brainstorming

Warn

Audited by Gen Agent Trust Hub on May 9, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill requires the user or agent to download multiple scripts, including start-server.sh, index.js, and package.json, from an external GitHub repository (github.com/obra/superpowers) to enable the 'Visual Companion' functionality.
  • [REMOTE_CODE_EXECUTION]: The 'Visual Companion' feature involves executing unverified external scripts (e.g., shell and Node.js code) provided in the external repository, posing a risk of remote code execution if the source content is malicious.
  • [COMMAND_EXECUTION]: The skill uses shell commands to interact with the local environment, specifically to start and stop a background web server process.
  • [PROMPT_INJECTION]: The skill reads project context (files, documentation, commits) and user interaction logs from a local file (.events) to inform its design process. The lack of explicit sanitization or structured boundary markers when processing this untrusted data creates a surface for indirect prompt injection attacks.
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 9, 2026, 05:17 AM