video-editing
Pass
Audited by Gen Agent Trust Hub on Jul 13, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: Indirect Prompt Injection surface exists via the transcript-based editing feature.
- Ingestion points: Video transcripts from
recording_transcripts.segments_json(untrusted external data) are processed by an AI agent to propose editing cuts. - Boundary markers: No explicit delimiters or boundary markers for the transcript data are defined in this skill's instructions.
- Capability inventory: The skill utilizes
ffmpeg.wasmfor client-side rendering, enqueues server-side export jobs, and performs database writes torecordings.edits_jsonandrecordings.chapters_json. - Sanitization: The skill specifies that edit actions validate for non-overlapping ranges and throw errors on invalid input. Crucially, agent-proposed edits are written to a draft state (
editor-draft) for human review and explicit approval before being committed. - [EXTERNAL_DOWNLOADS]: Lazy-loads the well-known
@ffmpeg/ffmpeglibrary and associated WASM binaries for client-side video processing when the user initiates an export.
Audit Metadata