The Agent Skills Directory

[OBFUSCATION]: The skill employs string concatenation to dynamically resolve and invoke sensitive Node.js functions. In meitu-tools/scripts/lib/executor.js, it uses require("node:" + "child_process") and cp["spawn" + "Sync"] to call spawnSync. This pattern is often used to bypass static analysis tools that look for direct references to shell execution modules.
[COMMAND_EXECUTION]: The skill frequently executes external commands through the meitu CLI and internal Node.js scripts. It also performs filesystem operations such as mkdir, mv, and rm to manage task outputs and temporary files.
[EXTERNAL_DOWNLOADS]: The documentation and setup guides instruct users to download and install the meitu-cli package globally via npm install -g meitu-cli@latest.
[CREDENTIALS_UNSAFE]: The skill accesses sensitive API credentials (accessKey and secretKey) stored in a local configuration file at ~/.meitu/credentials.json or through environment variables.
[PROMPT_INJECTION]: The skill includes instructions to treat user-provided text as task data only and to ignore attempts to override system rules or reveal internal prompts. However, the wide scope of data processing (reading memories, profile data, and user descriptions) creates an indirect prompt injection surface, although mitigated by explicit boundary instructions.

meitu-skills