/help

name: help
description: Show how to use deliberation here - what each command does, when to reach for it, and real prompts you can paste right now. Read-only.
allowed-tools: Bash
timeout: 15000

Help

Deliberation gives you a second opinion from GPT, Gemini, Grok, and OpenRouter, without leaving your editor. You ask; they answer independently; Claude reads the answers back to you. The experts advise by default - GPT and Gemini can also make changes when you ask them to.

Step 1: One quick look at your setup (ONE Bash call)

bash "${CLAUDE_PLUGIN_ROOT}/scripts/commands/help.sh"

If the host is not Claude Code, tell the user the commands work the same way but the exact syntax can differ on their host, and point them at the per-host guides in the project's `public-docs/hosts/`. Then show the rest of this guide.

Step 2: Show the guide

Render the sections below. Keep it tight - this is a cheat sheet, not a manual.

What each command is for

| Command | Use it when | |---------|-------------| | `/ask-gpt`, `/ask-gemini`, `/ask-grok`, `/ask-openrouter` | You want one model's take on something. | | `/ask-all` | You want several models in parallel and a side-by-side read. | | `/consensus` | The decision matters and you want the models to argue to one verdict. | | `/deliberation:analyze` | You want to see which models are slow or rarely add anything. | | `/deliberation:doctor` | Something looks broken - commands missing, providers failing, empty analyze. | | `/deliberation:setup` | First install, or to repair the install. |

Prompts you can paste right now

# Quick second opinion on what you're doing
/ask-gpt Is the token-expiry check in my auth middleware off-by-one? It uses < not <=.

# Several models at once, then compare
/ask-all Review the current diff for real bugs, security issues, and missing tests.

# Make a hard call with the models cross-checking each other
/consensus Should this be a config migration or a backward-compatible fallback? Tradeoffs: <paste them>

# Hand a stubborn bug to the debugger
/ask-gemini Debug why this test fails intermittently, ranked by likelihood: <paste the error>

# Weigh an architecture choice
/ask-gpt Keep provider routing in one module or split it per provider? Weigh maintainability vs testability.

# Research without leaving your editor
/ask-grok Best way to validate Node CLI auth state without mutating the user's config?

Tips

• Add an expert by name to sharpen the answer: "as the security-analyst, ..." or

"have the code-reviewer ...". The seven experts: architect, plan-reviewer, scope-analyst, code-reviewer, security-analyst, researcher, debugger.

• Point at files in your prompt - the local experts (GPT, Gemini) can read them.
• Stuck? Run `/deliberation:doctor`. Curious which models earn their slot? `/deliberation:analyze`.

Rules

• Read-only. This command only explains - it changes nothing.
• Keep it short and concrete. Lead with the paste-ready prompts; skip the theory.
• Be honest about the host: if it is not Claude Code, say the syntax may differ.