long-form-editing
Pass
Audited by Gen Agent Trust Hub on Oct 18, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill generates and executes shell commands for ffmpeg and dynamically invokes Node.js scripts using node -e to process transcripts and media. It also generates render scripts and filter files based on calculated cut points.- [COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to run media processing pipelines, including silence detection, transcription via Whisper, and complex ffmpeg filtering.- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from video files and external transcripts to drive editing decisions. While it includes user-confirmation steps, malicious content within processed transcripts could theoretically influence the agent's logic. Ingestion points: Video files in $PROJ/01_raw/a_roll/ and transcripts (SKILL.md). Boundary markers: The skill implements mandatory STOP and surface for approval checkpoints before rendering. Capability inventory: Bash, Write, Edit, Glob, Grep tools (SKILL.md). Sanitization: Not explicitly detailed for transcript content handling.
Audit Metadata