video-review-canvas
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill implements a feedback loop by fetching reviewer notes from the
here.nowservice, creating an attack surface where untrusted third-party content is processed by the agent.\n - Ingestion points: Notes are retrieved from the
here.nowplatform via thescripts/read-notes.mjsscript.\n - Boundary markers: The skill lacks explicit separators or instructions for the agent to ignore or isolate instructions embedded within these external notes.\n
- Capability inventory: The agent is designed to interpret these notes to inform subsequent video editing steps and project deliveries.\n
- Sanitization: External notes are handled as raw text strings without validation against prompt injection patterns.\n- [COMMAND_EXECUTION]: The skill utilizes
execFileSyncto run theffprobebinary for extracting video metadata inscripts/build-canvas.mjs. Additionally, it invokes a shell script (publish.sh) for project deployment.\n- [EXTERNAL_DOWNLOADS]: The skill performs network operations to exchange review data and publish project assets to thehere.nowplatform.
Audit Metadata