The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local commands for environment checks (e.g., nc, pg_isready), application startup (via project-defined npm or nx scripts), and testing (e.g., curl, npx playwright). These actions are properly scoped to the project environment and are initiated based on the user's explicit request to test changes.
[PRIVILEGE_ESCALATION]: Analysis shows no attempts to use sudo or bypass system permissions. Commands for starting apps use perl with setsid to manage process groups for clean teardown, which is a standard technical approach for background processes.
[CREDENTIALS_UNSAFE]: The skill follows safe handling of secrets. It specifically instructs the agent to read tokens from .env files using grep/cut while explicitly forbidding the printing of these secrets to the console or logs.
[PROMPT_INJECTION]: The skill uses natural instructional language. While it uses terms like 'Hard rules', these are internal constraints for the agent to ensure high-quality testing and do not attempt to bypass platform safety guidelines.
[DATA_EXFILTRATION]: Network operations are restricted to testing the local application under development (via curl and Playwright). There are no patterns suggesting data is being sent to external, unauthorized third-party domains.

testing-changes