The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The file SKILL.md contains a hardcoded Google Gemini API key (AIzaSyDvvGGRbH4Os3Er0dYi0kE_AzE3_2b_Az8) in the Phase 3 bash example and the API configuration table.\n- [COMMAND_EXECUTION]: The skill executes a local script scripts/generate_image.py via python3 to interact with the Gemini API and write image files to the local system.\n- [DATA_EXFILTRATION]: Hardcoded local paths in SKILL.md (e.g., /Users/ugreen/Documents/obsidian/09image/) expose specific user directory structures and the system username.\n- [PROMPT_INJECTION]: The skill processes untrusted user input for brand descriptions and interpolates it directly into the image generation prompt and the output filename.\n
Ingestion points: User input for brand name, industry, and philosophy collected during Phase 1 in SKILL.md.\n
Boundary markers: No delimiters or safety instructions are used when interpolating user data into the prompt templates provided in references/logo-styles.md.\n
Capability inventory: The script scripts/generate_image.py has the ability to create directories (os.makedirs) and write binary data to the local file system using open().write().\n
Sanitization: There is no validation or sanitization for the '品牌名' variable, which is used to construct the --output file path, creating a risk of directory traversal if malicious characters are provided.

logo-batch-generator