/ask-gemini

name: ask-gemini
description: Get Gemini second opinion on a question or current work. Single-shot, advisory, no contamination. Model pinned per call.
allowed-tools: mcp__deliberation-gemini__gemini, Read, Bash
timeout: 660000

Ask Gemini

Single-shot delegation to Gemini (Gemini 3 via the Antigravity CLI, `agy`) via MCP for an independent second opinion. Fresh thread, no shared context with prior calls. Advisory mode by default (read-only sandbox). The bridge passes the model to agy's `--model`, so the call wins over `~/.gemini/settings.json`; omit it to take `providers.gemini.model` from `config.json`. If agy rejects the shipped `auto-gemini-3` alias (`invalid model selection` - not in `agy models`), the bridge drops that pin with a stderr warning and the call runs on agy's own default (the result carries `pinDropped: true`); any other rejected id fails as `errorKind: "model-not-allowed"` with agy's catalog in the message.

Input

User question or topic: $ARGUMENTS

Workflow

1. **Identify expert** - match `$ARGUMENTS` against trigger patterns in `~/.claude/rules/deliberation/triggers.md`:

• Architecture / design / tradeoffs → Architect
• Plan validation → Plan Reviewer
• Requirements / scope → Scope Analyst
• Code review / find bugs → Code Reviewer
• Security / vulnerabilities → Security Analyst
• Default if unclear → Architect

2. **Read expert prompt** via this resolution sequence: 1. Glob `~/.claude/plugins/cache/*/deliberation/*/prompts/[expert].md`. Pick the match with the highest semver version segment (the segment immediately after `deliberation/`, parsed as semver - not lexical string compare). 2. If no match, look up the inlined fallback under the heading `## Inlined fallback - [Expert]` in this command file (see end of this file). 3. If neither found, abort with: `Error: deliberation plugin cache missing for expert "[Expert]". Run /plugin install deliberation or /reload-plugins.`

3. **Build 7-section delegation prompt** per `~/.claude/rules/deliberation/delegation-format.md`. Include:

• Verbatim user question from `$ARGUMENTS`
• Relevant code snippets / file paths from current conversation context
• Any specific constraints user has mentioned this session
• Time-sensitive facts: when the question turns on latest or current versions, pricing, a roadmap, or whether a model or tool exists, verify it with your own tools first and add the facts with as-of date and source (the delegate gets today's date from the server but cannot look anything up)

4. **Set cwd** - use `process.cwd()` as the MCP `cwd`; agy print mode needs no folder-trust pre-check.

5. **Call Gemini** - single-shot, advisory, model pinned:

   mcp__deliberation-gemini__gemini({
     prompt: "[7-section delegation prompt]",
     "developer-instructions": "[contents of expert prompt file]",
     sandbox: "read-only",
     cwd: "[current working directory]"
   })

6. **Synthesize response** - never paste raw output. Extract:

• Bottom-line recommendation
• Key reasoning points
• Where Gemini diverges from your prior analysis (if applicable)
• Your assessment of whether Gemini is correct

Rules

• **Single-shot only** - never reuse a `threadId` from a prior `/ask-gemini` call. Each invocation is independent.
• **Don't hardcode a model here** - concrete agy ids differ per install, so a pin in this command would hard-fail for anyone without that exact model. Set `providers.gemini.model` in `config.json` instead (run `agy models` for the local catalog). Resolution: per-call `model` > `providers.gemini.model` > `GEMINI_DEFAULT_MODEL` > `auto-gemini-3`. Ids fuse family and effort (`gemini-3.6-flash-high`); a bare family is rejected with `requires --effort`. Prefer a concrete pin over the `auto-gemini-3` alias where you can - agy's catalog includes Claude and GPT-OSS entries, so routing can seat a non-Gemini model in the Gemini slot.
• **Advisory by default** - use `sandbox: "read-only"` unless user explicitly asks for implementation.
• **No contamination** - do not include prior GPT opinions in the Gemini prompt. Each expert reasons independently.
• **Print status line** immediately before the MCP dispatch: `Gemini working (typical 30-60s)...`
• **Concurrent prep, single dispatch** - prep here is a single expert-prompt `Glob` followed by one dispatch (a fixed status line, no per-delegate config reads). Keep it that way; do not pad the preamble with extra sequential round-trips. See `rules/deliberation/orchestration.md` Step 5.5.

• **Final judgment is the orchestrator's** - the external model only advises. Claude reads its output, applies its own judgment, and is accountable for the synthesized answer shown to you. The model's raw verdict is not the final word.

<!-- GENERATED by scripts/sync-fallbacks.js from prompts/*.md - do not edit below. Required fallback if the plugin cache is missing. -->

Inlined fallback - Architect

> Adapted from [oh-my-opencode](https://github.com/code-yeongyu/oh-my-opencode) by [@code-yeongyu](https://github.com/code-yeongyu)

You are a software architect specializing in system design, technical strategy, and complex decision-making.

Context

You operate as an on-demand specialist within an AI-assisted development environment. You are invoked when a decision needs deep reasoning about architecture, tradeoffs, or system design. Each consultation is standalone: treat every request as complete and self-contained. Your available tools vary by where you run: some environments give you filesystem, repo, or shell access; others give you only the context in the request. Adapt to what you actually have - use tools when present, and when they are absent reason only from what was given. Never fabricate file paths, signatures, or repo details you have not actually seen.

What You Do

• Analyze system architecture and design patterns
• Evaluate tradeoffs between competing approaches
• Design scalable, maintainabl