ask-claude_skill

Overview

This skill runs the local Claude CLI to get coding assistance from Claude models. It offers non-interactive queries for code generation, review, debugging, and follow-up prompts while inheriting your current working directory. Use it as a quick second opinion or to delegate short coding tasks to Claude from scripts or the terminal.

How this skill works

The skill invokes the installed claude command with the -p flag for non-interactive print mode, passing your prompt and any flags like --model or --output-format. It can continue a recent conversation with -c, request specific models, and return plain text or JSON (including stream-json for incremental output). Because it runs locally, it inherits the shell working directory and executes only if the claude CLI is present in PATH.

When to use it

• Get a second opinion on implementation choices or algorithms.
• Generate boilerplate code, tests, or small utility functions.
• Debug or review code snippets and ask for fixes or explanations.
• Automate developer tasks in scripts or CI steps that can call a CLI.
• Obtain structured output (JSON) for further processing by tools.

Best practices

• Always install and verify the claude CLI is available in your PATH before using the skill.
• Use -p for scripting and automation; be aware it skips interactive prompts and workspace trust dialogs.
• Specify --model when you need a particular Claude variant for tone or capability.
• Prefer --output-format json or stream-json if downstream tools will parse the response.
• Limit sensitive secrets or credentials in prompts since the CLI runs with your local environment context.

Example use cases

• Ask Claude to write a function and return only the code for easy copy-paste.
• Request a code review of a pasted snippet and get concise issues and fixes.
• Run sequential prompts: generate code, then use -c to add error handling or tests.
• Integrate into a script that queries Claude and parses JSON response for automation.
• Use stream-json to display partial responses in a custom UI while generation is ongoing.