The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes repository-specific scripts such as bin/dev, bin/rails, and npm run dev to start the development server. It also relies on standard command-line tools like git and gh (GitHub CLI) to analyze branch differences and pull request metadata.
[CREDENTIALS_UNSAFE]: The skill attempts to read the PORT variable from sensitive environment files, specifically .env, .env.local, and .env.development. While the search is limited to finding the port number, accessing these files is a security-sensitive operation as they frequently contain application secrets.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources, including GitHub PR descriptions, file lists, and the content of live web pages during browser testing. This creates a vulnerability surface where untrusted content from a page or repository could attempt to influence the agent's browser interaction logic or test results.
[DYNAMIC_EXECUTION]: The skill utilizes the ToolSearch capability to dynamically load and configure browser automation tools (like Playwright MCP) at runtime based on the environment's availability. It also generates and executes shell scripts that incorporate dynamic port numbers and server paths determined during the discovery phase.

browser-smoke