hyperframes

Warn

Audited by Gen Agent Trust Hub on May 20, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The script scripts/package-loader.mjs includes a bootstrapWithNpmInstall function that executes npm install to download and install Node.js packages at runtime if they are not found locally. It then proceeds to dynamically import() these packages from a temporary directory.
  • [EXTERNAL_DOWNLOADS]: In references/typography.md, the skill provides instructions to download font metadata from fonts.google.com using curl and subsequently execute a Python script (fontquery.py) on the downloaded data.
  • [DATA_EXFILTRATION]: The file references/transcript-guide.md provides explicit curl command patterns for sending local audio files to external APIs (OpenAI and Groq) for transcription. These commands use environment variables for authentication tokens ($OPENAI_API_KEY, $GROQ_API_KEY).
  • [COMMAND_EXECUTION]: The references/design-picker.md reference instructs the agent to launch a background network service using python3 -m http.server on the local machine to host a design selection interface.
  • [COMMAND_EXECUTION]: Multiple scripts, including scripts/package-loader.mjs, utilize node:child_process.spawnSync to execute shell commands such as npm, npm.cmd, and node.
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 20, 2026, 12:19 PM
Security Audit — agent-trust-hub — hyperframes