| name | omega-claude-cli |
| description | Use when the user wants to use Claude (Claude Code CLI) for analysis, codebases, or brainstorming from another agent. Uses headless Claude Code CLI scripts (no MCP). Triggers on "use Claude", "analyze with Claude", "ask Claude", "brainstorm with Claude". |
| allowed-tools | Read, Grep, Bash |
Omega Claude CLI (headless)
This skill uses the Claude Code CLI in headless mode so other agents (Codex, Cursor, Copilot, etc.) can run Claude from scripts—no MCP server or MCP configuration required.
Overview
Run Claude via the headless script: scripts/ask-claude.mjs (from the skill root) or .claude/skills/omega-claude-cli/scripts/ask-claude.mjs (from the project root). It invokes claude -p "..." --dangerously-skip-permissions (and optional --model, --json, --sandbox, --timeout-ms) and returns the response. Requires Node.js and the Claude Code CLI to be installed. For one-time setup after copying this folder, run /omega-claude-setup. For install and auth, see references/installation.md and references/auth.md.
The wrapper strips CLAUDECODE from the child process environment, so it also works when invoked from inside an active Claude Code session.
First-time setup
If the user has not set up Claude Code CLI yet, direct them to run /omega-claude-setup. That runs the verification script (node .claude/skills/omega-claude-cli/scripts/verify-setup.mjs) and guides them to install Node, install Claude Code CLI, and complete one-time auth. No MCP config is needed.
How to run Claude (headless)
From the project root (where .claude lives), run:
node .claude/skills/omega-claude-cli/scripts/ask-claude.mjs "USER_PROMPT"
Options (append to the command):
--model MODEL — optional. CLI aliases: opus (5), sonnet (5), haiku (4.5), fable (5), best, opusplan, sonnet[1m], opus[1m], or a full claude-* model ID. Omit to use your account's CLI default. (Anthropic API resolutions; other providers may differ — see model-config.)--json — output JSON from Claude CLI; the script prints the .result text field.--sandbox — run or test code in Claude's sandbox mode.--timeout-ms N — optional timeout for automation.
Examples:
- Analysis:
node .claude/skills/omega-claude-cli/scripts/ask-claude.mjs "Review @src/main.js for bugs"
- With model:
node .claude/skills/omega-claude-cli/scripts/ask-claude.mjs "Summarize this" --model sonnet
- Stdin:
echo "Explain recursion" | node .claude/skills/omega-claude-cli/scripts/ask-claude.mjs
When the user says "ask Claude for feedback" (or similar), build the prompt from their request and any @ file refs, then run the script and return the script output to the user.
When to use this skill
- Analysis — Use Claude's context for code or doc review from another agent.
- Second opinion / feedback — User wants Claude's take on code or docs from Codex, Cursor, Copilot, etc.
- Brainstorming — Build a prompt that includes the challenge and optional methodology and run the script.
For full headless options and examples, see references/headless.md. For copying this skill into another project, see references/copy-and-run.md. Other surfaces: Cursor, Codex CLI, GitHub Copilot CLI, Antigravity IDE, VS Code.
Slash commands
- /analyze — Run the headless script with the user's prompt (and any @ refs). Example:
/analyze summarize @README.md.
- /omega-claude — Use the headless script for analysis or brainstorm as appropriate.
- /omega-claude-setup — Verify Node and Claude Code CLI; guide the user to install and auth. No MCP.
Natural language: "use Claude to explain index.html", "ask Claude for feedback on this function", "brainstorm with Claude using design thinking".
Auth
One-time sign-in and auth issues: references/auth.md.
Failure handling
- If shell execution is unavailable in the host agent (for example
@lydell/node-pty missing binary errors), report that the Claude call is blocked by runtime setup.
- Do not present fabricated Claude output. Return the real error and troubleshooting steps instead.
Copy and run in another project
Copy the entire .claude folder (so you have .claude/skills/omega-claude-cli/ and .claude/commands/) into the target project. Open that project in your agent and run from its root. Run /omega-claude-setup once, then "ask Claude" works. Details: references/copy-and-run.md.
