The Agent Skills Directory

[DATA_EXFILTRATION]: Accesses the configuration file ~/.meitu/credentials.json to retrieve API keys. This facilitates authentication between the meitu CLI and the vendor's OpenAPI by reading sensitive local configuration data.
[COMMAND_EXECUTION]: Executes the meitu CLI and system commands such as curl and rm. The curl command is utilized as a fallback to download intermediate image files from vendor-provided URLs to the /tmp/ directory, while rm manages the deletion of these temporary files.
[INDIRECT_PROMPT_INJECTION]: Ingests user-provided instructions and visual analysis results that are interpolated into command-line arguments for image processing. This creates an attack surface where maliciously crafted inputs could potentially influence the behavior of the underlying image editing models.
Ingestion points: User instructions and visual analysis results.
Boundary markers: None identified in the prompt construction logic.
Capability inventory: Execution of meitu, curl, and rm commands, and file system access.
Sanitization: No explicit input sanitization is performed.
[EXTERNAL_DOWNLOADS]: Recommends the installation of the meitu-cli tool from the official npm registry to ensure the necessary environment for processing is available.

meitu-image-fix