byted-mediakit-editing

Pass

Audited by Gen Agent Trust Hub on Nov 5, 2026

Risk Level: SAFE
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes the 'mediakit-cli' binary to perform media processing. All shell commands are structured around this specific tool, using defined parameters for both Local and Cloud modes. The use of environment variables like MEDIAKIT_RUNTIME is a standard practice for the agent to identify its environment.
  • [DATA_EXPOSURE_AND_EXFILTRATION]: The skill handles media URLs and local file paths for processing. It supports exfiltrating results to Volcengine services (VOD/TOS) or user-specified URLs (callback_url). These actions are consistent with the primary purpose of a media editing service and do not target sensitive system files or credentials.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external URLs and JSON-formatted inputs (like subtitle lists or image arrays). While this creates an attack surface for indirect prompt injection via media metadata, the instructions mitigate this risk by requiring the agent to perform pre-checks on metadata (e.g., probe-audio-metadata) and by delegating the actual processing to a non-LLM CLI tool.
  • [NETWORK_OPERATIONS]: In Cloud mode, the skill initiates network requests to Volcengine's infrastructure for asynchronous task submission and status polling. This is transparently documented in the reference files as the intended behavior for cloud-based media processing.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 5, 2026, 09:02 AM