The Agent Skills Directory

COMMAND_EXECUTION (HIGH): Path Traversal and Arbitrary File Write vulnerability in output resolution.
Evidence: In scripts/infographic_cli.py, the resolve_output function accepts an unvalidated string from the --output CLI argument and converts it directly into a Path object. It then calls mkdir(parents=True, exist_ok=True) on this path.
Impact: An attacker could provide a path like ../../../../.ssh/authorized_keys or ../../../../etc/passwd to attempt to overwrite sensitive system files during the infographic generation process.
REMOTE_CODE_EXECUTION (MEDIUM): Dynamic Module Loading via Computed Paths.
Evidence: The load_generator function in scripts/infographic_cli.py utilizes importlib.util.spec_from_file_location and spec.loader.exec_module to execute a Python script located at a path relative to the current file's parent directories.
Impact: If an attacker can influence the directory structure or place a file at the calculated GENERATOR_PATH, they can achieve arbitrary code execution when the CLI is run.
PROMPT_INJECTION (MEDIUM): Indirect Prompt Injection Surface (Category 8).
Ingestion points: The read_data function in scripts/infographic_cli.py ingests arbitrary JSON data via --data or --data-file arguments.
Boundary markers: None. Data is parsed and passed directly to the generator module without delimiters or instructions to ignore embedded commands.
Capability inventory: File system write access (PNG output) and directory creation.
Sanitization: Absent. The skill does not validate or sanitize strings within the JSON data (e.g., stat, label, title), which could contain malicious instructions for the downstream rendering engine or the agent itself.

infographic-generator