Skills
Navigate
skills/praxstack/skills-and-personas/teacher/Gen Agent Trust Hub

teacher

Fail

Audited by Gen Agent Trust Hub on Oct 20, 2026

Risk Level: HIGHDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill uses several techniques to execute or transform code at runtime.
  • skills/handdrawn-js-video/scripts/pydeps.py uses os.execve to restart the current script using a Python interpreter within a newly created virtual environment.
  • skills/explorable/scripts/check_code.mjs utilizes vm.runInContext to evaluate JavaScript logic from lesson files.
  • skills/explorable/assets/engine/code.js dynamically creates Web Workers using URL.createObjectURL and Blob to run the Pyodide environment in the browser.
  • [EXTERNAL_DOWNLOADS]: The skill fetches various resources from external services.
  • skills/handdrawn-js-video/scripts/pydeps.py invokes pip install to download Python dependencies if they are not already present in the environment.
  • skills/explorable/scripts/narrate.py downloads Kokoro and Piper neural voice models from GitHub and Hugging Face repositories.
  • [COMMAND_EXECUTION]: Arbitrary or semi-arbitrary command execution is present for testing and media processing.
  • skills/explorable/scripts/check_code.mjs uses spawnSync to execute g++ and python3 on code snippets defined in lesson files for verification purposes.
  • scripts/find_tool.py probes the host environment by checking for the existence of binaries on the system path and modules in various package registries.
  • Multiple scripts invoke ffmpeg and ffprobe for video validation and processing.
  • [INDIRECT_PROMPT_INJECTION]: The algo-sensei component and related tutoring modes ingest user-provided data structures, algorithm problems, and code snippets.
  • Ingestion Point: User input provided to the tutor modes (e.g., modes/tutor-mode.md, modes/review-mode.md).
  • Boundary Markers: Absent; the system prompts the user to "Share code" or "Share a problem" without explicitly defining markers or instruction-ignoring guards for that input.
  • Capability Inventory: The skill set includes capabilities to execute shell commands (subprocess), write files, and perform network requests.
  • Sanitization: Not explicitly implemented for the ingested algorithmic text.
Recommendations
  • HIGH: Downloads and executes remote code from: unknown (check file) - DO NOT USE without thorough review
Audit Metadata
Risk Level
HIGH
Analyzed
Oct 20, 2026, 06:44 AM