Gemini cli

Skill yu-iskw/coding-agent-skills/skills/gemini-cli

A curated collection of executable Agent Skills and SOPs for AI coding agents like Cursor, Claude Code, and Gemini CLI.

Install

npx -y skills add yu-iskw/coding-agent-skills --skill gemini-cli

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Executes tasks using the Gemini CLI (`gemini`). Automatically determines the safest approval mode (plan, auto_edit, or yolo) based on the task type.

SKILL.md

5.0 KB, as published. Nobody here has run it

executing-gemini

Purpose

Use this skill to perform coding, research, or automation tasks using the gemini CLI. It maps requests to Gemini's native approval modes to balance automation with security.

Permission Tiers

Tier	Gemini Mode	Capability	Approval Required	Typical Tasks
0	`plan`	Read-only access, analysis, research. (experimental)	No	Code review, audits, web research.
1	`auto_edit`	Auto-approves file edits; prompts for shell.	Yes	Refactoring, documentation, lint fixes.
2	`yolo`	Auto-approves ALL tools (edits + shell).	Yes (High Risk)	CI/CD, complex builds, automated testing.

Implementation Workflow

1. Quota Verification (Optional but Recommended)

Before initiating tasks, verify the remaining request quota to avoid session suspension.

Tools: Check Google AI Studio or Google Cloud project dashboard for current usage.
Constraints:
- Free/Unpaid: 250 requests/day, 10 requests/minute.
- Paid/Enterprise: 1500-2000 requests/day, 120 requests/minute.

2. Analyze & Classify

Analyze the user's intent to determine the required permission tier.

Tier 0 (Plan): Does the task only involve reading code or searching for information?
Tier 1 (Auto-Edit): Does the task involve modifying files but no command execution?
Tier 2 (YOLO): Does the task require running tests, build scripts, or managing dependencies autonomously?

2. Approval Protocol

If the task maps to Tier 1 or Tier 2, you MUST obtain user approval before executing the gemini command.

Use the AskQuestion tool to confirm:

"I've detected that this task requires [Auto-Edit/YOLO] permissions to [modify files/run shell commands]. OK to proceed?"

3. Execution

Execute gemini using the mode determined by the tier. If resuming a session, you can check /stats within the interactive prompt for token usage.

# Tier 0 (Plan — experimental read-only + sandbox)
gemini -p "<prompt>" --approval-mode plan --sandbox

# Tier 1 (Auto-Edit + sandbox — file edits allowed, shell gated, sandboxed)
gemini -p "<prompt>" --approval-mode auto_edit --sandbox

# Tier 2 (YOLO — sandbox is auto-enabled by --yolo)
gemini -p "<prompt>" --yolo
# equivalent: gemini -p "<prompt>" --approval-mode yolo  (sandbox auto-enabled)

Security Rules:

ALWAYS pass --sandbox for Tier 0 and Tier 1. --yolo enables it automatically for Tier 2.
Default sandbox backend uses Seatbelt (macOS) or bubblewrap (Linux). For stronger isolation, use GEMINI_SANDBOX=docker (requires Docker installed).
NEVER use --yolo / --approval-mode yolo without explicit confirmation of the risks.
ALWAYS use the most restrictive mode possible (prefer plan).
If you are unsure, default to plan and escalate only if gemini reports it cannot complete the task.
yolo mode can only be enabled via CLI flag. It cannot be persisted in settings.json to prevent accidental system-wide auto-approval.

Configuration

This skill leverages native gemini CLI flags to enforce the permission tiers.

Model Selection

The default model is gemini-2.5-pro. Switch to gemini-2.5-flash for faster, lower-cost responses:

gemini -m gemini-2.5-flash -p "<prompt>" --approval-mode plan --sandbox

Available models: gemini-2.5-pro (default, 1M token context), gemini-2.5-flash (fast fallback), gemini-3-flash.

Sandbox Backend and Network

For stronger container-level isolation use Docker as the sandbox backend:

GEMINI_SANDBOX=docker gemini -p "<prompt>" --approval-mode auto_edit --sandbox

Disable network access inside the sandbox via settings.json:

{
  "tools": {
    "sandboxNetworkAccess": false
  }
}

sandboxNetworkAccess defaults to false when sandbox is enabled — set it explicitly to make the policy auditable.

Remote Subagents (experimental — v0.33.0+)

Gemini CLI can delegate subtasks to remote agents via the Agent-to-Agent (A2A) protocol (JSON-RPC 2.0 over HTTPS). Delegation happens automatically when the main agent determines that a specialized subagent is better suited for a portion of the task.

# Enable experimental A2A features
gemini -p "<prompt>" --experimental-acp

# View active subagents in interactive mode
/subagents

Remote subagents use short-lived scoped tokens (expire in minutes) and communicate over authenticated HTTPS. Use --experimental-acp only in environments where you trust the remote agent endpoints.

Examples

Refer to references/usage-examples.md for concrete scenarios.