sora

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS] (LOW): The skill utilizes 'uv run --with openai' to dynamically install the OpenAI SDK. Under the [TRUST-SCOPE-RULE], this is a low-risk finding as 'openai' is a trusted package from a recognized organization.
  • [COMMAND_EXECUTION] (LOW): The skill executes a local Python script ('scripts/sora.py'). Security is enhanced by the recommendation to use '--prompt-file' and '--no-augment' flags to mitigate shell-escaping and injection risks when processing user-controlled strings.
  • [PROMPT_INJECTION] (LOW): The skill possesses a surface for indirect prompt injection. Evidence Chain: 1. Ingestion points: User-provided prompts, batch JSONL files (SKILL.md), and reference images (references/video-api.md). 2. Boundary markers: Use of structured prompt augmentation templates with labeled sections (Scene, Subject, Action, etc.). 3. Capability inventory: Subprocess execution via 'uv run', file system write access for temporary files, and outbound network access to the OpenAI API. 4. Sanitization: Relies on OpenAI API-side content guardrails (no real people, no copyrighted content) and provides explicit instructions for the agent to avoid shell-escaping issues.
  • [CREDENTIALS_UNSAFE] (SAFE): The skill correctly handles 'OPENAI_API_KEY' as an environment variable and explicitly instructs the agent never to ask the user to paste the key directly into the chat session.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 06:23 PM