com/openai/v1

# Claude configuration
micro-agent config set ANTHROPIC_KEY=<your token>
micro-agent config set MODEL=claude

Match a design by providing a design image and a local URL to your rendered code. Micro Agent will iterate until the rendered output matches the screenshot.

Requirements:

• Requires an Anthropic API key (recommended for visual matching).
• A screenshot file must exist next to the code being edited (e.g., ./app/about/page.png for ./app/about/page.tsx).

Flags:

• --visual <string>: The local URL of your rendered code (e.g., localhost:3000/about).

micro-agent ./app/about/page.tsx --visual localhost:3000/about

You can manage settings using the config command.

• Interactive UI: Run micro-agent config to open a visual interface for setting keys and models.
• Environment Variables: All config options can be overridden by environment variables (e.g., MODEL=gpt-3.5-turbo micro-agent ...).
• Max Runs: Use the -m flag to change the maximum number of iteration attempts (default is 10).

Full list of available commands and flags:

Commands:

• config: Configure the CLI
• update: Update Micro Agent to the latest version

Flags:

• -h, --help: Show help
• -m, --max-runs <number>: The maximum number of runs to attempt
• -p, --prompt <string>: Prompt to run
• -t, --test <string>: The test script to run
• -f, --test-file <string>: The test file to run
• -v, --visual <string>: Visual matching URL
• --thread <string>: Thread ID to resume
• --version: Show version

The interactiveMode function starts a CLI-based interactive session. It guides the user through:

1. Providing an OPENAI_KEY if one is not already configured.
2. Entering a natural language prompt describing the task (e.g., "A function that...").
3. Selecting a target file to create or edit (it uses an LLM to suggest a file path based on your prompt).
4. Generating a unit test file if one doesn't exist, allowing for iterative feedback.
5. Confirming the test command to run.
6. Executing the agent loop to fulfill the prompt.

It accepts a Partial<RunOptions> object to configure the session.

The getConfig function retrieves the current configuration. It follows a priority order:

1. Values passed via cliConfig (as a RawConfig object).
2. Values stored in the local configuration file (~/.micro-agent).
3. Environment variables (which are merged into the final object).

Returns a ValidConfig object where all values have been parsed according to their specific types (e.g., booleans are converted from strings, models have defaults).

The isInvalidCommand function identifies if a command execution failed because the command itself was not found or the script is missing. It checks for the following substrings in the output:

• command not found:
• command_not_found:
• npm ERR! Missing script: