The Agent Skills Directory

[CREDENTIALS_UNSAFE]: A hardcoded Zhipu AI API key (da8df5ba954341829f7afd05ca23a889.RrJoTsbaAkGYA6ZU) is present in INSTALL.md, QUICKSTART.md, and README-LITE.md. This could lead to unauthorized use of the API and associated costs.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input and task metadata directly into the LLM context.
Ingestion points: The userMessage and taskName fields are received via SkillInput in src/index.ts and passed to CompanionService.generateResponse in src/companion.ts.
Boundary markers: There are no delimiters or warnings to the LLM to ignore instructions embedded within the user message.
Capability inventory: The skill has the ability to make network requests to Zhipu AI APIs (axios) and write files to the local filesystem (fs.writeFileSync in src/image-generator.ts).
Sanitization: No validation or escaping of the user-supplied strings is performed before they are interpolated into the system or user prompts.
[COMMAND_EXECUTION]: The test script src/test.ts references an absolute local filesystem path (D:\tool\StepFun\resources\chat.png) for image processing, which exposes local environment details and creates a rigid dependency on the host's directory structure.

companion-skill