skills/getedgehq/skills/reel-machine/Gen Agent Trust Hub

reel-machine

Pass

Audited by Gen Agent Trust Hub on Oct 25, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The scripts/caption_review.py script constructs an LLM prompt that includes the full transcript of processed video clips to perform a secondary review of captions. If a video contains spoken text designed as a prompt injection, it could influence the output of the review model.
  • Ingestion points: scripts/caption_review.py reads transcript data from transcript.json (generated by transcribe.py using ElevenLabs Scribe).
  • Boundary markers: The transcript is interpolated into the prompt string without explicit delimiters or warnings to the model to ignore instructions within the transcript text.
  • Capability inventory: The review script can execute arbitrary shell commands defined in REEL_REVIEW_CMD, and the overall skill has capabilities for file writing, network operations, and remote server access.
  • Sanitization: The transcript text is included in the prompt without sanitization or escaping.
  • [COMMAND_EXECUTION]: The skill allows for the execution of user-defined shell commands for uploading files and reviewing content.
  • scripts/schedule_reel.py executes the REEL_UPLOAD_CMD from the environment/config using subprocess.run(shell=True). While it uses shlex.quote for the file argument, the command itself is dynamically loaded.
  • scripts/caption_review.py executes the REEL_REVIEW_CMD using shlex.split, allowing users to hook in their preferred CLI-based LLM for content validation.
  • [REMOTE_CODE_EXECUTION]: The scripts/remote_render.sh script facilitates remote rendering by transferring a bundle (including scripts and a generated job.sh) to a user-defined SSH host and executing it. This is a core feature for offloading heavy video processing but represents a remote execution capability.
  • [EXTERNAL_DOWNLOADS]: Several scripts perform network requests to fetch resources:
  • scripts/setup.sh downloads the 'Funnel Display' font from Google's official fonts repository on GitHub.
  • scripts/logo_png.py fetches SVG icons from the Simple Icons project via the jsDelivr CDN.
  • scripts/commons.py downloads images from Wikimedia Commons using their official API.
  • scripts/transcribe.py uploads audio to the ElevenLabs Scribe API.
  • scripts/schedule_reel.py interacts with the Buffer GraphQL API to schedule posts.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 25, 2026, 03:37 AM