cm-imagegen

CM Image Generation Skill

Generates or edits images for the current project (for example website assets, game assets, UI mockups, product mockups, wireframes, logo design, photorealistic images, or infographics).

This skill follows the same request semantics, prompt workflow, decision tree, and output expectations as the official imagegen skill. The only intentional difference is the execution layer: image requests are sent to CodexManager's OpenAI-compatible Images API using the current Codex provider base_url and auth.json API key.

Top-level modes and rules

This skill has one top-level mode:

• CodexManager CLI mode: bundled scripts/cm_image_gen.py CLI. It uses the current Codex configuration and calls CodexManager's image endpoints.

The CLI exposes two subcommands:

• generate
• edit

Rules:

• Use this skill only when the user explicitly asks for CodexManager image generation or when the official built-in image_gen tool is unavailable in API-key/provider mode.
• Do not ask the user for a separate image API key or image base URL by default.
• Do not create one-off SDK runners.
• Never modify the official system imagegen skill.

CodexManager config policy:

• If CODEXMANAGER_IMAGE_BASE_URL is set, use it as the image request base URL.
• Otherwise, read the active model_provider from $CODEX_HOME/config.toml.
• Read that provider's base_url from [model_providers.<provider>].
• Read OPENAI_API_KEY from $CODEX_HOME/auth.json.
• Use CODEXMANAGER_IMAGE_MODEL only for the image model override; otherwise use gpt-image-2 for OpenAI-compatible Images providers, or image-01 when the provider base URL targets the image_generation API (minimax.io / minimaxi.com hosts).
• When the provider base URL targets the image_generation API, the CLI calls POST /v1/image_generation with aspect_ratio/width/height/seed/n/prompt_optimizer/response_format request fields, passes --image inputs as subject_reference, and parses data.image_urls / data.image_base64 + base_resp.status_code; otherwise it calls the OpenAI Images /images/generations and /images/edits endpoints and parses data[].b64_json.
• Use CODEXMANAGER_IMAGE_OUTPUT_DIR only for the output directory override; otherwise save under the current working directory's generated-images/ folder.