The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/providers/google.ts uses execSync to run curl commands when a proxy is configured. The shell command is constructed by interpolating environment variables like HTTPS_PROXY and GOOGLE_BASE_URL without sufficient sanitization, which could allow shell injection if these environment variables are maliciously set.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to several external image generation services, including the vendor's own API at api.tu-zi.com and well-known services from Google, OpenAI, Alibaba (DashScope), and Replicate. These are documented neutrally as well-known service endpoints.
[DATA_EXFILTRATION]: The skill is designed to load and use sensitive credentials from the local file system. It reads .env files from the project root and the user's home directory (~/.tuzi-skills/.env), transmitting the discovered API keys to the respective external providers.
[PROMPT_INJECTION]: The skill handles untrusted data that could be used for indirect injection.
Ingestion points: Prompt data is ingested from CLI arguments and files (--promptfiles), and configuration is loaded from EXTEND.md files.
Boundary markers: No specific boundary markers are used to delimit user-provided prompt content when it is forwarded to external APIs.
Capability inventory: The skill has the capability to execute shell commands, perform network operations, and modify the local file system.
Sanitization: The implementation does not appear to sanitize prompt strings or configuration values sourced from the environment before use in API calls or shell commands.

tuzi-image-gen