The Agent Skills Directory

[COMMAND_EXECUTION]: The skill constructs shell commands by interpolating user-controlled variables (image, width, height, and extend_prompt) directly into the meitu image-transform command string. Without explicit sanitization or escaping of these inputs, a malicious user could potentially inject arbitrary shell commands via specially crafted input strings.
[INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface for indirect injection as it processes external image URLs and user-defined prompts that are passed to executable tools.
Ingestion points: 'image' (URL/path) and 'extend_prompt' fields in SKILL.md.
Boundary markers: None identified in the workflow instructions.
Capability inventory: Execution of the 'meitu' CLI tool with file read and write permissions in specified directories.
Sanitization: Instructions mention checking image accessibility and dimension ranges, but do not specify escaping or sanitization of input strings against shell injection.
[CREDENTIALS_UNSAFE]: The skill metadata and requirements specify access to sensitive credential files at ~/.meitu/credentials.json to retrieve API keys. While this is standard for the functionality of the associated vendor tool, it represents access to sensitive data within the agent's environment.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the meitu-cli package from the NPM registry to function. This is a vendor-provided tool required for the skill's core workflow.

meitu-image-adapt