The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill directs the agent to search for and read sensitive local files, including .env, .env.local, and database fixtures, specifically searching for strings like 'user', 'pass', 'email', and 'login' to harvest credentials for testing.
[COMMAND_EXECUTION]: The agent is instructed to execute arbitrary commands discovered within repository configuration files such as package.json, Makefile, and docker-compose.yml, which can lead to code execution from untrusted sources within a repository.
[EXTERNAL_DOWNLOADS]: The skill automatically initiates the download and installation of browser binaries using mcp__playwright__browser_install or npx playwright install chromium if they are not already present.
[PROMPT_INJECTION]: The skill exposes an indirect prompt injection surface because it reads and follows instructions from untrusted repository files (like VERIFICATION.md or other documentation) and then executes commands or browser interactions based on those instructions.
[COMMAND_EXECUTION]: The skill includes instructions to dynamically generate and execute Node.js scripts from the repository root to drive browser automation if the primary browser tools are unavailable.

exerciser