skill-manager

Warn

Audited by Gen Agent Trust Hub on Oct 29, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The install.sh script executes several high-impact shell commands including osacompile to create a macOS application, defaults write to modify the macOS Dock configuration, and killall Dock to restart system services.
  • [COMMAND_EXECUTION]: The open.sh script uses nohup to spawn background processes and curl to probe local ports, which are used to maintain the local server's lifecycle.
  • [DYNAMIC_EXECUTION]: The skill uses Bun.serve to run a TypeScript-based server (server.ts) that performs direct file system operations. It reads from and writes to ~/.claude/settings.json based on inputs received via its /api/skills POST endpoint.
  • [DATA_EXPOSURE]: The local server provides a web interface that lists all files in ~/.claude/skills and their full descriptions. While bound to 127.0.0.1, any local process or website (via CSRF if not carefully guarded, though the server is intended for local use) could potentially interact with this API to read skill metadata or disable security-related skills.
  • [PRIVILEGE_ESCALATION]: The installation script attempts to write to /Applications and falls back to ~/Applications if permissions are insufficient, effectively installing a persistent executable on the user's path.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 29, 2026, 03:43 PM