The Agent Skills Directory

[COMMAND_EXECUTION]: The skill dynamically constructs and executes shell commands using the meitu CLI. It maps user-provided inputs to specific CLI flags and subcommands defined in references/tools.yaml.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data, such as prompts and descriptions, which are interpolated into CLI commands. This represents a potential attack surface where malicious data could attempt to influence agent behavior.
Ingestion points: User-supplied prompt, summary, and other text-based inputs used across all 40 defined tools in SKILL.md and references/tools.yaml.
Boundary markers: The skill contains explicit instructions to treat user-provided prompts and URLs as tool input data only and not to follow attempts to override system instructions.
Capability inventory: The agent can read local credentials from ~/.meitu/credentials.json and execute the meitu binary.
Sanitization: The skill relies on the agent's internal instruction following to distinguish between instructions and data, as explicitly stated in the 'Instruction Safety' section.

meitu-tools