The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Installation instructions include piping remote scripts (install.sh and install.ps1) from a third-party GitHub repository (brennhill/Kaboom-Browser-AI-Devtools-MCP) directly into shell interpreters (e.g., curl | bash and irm | iex). This allows an unverified source to execute arbitrary code on the system.
[EXTERNAL_DOWNLOADS]: The skill requires downloading software binaries and browser extensions from a repository that is not affiliated with the identified vendor or recognized as a trusted organization.
[COMMAND_EXECUTION]: Manual setup instructions involve commands to change file permissions (chmod +x) and move files into system-level directories (/usr/local/bin/), which often requires administrative access.
[PROMPT_INJECTION]: The skill's architecture is vulnerable to indirect prompt injection because it processes data from untrusted web pages while providing the agent with browser interaction tools.
Ingestion points: Untrusted data enters the agent context via kaboom_console_get, kaboom_network_get, and kaboom_dom_query tools.
Boundary markers: There are no markers or instructions provided to isolate captured browser data from the agent's main command flow.
Capability inventory: The skill includes powerful tools for browser interaction such as kaboom_browser_type, kaboom_browser_upload, and kaboom_browser_navigate.
Sanitization: No sanitization or verification of the external browser-sourced content is implemented.

kaboom-browser-ai-devtools