The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a setup script (tool/setup.sh) and instructions to run commands for registration (narova providers add). These are standard for local development tools and do not involve remote command injection or suspicious execution patterns.
[DATA_EXFILTRATION]: The skill communicates with api.dev.runwayml.com to submit video generation tasks. This is the primary purpose of the skill. The code explicitly prevents the inclusion of the RUNWAYML_API_SECRET in local logs or metadata responses, adhering to safe data handling practices.
[CREDENTIALS_UNSAFE]: The skill requires RUNWAYML_API_SECRET but correctly instructs the user to provide it via environment variables rather than hardcoding. It uses the secret only for Authorization: Bearer headers in HTTPS requests to the official Runway API.
[EXTERNAL_DOWNLOADS]: The worker downloads generated video files from the Runway CDN to a local staging directory. The implementation includes safety checks such as a 1 GiB size limit, verification of HTTPS protocols, and atomic file replacement using temporary files.
[PROMPT_INJECTION]: While the skill processes user prompts, it acts as a pass-through to the Runway API. It implements length limits (1,000 characters) and structural validation but does not have local execution capabilities that could be subverted by prompt injection.
[DYNAMIC_EXECUTION]: The test suite (tests/test_worker.py) uses importlib to load the worker script for unit testing. This is a standard Python testing pattern and does not constitute a runtime vulnerability in the skill's operational path.

narova-runway