The Agent Skills Directory

[DATA_EXFILTRATION]: The skill requires read access to sensitive files including ~/.meitu/credentials.json and ~/.meitu/tool-registry.json. These files are vendor-specific and used to manage API credentials for the Meitu platform.
[COMMAND_EXECUTION]: The skill relies on the execution of the meitu CLI tool via shell commands. It defines workflows that assemble and run commands such as meitu image-transform --image_url <url> --rotate 90 ... using user-supplied parameters.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to how it handles untrusted user input.
Ingestion points: The skill ingests user-provided image_url and prompt strings in the image transformation and AI outpainting workflows.
Boundary markers: No boundary markers or delimiters are specified to isolate user-supplied content from the rest of the generated shell command.
Capability inventory: The skill is granted permission to execute shell commands (meitu), read system-level configuration files, and write to local directories (./output/).
Sanitization: The documentation does not describe any sanitization, validation, or escaping logic for the image_url or prompt inputs. Maliciously crafted inputs containing shell metacharacters (e.g., backticks or semicolons) could potentially lead to command injection.

image-transform