assemblyai
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFEMETADATA_POISONINGDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [METADATA_POISONING]: Significant contradictions exist between the skill's metadata and its technical implementation.
- The frontmatter and description claim the skill is for "Image generation/editing" and trigger on image-related requests.
- The implementation details, operation IDs, and sample requests (using
.mp3files) all describe "Speech to Text" functionality using AssemblyAI's Universal 3 Pro model. - The output processing section then reverts to claiming the model returns image URLs, which is inconsistent with speech-to-text transcription results.
- [DYNAMIC_EXECUTION]: The skill instructs the agent to dynamically load instructions and schemas at runtime.
- It directs the agent to fetch
https://www.pixazo.ai/models/assemblyai.mdto determine exact parameter names and accepted values. - Loading instructions from a remote URL allows for the behavior of the skill to change after installation.
- [EXTERNAL_DOWNLOADS]: The skill references the installation of additional components from external sources.
- It provides a command (
npx skills add Pixazo-AI/skills --skill '*') to download and install a broad set of additional skills from a GitHub repository. - [COMMAND_EXECUTION]: The skill instructs the agent to manage files and permissions on the local system.
- It requires creating a directory
~/.pixazo/and a fileapi-key. - It explicitly instructs the agent to run
chmod 600on the API key file to restrict permissions.
Audit Metadata