gemini-cli-helper

name	Gemini CLI Helper
description	Run Gemini CLI commands from Claude with correct Node version. Use when user asks to run Gemini CLI, test Gemini traces, or debug Gemini telemetry.

Gemini CLI Helper

Run Gemini CLI commands with the correct Node.js version and configuration.

Quick Start

Run a Gemini prompt:

.claude/skills/gemini-cli-helper/scripts/gemini_run.sh "Your prompt here"

Check Gemini CLI status:

.claude/skills/gemini-cli-helper/scripts/gemini_status.sh

When to Use This Skill

Invoke this skill when:

• User asks to run Gemini CLI
• User wants to test Gemini trace visibility
• User asks to debug Gemini telemetry
• User mentions "gemini" CLI commands
• Coordinator needs to execute Gemini tasks

Available Scripts

scripts/gemini_run.sh <prompt> [--json]

Run a Gemini CLI prompt with correct Node version.

Usage:

# Simple prompt
.claude/skills/gemini-cli-helper/scripts/gemini_run.sh "Say hello"

# With JSON output
.claude/skills/gemini-cli-helper/scripts/gemini_run.sh "Say hello" --json

# Complex prompt with quotes
.claude/skills/gemini-cli-helper/scripts/gemini_run.sh "List 3 things about AILANG"

scripts/gemini_status.sh

Check Gemini CLI installation and configuration status.

Usage:

.claude/skills/gemini-cli-helper/scripts/gemini_status.sh

Output:

Gemini CLI Status
━━━━━━━━━━━━━━━━━
Node Version: v22.20.0 (required: v20+)
Gemini CLI: /Users/mark/.nvm/versions/node/v22.20.0/bin/gemini
Version: 0.21.1
GCP Project: multivac-internal-dev
Telemetry: Enabled (GCP Cloud Trace)

Critical Knowledge

Node Version Requirement

Gemini CLI requires Node.js v20 or higher due to regex flag syntax.

# WRONG - Will fail with "Invalid regular expression flags"
gemini --version  # Uses default Node which may be v18

# CORRECT - Use full path with Node v22
/Users/mark/.nvm/versions/node/v22.20.0/bin/node \
  /Users/mark/.nvm/versions/node/v22.20.0/lib/node_modules/@google/gemini-cli/dist/index.js \
  --version

Common Errors

Error	Cause	Fix
Invalid regular expression flags	Node < v20	Use full Node v22 path
MODULE_NOT_FOUND	Wrong CLI path	Check actual symlink target
Cannot find module	Incorrect dist path	Use dist/index.js not bin/cli.mjs

Telemetry Configuration

⚠️ CRITICAL: Telemetry is DISABLED by default!

To enable Gemini CLI traces in AILANG Observatory:

Option 1: Direct to GCP (Recommended)

Create or edit ~/.gemini/settings.json:

{
  "telemetry": {
    "enabled": true,
    "target": "gcp",
    "logPrompts": true
  }
}

Or use environment variables:

export GEMINI_TELEMETRY_ENABLED=true
export GEMINI_TELEMETRY_TARGET=gcp
export GOOGLE_CLOUD_PROJECT=multivac-internal-dev

Option 2: Via OTLP Collector

For more control (e.g., sending to multiple backends):

1. Create ~/.gemini/collector-config.yaml:

receivers:
  otlp:
    protocols:
      grpc:
        endpoint: 0.0.0.0:4317
      http:
        endpoint: 0.0.0.0:4318

exporters:
  googlecloud:
    project: multivac-internal-dev
  otlp/ailang:
    endpoint: localhost:1957
    tls:
      insecure: true

service:
  pipelines:
    traces:
      receivers: [otlp]
      exporters: [googlecloud, otlp/ailang]

2. Enable collector mode in settings:

{
  "telemetry": {
    "enabled": true,
    "target": "local",
    "useCollector": true,
    "otlpEndpoint": "http://localhost:4317"
  }
}

Traces appear in:

• GCP Cloud Trace Console
• AILANG Observatory (via composite backend or direct OTLP)

Workflow

1. Check Status First

.claude/skills/gemini-cli-helper/scripts/gemini_status.sh

2. Run Prompt

.claude/skills/gemini-cli-helper/scripts/gemini_run.sh "Your prompt"

3. Verify Traces (Optional)

curl -s "http://localhost:1957/api/observatory/traces?limit=10" | jq '.[].service_name' | sort -u

Resources

Reference Guide

See resources/reference.md for:

• Full path configuration
• nvm setup instructions
• Troubleshooting guide
• GCP telemetry details

Notes

• Always use the scripts in this skill instead of direct gemini command
• Scripts handle Node version detection automatically
• Telemetry goes to GCP Cloud Trace, then imported to Observatory
• JSON output mode recommended for programmatic use