The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes utilities designed to execute shell commands. The documentation in references/browser-playwright.md describes scripts/with_server.py, which is used to launch arbitrary local servers (e.g., npm run dev) and run automation commands. Additionally, references/typescript-e2e/kafka/isolation.md provides an implementation of restartKafkaContainer that uses child_process.exec to execute docker restart commands on the host system. While functional for E2E testing, these patterns provide a surface for arbitrary command execution.
[INDIRECT_PROMPT_INJECTION]: The scripts/browser_trace.py utility is designed to capture complete browser state (DOM, console logs, network traffic) from target URLs and dump them into local files. The skill instructions in references/browser-trace.md and references/browser-playwright.md guide the agent to read and search these files using grep. This creates a high-risk surface for indirect prompt injection, as malicious content on a target website could influence the agent's behavior during analysis.
[EXTERNAL_DOWNLOADS]: The SKILL.md file contains instructions to install several external skills (e.g., browserbase-ui-test, browserbase-browser) using a script named scripts/install-external-skills.py. This installation script is not provided within the skill source files, making the source and safety of these external components unverifiable during this audit.

frontend-e2e