The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches documentation, prompt cookbooks, and model metadata from external repositories and API endpoints.\n
Evidence: scripts/refresh_references.sh downloads reference markdown files from raw.githubusercontent.com/ZenMux/zenmux-doc/ and raw.githubusercontent.com/YouMind-OpenLab/.\n
Evidence: scripts/list_models.sh fetches live model catalogs from zenmux.ai.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user prompts and reference image data, which are then used to build instructions for image generation models. This creates a surface where embedded instructions in the input could potentially influence agent behavior.\n
Ingestion points: User-provided prompt text and reference image URLs/paths processed in SKILL.md and scripts/image-common.ts.\n
Boundary markers: SKILL.md provides instructions for the agent to optimize prompts using specific scene/subject ordering and restating invariants.\n
Capability inventory: The skill calls zenmux.ai APIs and writes files to the project's .context directory via scripts/generate-openai.ts and scripts/generate-gemini.ts.\n
Sanitization: Input filenames are sanitized using the slugifyObjectName helper in scripts/image-common.ts.\n- [COMMAND_EXECUTION]: Executes local scripts to manage the workspace and perform generation tasks.\n
Evidence: SKILL.md instructs the agent to run node and bash scripts located in the scripts/ directory, such as workspace.mjs and refresh_references.sh.

zenmux-image-generation