The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill automatically installs the 'agent-browser' package globally using 'npm install -g' and downloads browser-related system dependencies. While attributed to Vercel in the description, the package source is not restricted to a specific trusted registry URL.
[COMMAND_EXECUTION]: The skill executes multiple shell commands to check the operating system, manage background processes ('npm run dev &'), and interact with databases using command-line tools like 'psql' and 'sqlite3'.
[REMOTE_CODE_EXECUTION]: For databases other than Postgres or SQLite, the instructions direct the agent to 'write a small ad hoc script in the application's language, run it, then delete it', which constitutes the generation and execution of dynamic code at runtime.
[PROMPT_INJECTION]: The skill is susceptible to indirect injection as it analyzes and acts upon untrusted codebase content. * Ingestion points: Codebase structure, database schemas, and application logic are researched by three parallel sub-agents. * Boundary markers: There are no explicit delimiters or specific instructions provided to the sub-agents to ignore or isolate potentially malicious instructions embedded in the code or comments of the application being tested. * Capability inventory: The agent can execute shell commands, perform database queries, and directly modify source files ('Fix the code — make the correction directly'). * Sanitization: The skill does not perform sanitization or validation of the code content before using it to derive testing steps or logic.

e2e-test