lazyedit-publish-workflow

Pass

Audited by Gen Agent Trust Hub on Oct 19, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill makes extensive use of system commands and subprocess calls to manage the video publishing pipeline.
  • Executes local Python scripts such as scripts/lazyedit_publish.py and scripts/lazyedit_music_package.py for orchestration.
  • Uses ssh to execute commands on remote hosts (e.g., lazyingart or $AUTOPUBLISH_SSH) to tail logs, manage tmux sessions, and run management scripts like scripts/manage_shipinhao_videos.py.
  • Performs direct database modifications using psql to repair transcription job statuses (e.g., inserting records into the transcriptions table).
  • Utilizes curl to interact with local and remote APIs (e.g., $LAZYEDIT_API/api/autopublish/queue) for monitoring and configuration management.
  • Employs ffmpeg for video processing and frame extraction.
  • [INDIRECT_PROMPT_INJECTION]: The skill defines a workflow that ingests external, potentially untrusted text data to influence AI-driven outputs.
  • Ingestion points: The skill reads external story, script, and prompt files via the --prompt-file, --correction-prompt-file, and --metadata-prompt-file arguments in SKILL.md.
  • Boundary markers: The skill relies on external validation scripts like scripts/validate_subtitle_timeline.py to ensure that AI-corrected subtitles maintain the original source timeline and cue structure.
  • Capability inventory: The environment has broad capabilities including shell command execution (ssh, psql), local network access (curl), and file system operations.
  • Sanitization: Instructions emphasize manual inspection of generated metadata and automated validation of subtitle timestamps to mitigate incorrect or malicious AI outputs.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 19, 2026, 06:58 PM