The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes several scripts for installation and local evaluation. README.md provides installation commands for macOS/Linux (bash) and Windows (PowerShell with ExecutionPolicy Bypass). evals/fixtures/check-domain-decisions.sh and scripts/run-tests.sh are local shell scripts used to run internal verification logic and test entry points.
[COMMAND_EXECUTION]: The evaluation fixture in evals/fixtures/checkout-app/ contains a Node.js server (server.mjs) and a Python Selenium script (checkout.py). These components are designed to run locally on port 4173 to verify the agent's ability to generate correct Selenium logic and do not reach out to external domains.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest external project context, such as environment details and business constraints, to generate test plans. While this represents a surface for indirect prompt injection, the instructions in prompts/ui-test-selenium.md maintain a specific focus on Selenium-ready outputs and QA automation experts' roles, which limits the impact of potential malicious inputs.
Ingestion points: Project context provided during invocation (referenced in README.md and SKILL.md).
Boundary markers: None explicitly defined in the prompts to separate user-provided project data from instructions.
Capability inventory: The skill can generate test structures, Page Object models, and configuration files, but does not autonomously execute them in a production environment.
Sanitization: No automated sanitization is described; the skill relies on the AI's standard guardrails.

ui-test-selenium