The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the meitu command-line tool to perform image generation and authentication verification (meitu auth verify). This is the primary mechanism for fulfilling user requests.
[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the meitu-cli package via NPM (npm install -g meitu-cli). This package is the official tool for the Meitu AI platform, which matches the skill's authorship.
[DATA_EXFILTRATION]: The skill requests read access to ~/.meitu/credentials.json. This is the standard configuration file for the Meitu CLI and is necessary for authenticating API calls to the vendor's service. The use of environment variables for keys is also supported as a safe alternative.
[PROMPT_INJECTION]: An analysis of the instructions and reference files found no attempts to bypass safety filters, extract system prompts, or override agent constraints.
[COMMAND_EXECUTION]: The skill contains a vulnerability surface for indirect prompt injection as it processes untrusted user input (design briefs and articles) to construct prompts for the meitu CLI.
Ingestion points: User-provided text processed in the Execute step of SKILL.md.
Boundary markers: The skill uses structured JSON templates and reasoning steps (defined in references/output-formats.md) to isolate design instructions, though explicit sanitization is handled by the downstream generative model.
Capability inventory: The skill has the capability to execute shell commands (meitu) and write files to the workspace (~/.openclaw/workspace/visual/).
Sanitization: Content is passed to an image generation model rather than being executed as code, which significantly limits the impact of potential injection.

meitu-poster