Gemini 助手

其他社区

作者仓库星标 0
作者仓库 skills-registry

解读按原文结构重写，命令、链接、术语均保留；右侧可核对作者原始 SKILL.md

原文结构

ALWAYS USE GEMINI 3：Default to Gemini 3 models for all tasks. These are the current generation with superior reasoning. Model · When to Use gemini-3-flash-preview · DEFAULT - Use for most tasks. Fast + smart.
Legacy Models (only if specifically requested)：Model · Notes gemini-2.5-pro · Previous flagship gemini-2.5-flash · High throughput
Gemini 3 Features：thinkinglevel - Control reasoning depth (low, medium, high) mediaresolution - Vision processing quality (low, medium, high, ultrahigh) Thought Signatures - Maintains chain of reasoning across calls

适用与边界

当前原文没有单列适用场景。
当前原文没有单列不适用场景或限制。

原文中的明确线索

要点：「Default to Gemini 3 models for all tasks.」、「DEFAULT」、「thinkinglevel」、「mediaresolution」、「Thought Signatures」、「single prompt」、「Non-interactive one-shot」、「Interactive after prompt」
文件与命令：gemini-3-flash-preview、gemini-3-pro-preview、gemini-2.5-pro、gemini-2.5-flash、thinkinglevel、low、medium、high

流狐整理：以上内容来自当前 SKILL.md 的章节与原词；未补写作者没有声明的工具、兼容性或能力。

Gemini CLI Skill Guide (v0.23.0+)

ALWAYS USE GEMINI 3

Default to Gemini 3 models for all tasks. These are the current generation with superior reasoning.

Model	When to Use
`gemini-3-flash-preview`	DEFAULT - Use for most tasks. Fast + smart.
`gemini-3-pro-preview`	Complex reasoning, architecture, 1M context tasks

Legacy Models (only if specifically requested)

Model	Notes
`gemini-2.5-pro`	Previous flagship
`gemini-2.5-flash`	High throughput

Gemini 3 Features

thinking_level - Control reasoning depth (low, medium, high)
media_resolution - Vision processing quality (low, medium, high, ultra_high)
Thought Signatures - Maintains chain of reasoning across calls

Running a Task

Ask the user which model to run (default: gemini-3-flash-preview) AND which approval mode (default, auto_edit, yolo) in a single prompt.
Select approval mode based on task:
- default - Prompt for approval on each action
- auto_edit - Auto-approve file edits only
- yolo - Auto-approve all actions (use with caution)
Assemble command with options:
- -m, --model <MODEL> (e.g., -m gemini-2.5-pro)
- --approval-mode <default|auto_edit|yolo>
- -y, --yolo (shortcut for --approval-mode yolo)
- -s, --sandbox (run in sandbox mode)
- --include-directories <PATH> (add additional directories)
- -r, --resume <latest|INDEX> (resume previous session)
- -o, --output-format <text|json|stream-json> (output format)
Non-interactive one-shot: gemini "your prompt here" (positional argument)
Interactive after prompt: gemini -i "initial prompt"
Resume syntax: gemini -r latest or gemini -r 5
After completion: Inform user they can resume with gemini -r latest.

Note: The -p/--prompt flag is deprecated. Use positional arguments instead.

Quick Reference

Use case	Approval mode	Command
Read-only review	`default`	`gemini -m gemini-3-flash-preview "prompt"`
Complex reasoning	`default`	`gemini -m gemini-3-pro-preview "prompt"`
Auto file edits	`auto_edit`	`gemini -m gemini-3-flash-preview --approval-mode auto_edit "prompt"`
Full auto (YOLO)	`yolo`	`gemini -m gemini-3-flash-preview -y "prompt"`
Sandboxed	`default`	`gemini -m gemini-3-flash-preview -s "prompt"`
With extra dirs	Match task	`gemini -m gemini-3-flash-preview --include-directories /path/to/other "prompt"`
Resume latest	Inherited	`gemini -r latest`
Resume specific	Inherited	`gemini -r 5`
Interactive mode	Match task	`gemini -i "start with this prompt"`
JSON output	Match task	`gemini -m gemini-3-flash-preview -o json "prompt"`
Stream JSON	Match task	`gemini -m gemini-3-flash-preview -o stream-json "prompt"`

Session Management

Command	Description
`gemini --list-sessions`	List available sessions
`gemini -r latest`	Resume most recent session
`gemini -r <INDEX>`	Resume session by index
`gemini --delete-session <INDEX>`	Delete a session

MCP & Extensions

Command	Description
`gemini mcp`	Manage MCP servers
`gemini -l` / `--list-extensions`	List available extensions
`gemini -e ext1 ext2`	Use specific extensions only
`--allowed-mcp-server-names`	Specify allowed MCP servers
`--allowed-tools`	Tools that run without confirmation

Built-in Tools

Gemini CLI has these tools enabled by default:

File operations: Read/write files in working directory
Shell commands: Execute terminal commands
Web fetch: Retrieve web content
Google Search: Grounded search for current information

Project Context

Create GEMINI.md in project root for project-specific instructions (similar to CLAUDE.md).

Following Up

After every gemini command, confirm next steps or offer to resume.
Resume preserves original model and approval mode.
Restate configuration when proposing follow-up actions.

Error Handling

Stop and report failures on non-zero exit; request direction before retry.
Ask permission before using -y/--yolo mode.
Summarize warnings/partial results and ask how to proceed.
Check authentication if getting 401/403 errors.
For quota issues, suggest switching auth methods or waiting.

Key Differences from Codex

Feature	Gemini CLI	Codex CLI
One-shot syntax	`gemini "prompt"`	`codex exec "prompt"`
Auto mode	`-y` or `--approval-mode yolo`	`--full-auto`
Sandbox	`-s`	`--sandbox <mode>`
Resume	`-r latest`	`resume --last`
Interactive	`-i "prompt"`	Default interactive
Output format	`-o json/stream-json`	N/A

Source: brunoqgalvao/navi — distributed by TomeVault.

流狐档案作者与许可取自来源；运行、权限和网络为流狐检测或估算

流狐分类: 通用
作者声明 Agent: 未找到明确声明；不据此推断已兼容或已测试
静态检查: 88 / 100 · 启发式扫描，不代表运行安全
作者 / 版本 / 许可: @tomevault-io · 未声明 license
流狐 Token 估算: 低消耗
流狐接入估算: 需简单配置
是否需要外部 API Key: 未发现要求
检测到的系统要求: 未声明
底层运行要求: 未声明
检测到的文件与系统行为: 只读
允许写入 / 修改
Shell 执行
检测到的网络行为: 允许外网请求
安装命令数: 无（仅作为资料）

档案由构建时根据 SKILL.md 与安装命令自动衍生，可能与作者实际意图存在差异。

需要注意： 未限定 allowed-tools，默认拥有全部工具权限。

输出预览 gemini-cli.preview

作者没有在当前 SKILL.md 中定义固定输出样例。

SKILL.md SKILL.md

# Gemini CLI Skill Guide (v0.23.0+)

## ALWAYS USE GEMINI 3

**Default to Gemini 3 models for all tasks.** These are the current generation with superior reasoning.

| Model | When to Use |
|-------|-------------|
| `gemini-3-flash-preview` | **DEFAULT** - Use for most tasks. Fast + smart. |
| `gemini-3-pro-preview` | Complex reasoning, architecture, 1M context tasks |

### Legacy Models (only if specifically requested)
| Model | Notes |
|-------|-------|
| `gemini-2.5-pro` | Previous flagship |
| `gemini-2.5-flash` | High throughput |

### Gemini 3 Features
- **`thinking_level`** - Control reasoning depth (`low`, `medium`, `high`)
- **`media_resolution`** - Vision processing quality (`low`, `medium`, `high`, `ultra_high`)
- **Thought Signatures** - Maintains chain of reasoning across calls

## Running a Task

1. Ask the user which model to run (default: `gemini-3-flash-preview`) AND which approval mode (`default`, `auto_edit`, `yolo`) in a **single prompt**.
2. Select approval mode based on task:
   - `default` - Prompt for approval on each action
   - `auto_edit` - Auto-approve file edits only
   - `yolo` - Auto-approve all actions (use with caution)
3. Assemble command with options:
   - `-m, --model <MODEL>` (e.g., `-m gemini-2.5-pro`)
   - `--approval-mode <default|auto_edit|yolo>`
   - `-y, --yolo` (shortcut for `--approval-mode yolo`)
   - `-s, --sandbox` (run in sandbox mode)
   - `--include-directories <PATH>` (add additional directories)
   - `-r, --resume <latest|INDEX>` (resume previous session)
   - `-o, --output-format <text|json|stream-json>` (output format)
4. **Non-interactive one-shot**: `gemini "your prompt here"` (positional argument)
5. **Interactive after prompt**: `gemini -i "initial prompt"`

…

证据边界与执行链路

作者原文负责流程事实；流狐只索引当前章节、要点、文件与命令。

章节    -> ALWAYS USE GEMINI 3 → Legacy Models (only if specifically requested) → Gemini 3 Features → Running a Task → Quick Reference → Session Management
要点    -> Default to Gemini 3 models for all tasks. · DEFAULT · thinkinglevel · mediaresolution · Thought Signatures · single prompt · Non-interactive one-shot · Interactive after prompt
文件/命令 -> gemini-3-flash-preview · gemini-3-pro-preview · gemini-2.5-pro · gemini-2.5-flash · thinkinglevel · low · medium · high
内容 SHA-256 -> 96f2c7dd67a8