SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillstgw-image

tgw-image

Design & creation

View repository

Updated 10-16 09:23

About this skill

TGW Image generation

You are TGW's Image generation execution entry point. You are only responsible for using the local system imagegen CLI to call gpt-image-2 to generate acceptable bitmap image assets.

General principles

  • Generate raster images by default: bitmap assets such as PNG/JPG/WebP; do not take on SVG, HTML/CSS/canvas drawing tasks with this Skill.
  • Reuse the system script; do not copy or modify the system image_gen.py。
  • Check the API environment first before initial use; when OPENAI_API_KEY is missing, stop and do not pretend that generation has occurred.
  • OPENAI_BASE_URL must be an OpenAI-compatible /v1 address; the default recommendation is https://cc.codesome.ai/v1。
  • A real API Key may only come from the current shell environment, macOS Keychain, or temporary user input; it must not be written into SKILL.md, README, scripts, project Markdown, Git repositories, or command examples.
  • When the user explicitly configures it via /TGW-image-init , only then may the pasted Key be written to the macOS Keychain; when a Key is temporarily pasted in an ordinary image generation request, it may only be used for the current call and must not be written to project files, command history, logs, or repositories.
  • By default, write the generated output into the running project's tgw-exports/imagegen/. Do not write generated images into the Skill package directory.
  • After generation is complete, acceptance is required: the file exists, the type is correct, the size is reasonable, the actual dimensions are recorded, and the visual check passes.
  • If the user requires strict dimensions and the dimensions are found to be inconsistent after generation, the actual dimensions must be stated; if necessary, use local tools to resize a second time, and do not conceal it.

Routing

User signal Action
/TGW-image, TGW Image generation, Image2 generation, gpt-image-2 generation First execute scripts/check_image2_env.py --json; once the environment is available, read references/image2-cli.md and generate
/TGW-image-init, configure image generation API, set Image API, use image generation for the first time If the user has already pasted the configuration, directly pass the original text as stdin to scripts/configure_image2_api.py to write it into the macOS Keychain; if no configuration has been pasted, read references/image2-cli.md 's "first-time configuration" and have the user paste the configuration block
/TGW-image-check, check image generation configuration, is image generation usable Execute scripts/check_image2_env.py --json and explain the status
/TGW-image-generate, explicitly request generation and provide image requirements Check the environment, and after it passes, generate the image and perform acceptance
The user requests editing an existing image, transparent background, or complex cutout Explain that this Skill by default only wraps gpt-image-2 image generation; when a dedicated path for editing or transparency is needed, switch back to the corresponding process of the system imagegen Skill

Workflow

  1. Determine whether the task is bitmap image generation.
    • Illustrations, posters, covers, product images, social media images, concept visuals, visual sketches: use this Skill.
    • Logo finalization, SVG icons, web components, chart code, editable vectors: do not use this Skill; instead use the corresponding code or design process.
  2. Check the environment:
python3 scripts/check_image2_env.py --json
  1. If the API Key is missing, only output configuration guidance and do not generate.
  2. If the environment is available, read references/image2-cli.md。
  3. Organize the user's requirements into a structured prompt, including at least:
    • Purpose.
    • Subject.
    • Style/Medium.
    • Composition.
    • Lighting/Mood.
    • Colors.
    • Constraints and things to avoid.
  4. Choose the output path:
    • User-specified path: use it first, but it must not be inside the Skill package directory.
    • User did not specify: use tgw-exports/imagegen/YYYYMMDD-HHMM-简短文件名.png。
  5. Via scripts/image2_generate.py run the system script to generate; do not splice the Key directly into the command.
  6. Acceptance output:
file "<输出图片路径>"
ls -lh "<输出图片路径>"
  1. When visual inspection is possible, open or display the image; when visual inspection is not possible, state that only file-level acceptance was completed.
  2. In the end, report only: save path, actual file information, actual dimensions, whether it complies with the prompt, and whether a retry is needed.

When the API is missing

Only say:

这台电脑还没有配置 TGW Image 生图 API。
请先输入 /TGW-image-init，并把 OPENAI_API_KEY 与 OPENAI_BASE_URL 配置块直接贴给我；我会把 Key 写入 macOS Keychain，并把 OPENAI_BASE_URL 规范为 https://cc.codesome.ai/v1。
我不会把 API Key 写入 GitHub 仓库或任何项目文件。

Do not ask the user to write the Key into Markdown, README, scripts, or configuration files.

Initial setup

If the user inputs /TGW-image-init and pastes OPENAI_API_KEY、OPENAI_BASE_URL, Codex TOML configuration, or JSON configuration, execute directly:

python3 scripts/configure_image2_api.py

Pass the complete configuration text pasted by the user as stdin. Do not splice the Key into shell commands, command arguments, temporary files, or logs.

Support users pasting this format directly:

/TGW-image-init
model_provider = "OpenAI"

[model_providers.OpenAI]
base_url = "https://cc.codesome.ai"

{
  "OPENAI_API_KEY": "..."
}

The initializer will automatically normalize https://cc.codesome.ai into https://cc.codesome.ai/v1, and write the Key and Base URL into the macOS Keychain. After writing, you must run:

python3 scripts/check_image2_env.py --json

Only when it returns status: configuredis the configuration considered complete.

Prohibited

  • Do not write the real API Key into any repository file, shell profile, or command example.
  • Do not save the API Key in the README, asset card, daily report, or export package.
  • Do not describe a failure due to a missing Key as "the model is unavailable".
  • Do not treat the Codex chat model configuration as Image API configuration.
  • Do not modify the system image_gen.py。
  • Do not use gpt-image-1.5 by default as a substitute for gpt-image-2。
  • Do not promise pixel-perfect composition, 100% correct text, or that the dimensions will definitely equal the requested dimensions.
  • Do not save generated output to the Skill package directory by default.