The Agent Skills Directory

[DYNAMIC_EXECUTION]: The scripts/browser_server.py file implements a mechanism to execute arbitrary Python strings using eval() and exec(). These strings are received via a local Unix socket, which allows the browser session to be controlled dynamically but also provides a broad execution primitive.\n- [DYNAMIC_EXECUTION]: The script scripts/run_browser_ops.py enables the execution of arbitrary code provided through command-line arguments using eval() and exec() to bypass standard script structure.\n- [EXTERNAL_DOWNLOADS]: The skill uses npx to download and run the Playwright MCP server from the npm registry during the initialization of browser tools in scripts/utils.py.\n- [COMMAND_EXECUTION]: The skill invokes system commands to manage the browser environment and, in scripts/utils_test.py, configures a debugger to listen on a local network port (localhost:5678).\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process text snapshots from external websites, which could contain malicious instructions. If the agent acts on these instructions, it could leverage the skill's dynamic execution capabilities to perform unauthorized actions.\n
Ingestion points: browser.snapshot() calls within the various search and checkout scripts.\n
Boundary markers: None identified.\n
Capability inventory: Arbitrary Python execution (eval/exec), browser automation, and network access.\n
Sanitization: None identified.

shopping-browser-automation