video-publisher
Warn
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill dynamically assembles Node.js scripts from local module fragments (core logic, platform-specific adapters, and dispatchers) and executes them within an Ego Lite browser context to automate complex platform interactions. This is a core architectural pattern for the tool's automation logic.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from subtitle files and video metadata to generate and verify publishing information, which presents a surface for injection since it interprets content from external files.
- Ingestion points: Subtitle files (.ass, .srt, .vtt) and content package fields parsed during the intake workflow (references/intake-workflow.md).
- Boundary markers: Absent for subtitle content interpolation into metadata suggestions; the system relies on LLM reasoning and user confirmation.
- Capability inventory: Full browser automation, local file system read/write access for state management, and execution of shell commands through the publisher orchestrator.
- Sanitization: Basic hashtag and description normalization is implemented (scripts/lib/content-package.mjs), but it lacks rigorous validation of external file content.
- [COMMAND_EXECUTION]: The skill manages the publishing lifecycle by spawning multiple Node.js subprocesses and external browser automation controllers through a specialized orchestrator (scripts/v2/publisher.mjs).
Audit Metadata