claude-typer

Warn

Audited by Gen Agent Trust Hub on Nov 21, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill uses npx or bunx to download and execute Node.js packages from the npm registry at runtime, specifically @remotion/cli and @remotion/tailwind-v4 (version 4.0.440).
  • [REMOTE_CODE_EXECUTION]: The video rendering process loads and executes a Remotion composition from a remote, third-party domain: https://www.laosunwendao.com.
  • [COMMAND_EXECUTION]: The Python script scripts/render_claude_typer.py uses subprocess.run() to execute shell commands (specifically npx and bunx runners).
  • [INDIRECT_PROMPT_INJECTION]: This category identifies an attack surface for processing untrusted data. * Ingestion points: The user-supplied prompt text enters the agent context in SKILL.md and is passed as a command-line argument to the rendering script. * Boundary markers: The prompt is encapsulated within a JSON string inside the --props argument to provide a structural boundary for the Remotion CLI. * Capability inventory: The skill has the capability to execute subprocesses, access the network to fetch rendering templates, and write video files to the current working directory. * Sanitization: The condense_prompt function sanitizes the text for use as a filename by removing illegal characters, but the full prompt is passed into the rendering engine for processing without sanitization.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 21, 2026, 01:18 AM