The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is instructed to identify and execute arbitrary build, start, and test commands found within the target repository (e.g., Playwright, Cypress, npm scripts). This is a core part of its E2E verification purpose but involves executing potentially untrusted code from the project environment.
[DYNAMIC_EXECUTION]: The skill generates new verification logic in the form of scripts and agent skills (e2e-<area>) and is explicitly directed to run this generated code (WORKFLOW @prove_before_handoff) to verify its effectiveness. This constitutes the execution of dynamically created code.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the repository, including requirement documents, public contracts, and commit messages, to derive its verification criteria and logic. This presents an attack surface where malicious instructions in project documentation could influence the generated verification steps or findings.
[DATA_EXPOSURE]: The skill's workflow involves capturing "evidence" such as logs, screenshots, and command outputs. Although the instructions include guidelines to avoid sensitive data and use secure sources for credentials, the broad access required for E2E testing creates a risk of sensitive information being captured in test artifacts.

create-e2e-verification-skill