The Agent Skills Directory

[COMMAND_EXECUTION]: The script executes ffprobe to verify media files. This is limited to a specific binary and utilizes subprocess.run with a static list of arguments, minimizing the risk of shell injection.
[DATA_EXFILTRATION]: The skill implements strict data boundaries. It explicitly filters incoming and outgoing data, removing sensitive tokens, signed URLs, and provider secrets before producing a verdict.
[PROMPT_INJECTION]: The skill is configured with disable-model-invocation: true and does not interact with an LLM, making it immune to traditional prompt injection attacks.
[CREDENTIALS_UNSAFE]: The code contains robust regex-based sanitization (_safe_token and _safe_policy_code) to detect and discard strings resembling API keys (e.g., 'sk-', 'bearer') or internal request IDs from its output.

short-drama-delivery-audit