Skip to content
BENIBIX Learn

Claude Code Action Inputs

AI⌨️ Claude CodeIntegrations

Also known as prompt, claude_args, anthropic_api_key, claude_code_oauth_token

The settings that tell the GitHub Action what Claude should do.

Think of it like

It is like the boxes on a form. You fill them in, then hand it over.

Example

Jack wants a review on every new pull request. He sets prompt to his review skill, and Claude runs it each time with no @claude needed.

Why it matters

With the inputs, the action is not just a helper you call. It can be a job that runs by itself.

Definition

The action inputs are the settings in the step that runs Claude. They sit under the with key. prompt gives Claude a task, so it runs without waiting for @claude. claude_args passes any Claude Code flag, such as --max-turns or --model. anthropic_api_key or claude_code_oauth_token signs Claude in. Others add settings, plugins, a new trigger word or a cloud provider.

Example code

yaml
with:
  prompt: "/review"
  claude_args: "--max-turns 5 --model sonnet"
A set task and a cap on how many turns it takes.
The inputs say what the workflow asks Claude to do.

Common mistake

Putting model or max_turns as their own inputs, as the old beta did. In version 1 they go inside claude_args.

Key takeaways

  1. 1
    prompt. Runs with no mention.
  2. 2
    claude_args. Any CLI flag.
  3. 3
    A sign-in input. Key or token.