oma-video
Warn
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
- [DYNAMIC_EXECUTION]: The skill utilizes internal driver scripts that perform dynamic loading of code from paths supplied via arguments.
- In
resources/mpt/driver.py, the script usessys.path.insert(0, mpt_dir)wherempt_diris a path provided in the input JSON, followed by importing modules from that path. - In
resources/playwright/record.mjs, the driver usesnode:module'screateRequireand dynamicimport()to load Playwright packages from a directory path provided via the--playwright-dirflag. - [COMMAND_EXECUTION]: The skill frequently spawns subprocesses to interface with media processing tools.
resources/mpt/driver.pyandresources/playwright/record.mjsexecuteffmpegandffprobefor video synthesis and probing.- The skill invokes
npx remotion renderto execute the vendored Remotion composition logic. - [INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it processes untrusted data for script generation.
- Ingestion points: The skill accepts a
brief(which can be a topic, README path, or code) and a--urlfor web-app capture. - Boundary markers: Asset schemas are validated using Zod (
RenderSpecSchema), though the core narrative content is generated from the brief. - Capability inventory: The skill can write to the local filesystem, execute shell commands, and interact with external image/voice APIs.
- Sanitization: While manifest data and logs redact tokens, the input brief text is interpolated into the script generation phase without explicit sanitization mentioned for instruction-like text patterns.
- [CREDENTIALS_UNSAFE]: The skill handles sensitive environment variables such as
PEXELS_API_KEYandRUNNINGHUB_API_KEYto authenticate with third-party providers. The implementation documentation claims to mask these keys in logs and manifests to prevent exposure.
Audit Metadata