The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to ingest and process sensitive file formats such as .env, .tf (Terraform), and Dockerfile. These files often contain secrets, credentials, and private infrastructure details. Reading these files brings sensitive data into the agent's context during the diagramming process.\n- [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands (open on macOS or xdg-open on Linux) to launch the generated HTML diagram file. Executing commands on content generated by the agent is a risky operational pattern.\n- [INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface as it accepts and processes many file types (.md, .json, .py, etc.) from users to generate diagrams.\n
Ingestion points: Reads contents from various file paths provided by the user.\n
Boundary markers: There are no instructions or delimiters provided to ensure the agent ignores malicious instructions embedded within the processed files.\n
Capability inventory: Includes file writing (HTML diagram) and shell command execution (opening the file).\n
Sanitization: No validation or sanitization of input content is performed.\n- [EXTERNAL_DOWNLOADS]: The generated diagrams include a reference to fetch fonts from Google Fonts (fonts.googleapis.com) using @import. This is an external reference to a well-known and trusted service.\n- [DYNAMIC_EXECUTION]: The skill generates HTML files containing custom JavaScript for functionality like pan/zoom and coordinate calculations, which is then executed when the file is opened. This dynamic generation and execution of code represents a security risk.

diagram-creator