clipping-streamer-shorts
Warn
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATIONREMOTE_CODE_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions and scripts execute several external command-line tools. Scripts
scripts/clip.py,scripts/captions.ts, andscripts/verify-bleeps.tsusesubprocess.runorBun.spawnSyncto invokeyt-dlp,ffmpeg,ffprobe, anddiffusion.SKILL.mdprovides instructions for the agent to manually execute these tools with user-provided parameters.\n- [EXTERNAL_DOWNLOADS]: The skill downloads content from external sources.yt-dlpis used to fetch video stream URLs and metadata from YouTube.scripts/clip.pyusesurllib.requestto download specific byte ranges of video and audio streams. The skill also suggests usingyt-dlp -Uto update the tool, which downloads and installs code from a remote server.\n- [DYNAMIC_EXECUTION]: The skill uses dynamic code loading mechanisms.scripts/captions.tsandscripts/verify-bleeps.tsboth useimport(pathToFileURL(...))to load local configuration files (segments.tsandbleeps.ts) from a computed path based on the project directory argument.\n- [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through external data processing. \n - Ingestion points: The skill reads YouTube auto-captions (
json3format) and local documentation files (decisions.md). \n - Boundary markers: No explicit delimiters or instructions are used to prevent the agent from obeying commands embedded in these data sources. \n
- Capability inventory: The environment allows for file system writes, network requests, and arbitrary command execution. \n
- Sanitization: There is no sanitization or validation of the ingested text to filter out potential prompt injection attacks.\n- [DATA_EXFILTRATION]: The skill references absolute file paths on the local system.
SKILL.mdcontains hardcoded paths to a specific user's directory structure (C:\Users\DavidOdunjo\Codebase\YouTube\...andC:\Users\DavidOdunjo\Videos\TheShortlist\...). While these appear to be configuration for the author's environment, they expose internal directory structures.\n- [REMOTE_CODE_EXECUTION]: The skill instructions suggest usingyt-dlp -Uto update the tool, which involves downloading and executing new code from a remote server, constituting a remote code execution pattern.
Audit Metadata