The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides shell command templates, such as ls <编译产物目录>/project.config.json, which interpolate user-provided directory paths. This creates a potential surface for command injection if the input is not sanitized by the agent using the skill.
[EXTERNAL_DOWNLOADS]: The skill instructs the installation of the miniprogram-ci package from the official registry. This is the standard library for WeChat mini-program CI/CD automation.
[DATA_EXFILTRATION]: The generated scripts handle sensitive WeChat private keys (private.*.key) required for authentication. The skill correctly identifies these as sensitive and includes mandatory safety warnings to prevent them from being committed to version control, recommending CI/CD secrets instead.
[PROMPT_INJECTION]: This skill presents an indirect prompt injection surface because it processes untrusted user-supplied project metadata which is then used in file system and network operations.
Ingestion points: Variable inputs such as MP_PROJECT_PATH, MP_APPID, and MP_ROBOT defined in SKILL.md.
Boundary markers: No boundary markers or 'ignore' instructions are present in the provided bash templates.
Capability inventory: File system commands (ls), Node.js file operations (fs.mkdirSync, fs.writeFileSync), and network requests through the miniprogram-ci library.
Sanitization: The documentation does not provide sanitization or validation logic for user-provided strings before they are interpolated into the setup scripts.

miniprogram-ci