The Agent Skills Directory

[DATA_EXFILTRATION] (HIGH): The generated API route recursively reads files from a path defined in help.config.json (articlesDir) and sends their contents to the Anthropic API. If this directory is maliciously or accidentally pointed to sensitive locations (e.g., .ssh/, .env, or /etc/), private credentials and system information will be exfiltrated to the AI provider and potentially the chat user. \n- [PROMPT_INJECTION] (HIGH): The skill creates an indirect prompt injection surface (Category 8). Ingestion points: User-provided messages and local documentation files. Boundary markers: Minimal string separators are used, which are insufficient for adversarial inputs. Capability inventory: The generated code has filesystem read and network request capabilities. Sanitization: No sanitization or validation is performed on the user input before it is sent to the AI model. \n- [COMMAND_EXECUTION] (MEDIUM): The skill performs automated file system operations to scaffold components and API routes into the user's project. The lack of explicit path validation during the file-writing process poses a risk of overwriting critical project files if the framework detection or configuration is manipulated.

help-beacon