The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script (scripts/generate_image.py) using shell commands that incorporate user-provided text as arguments.
Evidence: SKILL.md provides a command template: python ${CLAUDE_SKILL_DIR}/scripts/generate_image.py --prompt "<提示词>".
Risk: There is a potential for command injection if the user-supplied prompt text contains shell metacharacters and is not correctly escaped by the agent when invoking the shell.
[EXTERNAL_DOWNLOADS]: The skill initiates network connections to an external service to perform its primary function.
Evidence: scripts/generate_image.py uses urllib.request to send POST requests to https://zenmux.ai/api/v1/images/generations and download image data.
Context: This behavior is documented as the core functionality for image generation through the ZenMux platform.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input to generate content, which could be exploited to influence the agent's behavior or output.
Ingestion points: User descriptions and instructions are ingested via SKILL.md to design image prompts.
Boundary markers: The skill explicitly requires the agent to present the finalized prompt and parameters to the user for confirmation before any script execution occurs.
Capability inventory: The skill has the capability to execute a local Python script, perform network operations, and write files to the AI_output/ directory.
Sanitization: The Python script uses argparse for parameter handling, and the skill instructions include a mandatory visual quality check of the generated output.

generate-image