Installs into .claude/skills of the current project.
Are you the author of Ask Claude?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/diegosouzapw-ask-claude)
---
name: ask-claude
description: Asks Claude CLI for coding assistance. Use for getting a second opinion, code generation, debugging, or delegating coding tasks.
allowed-tools: Bash(claude:*)
---
# Ask Claude
Executes the local `claude` CLI to get coding assistance.
**Note:** This skill requires the `claude` CLI to be installed and available in your system's PATH.
## Quick start
Run a single query with `-p` (print mode):
```bash
claude -p "Your question or task here"
```
## Common options
| Option | Description |
|--------|-------------|
| `-p` | Non-interactive print mode (required for scripting) |
| `--model MODEL` | Specify model (e.g., `sonnet`, `opus`, `haiku`) |
| `--output-format FORMAT` | Output format: `text`, `json`, `stream-json` |
| `-c, --continue` | Continue the most recent conversation |
> For all available options, run `claude --help`
## Examples
**Ask a coding question:**
```bash
claude -p "How do I implement a binary search in Python?"
```
**Use a specific model:**
```bash
claude -p --model opus "Review this code for potential issues"
```
**Get JSON output:**
```bash
claude -p --output-format json "List the main functions in this file"
```
**Continue a previous conversation:**
```bash
claude -p -c "Now add error handling to that function"
```
## Notes
- The `-p` flag runs Claude non-interactively and outputs result to stdout
- If `--model` is not specified, the default model is used
- `stream-json` outputs responses incrementally as they are generated
- **Warning**: The `-p` flag skips the workspace trust dialog. Only use in directories you trust.
- The command inherits the current working directory