| name | call-claude |
| description | Delegate a prompt to Claude via `claude -p`. Use when the user wants to offload a task to Claude from another agent (Codex, Cursor, another Claude session), get a second opinion, or run a parallel implementation. Default model is opus. |
| license | MIT |
| metadata | {"author":"stablyai","version":"0.1.0"} |
call-claude
Delegate an arbitrary prompt to Claude and return the output.
Arguments
Required: a prompt.
Optional flags (parsed from the user's message):
| Flag | Values | Default |
|---|
--model | opus, sonnet, haiku, or a full model ID | opus |
--effort | low, medium, high, xhigh, max | (omitted) |
Process
- Parse the user's message — extract the prompt and any
--model / --effort overrides.
- Invoke the wrapper script that ships alongside this SKILL.md:
"$(dirname "$0")/call-claude.sh" [--model <alias>] [--effort <level>] "<prompt>"
For multi-line prompts, pipe via stdin:
echo "<prompt>" | "$(dirname "$0")/call-claude.sh" --model opus
- Return Claude's full output to the user. Don't summarize unless asked.
claude -p is non-interactive and prints the final answer to stdout, so the wrapper output can be captured directly.
Prerequisite
Claude Code must be installed — verify with claude --version.
Examples
./call-claude.sh "explain what this repo does"
./call-claude.sh --model sonnet "summarize the README"
./call-claude.sh --model opus --effort high "design a caching layer"
