The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/create_pptx.py uses subprocess.run to call an external image generation tool.
Evidence: The generate_image function executes uv run on a script path located in a sibling directory (nano-banana-pro).
Risk: The --prompt argument passed to this command is derived directly from user-supplied outline or JSON files, allowing user-influenced strings to be passed as shell arguments.
[PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection due to how it processes external data files.
Ingestion points: scripts/create_pptx.py reads and parses input from files specified via the --outline or --json arguments.
Boundary markers: Absent. The script does not use delimiters or provide instructions to the underlying logic to ignore potentially malicious directives embedded in the content.
Capability inventory: The skill possesses command execution capabilities via subprocess.run in scripts/create_pptx.py.
Sanitization: Absent. There is no evidence of validation, escaping, or filtering of the content extracted from the data files before it is used to generate presentations or passed to external scripts.
[CREDENTIALS_UNSAFE]: The skill's main generation script accesses sensitive authentication tokens from the environment.
Evidence: scripts/create_pptx.py retrieves TWENTY_API_TOKEN and TWENTY_API_URL using os.environ.get for CRM data integration.
Risk: While used for legitimate functionality, these credentials are accessible within a script that also processes untrusted user data and executes external commands.

pptx-creator