gemini

name	gemini
description	Local Gemini for large-context analysis, multimodal reasoning, architecture review. Runs on the Antigravity CLI (`agy`, Gemini 3.5 Flash) by default, with the legacy `gemini` CLI as fallback. Use when user says /gemini, asks to consult Gemini, or when Gemini's strengths (large context, grounded search) complement Claude's analysis.
user-invocable	true
argument-hint	<prompt or review request>
compatibility	Requires the Antigravity CLI (`agy`) or the legacy `gemini` CLI installed and authenticated.

Gemini

Use the local Gemini backend through the wrapper scripts. The skill auto-selects its backend (see below); the public flags and output shapes are identical regardless of which backend runs, so consumer skills (/coordinate, /pr, /handover, /verify, debate mode) need no changes.

Backends

The wrappers resolve a backend via the GEMINI_BACKEND env var (default auto):

GEMINI_BACKEND	Backend	Model	Notes
auto (default)	agy if installed, else gemini	Gemini 3.5 Flash (agy)	Prefer the modern CLI
agy	Antigravity CLI (agy)	Gemini 3.5 Flash (pinned default, auto-selected)	Required after 18/6/2026 when the legacy CLI stops serving the Google One AI Pro plan
gemini	Legacy Gemini CLI (gemini)	auto → 3.1 Pro, 2.5 fallback	Needed for --list (session listing) and very large prompts

Key differences on the agy backend (handled transparently by the wrappers):

• No per-call model flag — --fast/--deep/--model all resolve to the pinned Gemini 3.5 Flash default (output shape unchanged).
• JSON output comes from the prompt instruction (--structured still works, prints clean JSON).
• --plan is implicit (single-shot --print is non-interactive); --yolo maps to --dangerously-skip-permissions.
• --resume latest → agy -c; --resume <id> → agy --conversation <id>. Numeric session indices have no agy mapping.
• --list (session listing) is gemini-only and transparently falls back to the legacy CLI.

Context

• Working directory: !pwd
• Current branch: !git branch --show-current 2>/dev/null || true
• Git status: !git status -sb 2>/dev/null | head -20 || true

Workflow

1. Decide whether the request is a consultation (ask) or a code review.
2. For consultations, decide the session strategy (see below).
3. For reviews, use the review wrapper (see below).

Consultations

Start a fresh consultation (default — most common case):

bash "${CLAUDE_SKILL_DIR}/scripts/gemini-ask.sh" "<prompt>"

Resume the latest Gemini session when the user is clearly continuing a previous Gemini conversation:

bash "${CLAUDE_SKILL_DIR}/scripts/gemini-ask.sh" --resume latest "<follow-up prompt>"

Resume a specific session by index:

bash "${CLAUDE_SKILL_DIR}/scripts/gemini-ask.sh" --resume 2 "<follow-up prompt>"

List available sessions for the current project:

bash "${CLAUDE_SKILL_DIR}/scripts/gemini-ask.sh" --list

Approval Modes

• --plan: Read-only mode. Best for analysis, reviews, and when Gemini should not touch files.
• --yolo: Auto-approve all tool calls. Only for trusted automated tasks.
• --approval <mode>: Explicit mode — one of default, auto_edit, yolo, plan.

Default is default for consultations. Reviews default to plan.

Code Reviews

Use the dedicated review wrapper. It defaults to plan mode for safety.

bash "${CLAUDE_SKILL_DIR}/scripts/gemini-review.sh" --uncommitted
bash "${CLAUDE_SKILL_DIR}/scripts/gemini-review.sh" --base main
bash "${CLAUDE_SKILL_DIR}/scripts/gemini-review.sh" --commit <sha>

Add custom instructions with --prompt:

bash "${CLAUDE_SKILL_DIR}/scripts/gemini-review.sh" --uncommitted --prompt "Focus on security"

Depth Control

Choose the depth level based on task complexity. On the agy backend these flags resolve to the pinned Gemini 3.5 Flash default (no per-call model override); the depth distinction below applies to the legacy gemini backend:

• --fast: Uses gemini-2.5-flash. Best for quick lookups, simple questions, high-throughput tasks. Fast and lightweight.
• (default): Uses auto (Gemini 3.5 Flash on agy, 3.1 Pro on gemini). Good for most tasks.
• --deep: Explicitly uses pro (Gemini 3.1 Pro on gemini). Best for complex architecture analysis, deep reasoning, security audits.

# Quick question
bash "${CLAUDE_SKILL_DIR}/scripts/gemini-ask.sh" --fast "What's the default BLE MTU size on iOS?"

# Deep analysis
bash "${CLAUDE_SKILL_DIR}/scripts/gemini-ask.sh" --deep "Analyze this entire module for architectural issues: $(cat src/ble/)"

# Fast review
bash "${CLAUDE_SKILL_DIR}/scripts/gemini-review.sh" --fast --uncommitted

Structured Output

Use --structured to get JSON output for machine-readable results. This is essential for cross-model chaining — when output from Gemini will be compared with or fed to Codex.

bash "${CLAUDE_SKILL_DIR}/scripts/gemini-ask.sh" --structured "Review this function for bugs: $(cat src/utils.ts)"

Output schema: { findings[], summary, model } — each finding has id, severity, category, file, line, title, detail, recommendation, confidence.

Session Strategy

• Start a fresh session for new questions, isolated tasks, or one-shot consultations.
• Use --resume latest when the user says to continue, iterate on, or follow up the most recent Gemini conversation.
• Use --resume <index> when the user refers to a specific past session or when multiple ongoing threads make latest ambiguous.
• If the continuation target is ambiguous, ask one short clarifying question.

Review Defaults

• No target specified → --uncommitted.
• User refers to a specific commit → --commit.
• User refers to branch or PR changes against a base → --base.
• If the review target is ambiguous and the wrong target would be misleading, ask one short clarifying question.

Multi-Agent Routing Guide

Use this guide when deciding whether to delegate a task to Gemini, Codex, or both:

Task Type	Best Model	Depth	Why
Platform-specific (iOS/Android/BLE)	Gemini	--deep	Excellent at platform quirks, real-world edge cases, battery/lifecycle
Large codebase analysis	Gemini	default	Largest context window, can process entire modules
Code review (general)	Both in parallel	default	Different blind spots — merge findings
Security audit	Both in parallel	--deep	Gemini finds integration risks, Codex finds logic flaws
Architecture design	Codex first, Gemini validates	--deep	Gemini adds real-world platform considerations
Logic/algorithm bugs	Codex first	--deep	Better at state machine reasoning
Quick syntax/API question	Gemini	--fast	Flash model is very fast for simple lookups
Multimodal analysis (images, video)	Gemini	default	Native multimodal capabilities

Multi-Agent Collaboration

When Claude delegates work to Gemini as part of a broader multi-model workflow:

One-Shot Consultation

For a focused question where Claude needs Gemini's perspective once:

bash "${CLAUDE_SKILL_DIR}/scripts/gemini-ask.sh" --plan "<specific question with full context>"

Use --plan to ensure Gemini only analyzes and does not modify files. Include all necessary context in the prompt — Gemini does not share Claude's conversation history.

Cross-Model Session Tracking

For tasks involving both Codex and Gemini, use the shared tracker to link sessions:

# Create a shared thread
bash "${CLAUDE_SKILL_DIR}/scripts/cross-model-tracker.sh" new "auth-redesign"

# Link Gemini session
bash "${CLAUDE_SKILL_DIR}/scripts/cross-model-tracker.sh" link "auth-redesign" gemini "latest"

# Log a turn summary after each model interaction
bash "${CLAUDE_SKILL_DIR}/scripts/cross-model-tracker.sh" log "auth-redesign" gemini "Validated JWT rotation approach, flagged iOS background refresh limits"

# Export context for injection into the next model's prompt
context=$(bash "${CLAUDE_SKILL_DIR}/scripts/cross-model-tracker.sh" export "auth-redesign")

Debate Mode

For thorough analysis with automatic cross-model critique:

GEMINI_SKILL_DIR="${CLAUDE_SKILL_DIR}" \
bash "${CLAUDE_SKILL_DIR}/scripts/debate.sh" \
  --topic "Should we use WebSockets or SSE for real-time BLE data streaming?" \
  --first gemini \
  --rounds 1 \
  --deep \
  --output-dir /tmp/debate-ble-streaming

The debate script automates: Model A responds → Model B critiques → Model A addresses critique. Multiple rounds are supported. After the debate, read all round files and synthesize: consensus points, divergences, unique insights, and corrected errors.

Providing Context to Gemini

Gemini cannot see Claude's conversation. When delegating, always embed the relevant context directly in the prompt:

• For code analysis: include the file contents or a diff inline.
• For architecture questions: summarize the current design and constraints.
• For follow-ups on a previous Gemini session: use --resume so Gemini has its own prior context.

Interpreting Results

After receiving Gemini's response:

1. Summarize the key findings and attribute them to Gemini.
2. Preserve concrete details: file paths, line references, specific recommendations.
3. If Gemini's analysis conflicts with Claude's own, present both perspectives and let the user decide.
4. If Gemini's output will be forwarded to another model (e.g., Codex), extract the actionable parts cleanly.

Environment Variables

• GEMINI_BACKEND: Backend selection — auto (default, prefer agy), agy (force Antigravity CLI / Gemini 3.5 Flash), or gemini (force legacy CLI). Use gemini when you specifically need 3.1 Pro depth or session listing.
• GEMINI_SKILL_MODEL: Override the default model on the gemini backend (default: auto → Gemini 3.1 Pro). Use CLI aliases: pro, flash, flash-lite. Ignored on the agy backend (3.5 Flash is pinned).
• GEMINI_SKILL_APPROVAL: Override the default approval policy.

Worker Mode (for /coordinate and multi-agent orchestration)

When invoked as a worker by a coordinator skill, use --worker --scratchpad <dir>:

bash "/Users/administrator/.claude/skills/gemini/scripts/gemini-ask.sh" --worker --scratchpad /tmp/robottino-scratchpad/session-123 --deep --plan "Analyze the auth module architecture"

Worker mode behavior:

• Output is written to {scratchpad}/workers/gemini.md (with frontmatter metadata) and gemini.json (raw)
• --structured is forced on automatically (JSON findings schema)
• No interactive output — all goes to scratchpad files
• The coordinator reads the scratchpad to synthesize findings across workers

Notes

• The wrappers live inside this skill, so the skill remains portable and independent from gemini_mcp.
• Optional environment variables and detailed wrapper documentation are in references/gemini-cli.md.
• For advanced Gemini CLI features, run gemini <command> directly: gemini extensions list, gemini skills list, gemini hooks list.