The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill is designed to read and use Meitu-specific credentials from ~/.meitu/credentials.json or environment variables (MEITU_OPENAPI_ACCESS_KEY, MEITU_OPENAPI_SECRET_KEY) to authenticate CLI operations.
[EXTERNAL_DOWNLOADS]: The instructions direct the user to install the meitu-cli package via npm install -g meitu-cli@latest. This is a vendor-provided dependency required for the skill's primary functionality.
[COMMAND_EXECUTION]: The skill executes multiple shell commands using the meitu CLI, including meitu auth verify, meitu image-cutout, meitu image-edit, and meitu image-superres-enhance to process images.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it processes external image data and interpolates user-provided product descriptions into prompt templates.
Ingestion points: User-uploaded images and product descriptions are ingested in SKILL.md (Execute Step 1).
Boundary markers: The skill uses structured prompt templates defined in references/prompts.md to wrap user-provided descriptions and scene data.
Capability inventory: The skill has the capability to execute shell commands (meitu CLI) and write files to the local project and workspace directories (SKILL.md permissions section).
Sanitization: The skill implements a 5-level error degradation strategy to simplify prompts if generation fails, which acts as a basic form of sanitization against complex malicious prompt constructions.

meitu-product-view