gemini

NAME

gemini - execute non-interactive Google Gemini prompts

SYNOPSIS

state-name:
  tool: gemini
  arguments:
    prompt: <prompt-text>
    model: <model-name>       # optional
    timeout: <duration>       # optional
    output-format: <format>   # optional

DESCRIPTION

The gemini tool invokes Google’s Gemini CLI for AI-powered tasks. It sends a prompt using the -p flag and captures the response.

The tool runs in the project root directory for full codebase context.

ARGUMENTS

Argument	Required	Default	Description
`prompt`	Yes	-	The prompt to send to Gemini
`model`	No	Default model	Model to use (e.g., `gemini-flash`)
`timeout`	No	`5m`	Maximum execution time
`output-format`	No	-	Output format

Timeout Format

30s - 30 seconds
5m - 5 minutes
1h - 1 hour

EXIT CODES

Code	Meaning
`0`	Success
`1`	Gemini error or authentication issue
`2`	Gemini CLI not found
`124`	Timeout exceeded

OUTPUT

Both stdout and stderr are merged for capture:

analyze:
  tool: gemini
  arguments:
    prompt: "Summarize this document"
  output: var(summary)

EXAMPLES

Simple Prompt

quick-question:
  tool: gemini
  arguments:
    prompt: "What's the best practice for error handling in Go?"

With File Content

review-code:
  tool: gemini
  arguments:
    prompt: |
      Review this code for potential improvements:

      {{ file(./src/main.go) }}
  output: var(review)

With Model Selection

fast-task:
  tool: gemini
  arguments:
    prompt: "Generate a brief summary"
    model: gemini-flash

Error Handling

gemini-task:
  tool: gemini
  arguments:
    prompt: "Analyze codebase architecture"
    timeout: 10m
  on-error:
    1: check-auth
    2: gemini-not-installed
    124: handle-timeout
    _: general-error
  next: use-result

check-auth:
  tool: bash
  arguments:
    command: |
      echo "Gemini authentication issue. Run 'gemini' interactively to log in."
      exit 1

gemini-not-installed:
  tool: bash
  arguments:
    command: |
      echo "Gemini CLI not found. Install from https://github.com/google-gemini/gemini-cli"
      exit 1

In a Multi-AI Workflow

# Get perspectives from multiple AIs
start-at: get-claude-opinion

states:
  get-claude-opinion:
    tool: claude
    arguments:
      prompt: "How would you implement {{ feature }}?"
    output: var(claude_response)
    next: get-gemini-opinion

  get-gemini-opinion:
    tool: gemini
    arguments:
      prompt: "How would you implement {{ feature }}?"
    output: var(gemini_response)
    next: compare

  compare:
    tool: bash
    arguments:
      command: |
        echo "=== Claude's Approach ==="
        echo "{{ claude_response }}"
        echo ""
        echo "=== Gemini's Approach ==="
        echo "{{ gemini_response }}"

INSTALLATION

Install the Gemini CLI:

# Follow instructions at https://github.com/google-gemini/gemini-cli

Authenticate by either:

Running gemini interactively to complete OAuth login
Setting the GEMINI_API_KEY environment variable

RATE LIMITS

The free tier has limits:

60 requests per minute
1,000 requests per day

If rate limited (exit code 1 with rate limit message), consider:

Using gemini-flash model (lower limits)
Adding delays between requests
Upgrading to a paid plan

WORKING DIRECTORY

The tool runs in the project root directory, giving Gemini access to the full project context.