Detalle del Skill

gemini-cli

Strong local Gemini CLI usage guidance but assumes it is already installed.

CoincidenciaPosibleRevisado para gemini cli
Fuentex-zero-l/agent-skillsFuente externa
Instalaciones reportadas220Solo señal de popularidad

Revisar antes de usar

La revisión automática comprueba relevancia, no seguridad ni respaldo. Lee las instrucciones de la fuente antes de usar este Skill.

Vista previa guardada

SKILL.md

Este extracto es una copia guardada durante la revisión. La fuente externa contiene la versión completa y actual.

---
name: gemini-cli
description: Ask Gemini via the local `gemini` CLI (no MCP). Use when the user says "ask gemini" / "use gemini", wants a second opinion, needs large-context `@path` analysis, sandbox runs, or structured change-mode edits.
---

# Gemini CLI Skill

Use Google’s `gemini` CLI as an external “second brain” when you need a very large context window, a second opinion, or safe sandbox execution. This skill does **not** use MCP; it runs the local CLI directly.

## When To Use

Use this skill when the user asks to:

- “Ask Gemini” / “Use Gemini”
- Analyze or summarize large files or directories (especially with `@path` references)
- Brainstorm options with a specific framework (SCAMPER, design thinking, etc.)
- Generate structured edit suggestions (“change mode”) that can be applied deterministically

## Prerequisites

- `gemini` CLI is installed and configured (e.g. `gemini --version` works)
- You are allowed to run local shell commands in this environment
- For sandbox execution, Gemini CLI supports `-s/--sandbox` (see `gemini --help`)

## Core Commands

### One-shot query (plain)

```bash
gemini "Explain what this code does: @src/main.ts"
```

### Choose a model

```bash
gemini -m gemini-3-pro-preview "..."
gemini -m gemini-3-flash-preview "..."
```

### Sandbox mode

```bash
gemini -s "Safely test @scripts/example.py and explain the results."
```

## Helper Script (Recommended)

This skill ships with a dependency-free wrapper that mirrors the core behavior of `gemini-mcp-tool` without MCP:

`skills/gemini-cli/scripts/gemini-tool.mjs`

Important: Gemini CLI can only read files under its current “project directory” (typically the process working directory). Run the wrapper from the repo you want Gemini to read, or pass `--cwd`.

```bash
node skills/gemini-cli/scripts/gemini-tool.mjs help
```

### Ask Gemini (wrapper)

```bash
node skills/gemini-cli/scripts/gemini-tool.mjs ask --prompt "Summarize @. in 5 bullets."
node skills/gemini-cli/scripts/gemini-tool.mjs ask --cwd /path/to/project --prompt "Summarize @. in 5 bullets."
node skills/gemini-cli/scripts/gemini-tool.mjs ask --prompt "..." --model gemini-3-pro-preview
node skills/gemini-cli/scripts/gemini-tool.mjs ask --prompt "..." --sandbox
```

### Brainstorm (wrapper)

```bash
node skills/gemini-cli/scripts/gemini-tool.mjs brainstorm --prompt "How should we design onboarding?" --methodology scamper --ideaCount 12
```

### Change Mode (structured edits + chunk caching)

Use change mode when you want Gemini to output **deterministic** OLD/NEW blocks that can be applied by exact string matching.

```bash
node skills/gemini-cli/scripts/gemini-tool.mjs ask --changeMode --prompt "In @src/foo.ts, rename Foo to Bar and update call sites."
```

If the response is chunked, the output includes a `cacheKey`. Fetch the next chunk:

```bash
node skills/gemini-cli/scripts/gemini-tool.mjs fetch-chunk --cacheKey <key> --chunkIndex 2
```

## Safety Rules

- Never execute destructive commands from Gemini output without explicit user approval.
- In change mode, the `OLD` block must match the file content **exactly**; otherwise, do not apply the edit.
- Prefer read-only analysis unless the user explicitly requests changes.
Leer la fuente completa en GitHub (abre una página externa)
Contexto

Trabajo relacionado