What Is Mode C in the gpt-image-2 Skill? The Complete Guide to Prompt-Advisor Fallback

Mode C is the "Advisor" fallback mode in the gpt-image-2 skill that acts as a pure prompt-consultant, generating high-quality image prompts without attempting any API calls when the host environment lacks image generation capabilities.

The gpt-image-2 skill in the ConardLi/garden-skills repository operates in three distinct runtime modes, with Mode C serving as a critical fallback that ensures users retain access to professional prompt engineering even when image generation tools are unavailable. This mode transforms the skill into a structured prompt-writing assistant that prepares ready-to-use prompts for external tools like ChatGPT, Midjourney, or DALL·E 3.

How Mode C Detection Works

The skill determines its runtime mode by executing skills/gpt-image-2/scripts/check-mode.js immediately upon task initiation. This script evaluates the host environment to determine whether full image generation is possible.

Mode C activates when two specific conditions are met:

• The environment variable ENABLE_GARDEN_IMAGEGEN is not set
• The host agent provides no built-in image-generation tool

When both conditions evaluate to true, the skill automatically falls back to Mode C, bypassing all API-dependent functionality. According to the skill documentation in skills/gpt-image-2/SKILL.md, this detection occurs before any template processing begins, ensuring the skill never attempts unsupported API calls.

The Mode C Workflow: Prompt Generation Without APIs

Unlike Mode A (which calls scripts/generate.js) or Mode B (which uses host-native tools), Mode C completes the full prompt preparation workflow but stops before image generation. The skill follows the standard template selection and field-filling process, then renders the final prompt for manual use.

Prompt Preparation and Rendering

When operating in Mode C, the skill executes the following sequence:

1. Template Selection: Loads the specified template from the references/ directory
2. Field Population: Injects user-provided values into template variables
3. Prompt Rendering: Generates the complete, formatted prompt text
4. File Persistence: Saves the rendered prompt to disk
5. User Display: Outputs the prompt with usage instructions

Output Location and File Structure

Mode C saves all generated prompts to a dedicated directory using a timestamped naming convention:

garden-gpt-image-2/prompt/<task-slug>-<timestamp>.md

For example, a brand poster task started on April 24, 2026 at 15:30:45 would save to garden-gpt-image-2/prompt/brand-poster-20260424-153045.md. Each file contains the complete, formatted prompt ready for copy-paste into external image generation platforms.

User Interaction Model

After saving the file, Mode C explicitly informs the user that "it has not generated an image" and provides a recommendation to paste the rendered prompt into compatible tools such as ChatGPT, Midjourney, or DALL·E 3. This transparent communication ensures users understand they are receiving a prompt advisory service rather than image files.

Mode C vs. Mode A: Key Differences

Understanding the boundaries between operational modes prevents workflow errors:

Mode A (Full Generation)

• Calls skills/gpt-image-2/scripts/generate.js for image creation
• Calls skills/gpt-image-2/scripts/edit.js for image modification
• Requires ENABLE_GARDEN_IMAGEGEN environment variable or native host tools
• Produces actual image files

Mode C (Prompt Advisor)

• Never invokes generate.js or edit.js
• Requires only Node.js runtime and template access
• Produces markdown prompt files only
• Explicitly notifies users of the non-generation status

As implemented in the ConardLi/garden-skills repository, Mode C ensures the skill remains functional in restricted environments while maintaining the structured approach to prompt engineering defined in skills/gpt-image-2/references/prompt-writing.md.

Practical Example: Running Mode C Locally

To verify you are operating in Mode C or to intentionally use the prompt-advisor functionality, execute the mode detection script:

# Detect current runtime mode

node skills/gpt-image-2/scripts/check-mode.js

# Expected output: mode = C

To run a complete task in Mode C without image generation:

node skills/gpt-image-2/scripts/run-task.js \
  --template references/poster-and-campaigns/brand-poster.md \
  --field title="Eco-Friendly Water Bottle" \
  --field style="minimalist, pastel"

Expected Output:

✅ Mode C selected – acting as a prompt advisor
📄 Prompt saved to garden-gpt-image-2/prompt/brand-poster-20260424-153045.md
🗒️ Prompt:
"Create a minimalist pastel poster for an eco-friendly water bottle..."

💡 You can now paste the above prompt into any image tool (ChatGPT, Midjourney, DALL·E 3, etc.).

Note that in Mode C, the --template and --field arguments function identically to other modes, but the execution stops after the prompt rendering phase.

Summary

• Mode C is the prompt-advisor fallback in gpt-image-2 that activates when ENABLE_GARDEN_IMAGEGEN is unset and no host image tools are available.
• The mode is detected by scripts/check-mode.js before any task processing begins.
• Mode C executes the full template workflow but saves output to garden-gpt-image-2/prompt/<task-slug>-<timestamp>.md rather than generating images.
• The skill explicitly excludes scripts/generate.js and scripts/edit.js from Mode C execution, reserving these for Mode A only.
• Users receive ready-to-use prompts formatted for direct input into external generation platforms like ChatGPT, Midjourney, or DALL·E 3.

Frequently Asked Questions

What triggers Mode C in the gpt-image-2 skill?

Mode C triggers automatically when the scripts/check-mode.js script detects that the ENABLE_GARDEN_IMAGEGEN environment variable is not defined and the host agent provides no built-in image-generation capability. This dual-condition check ensures the skill only attempts API calls in supported environments.

How is Mode C different from Mode A?

Mode A executes scripts/generate.js and scripts/edit.js to produce actual image files, requiring either the ENABLE_GARDEN_IMAGEGEN environment variable or native host tools. Mode C skips these scripts entirely, acting solely as a prompt-consultant that prepares text files for manual use in external image generators.

Where does Mode C save the generated prompts?

Mode C persists rendered prompts to garden-gpt-image-2/prompt/<task-slug>-<timestamp>.md using the task identifier and current timestamp to ensure unique filenames. This directory structure keeps prompt assets organized and easily accessible for copy-paste workflows.

Can I manually force the skill to use Mode C?

While Mode C typically activates automatically based on environment detection, you can ensure it runs by unsetting ENABLE_GARDEN_IMAGEGEN and removing access to host image tools before executing scripts/run-task.js. The check-mode.js script will then confirm Mode C selection before proceeding with the prompt-only workflow.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}