AI Test Platform — Testing skill for Claude Code

AI Test Execution Platform

An AI-driven test execution platform where QA engineers save structured test cases, organized by project, and an AI agent autonomously executes them against a real web application — reasoning through each step, verifying outcomes at the database level, and reporting a structured, trustworthy verdict.

This is built specifically around the hardest, least-solved problem in AI-driven testing: **reliability**. Most tools in this space focus on generating tests from natural language; this project focuses on making sure the AI's actions and conclusions can actually be trusted.

Built to run against [HrManagement.Api](https://github.com/jiva19/HrManagement.Api) and [HrManagement.Frontend](https://github.com/jiva19/HrManagement.Frontend), a full-stack HR management app built as its target.

Core Reliability Design

• DOM-grounded execution — the agent never guesses at selectors. A read_page tool lets it inspect the real page for stable data-testid attributes before acting, added specifically after an early failure where a guessed CSS selector matched the wrong element.
• Independent database verification — a constrained, read-only query_database tool lets the agent confirm outcomes directly against the database, rather than trusting only what the UI displays.
• Structured, three-state verdicts — every run resolves to PASS, FAIL, or BLOCKED, so the agent correctly distinguishes a genuine application defect from a test whose precondition wasn't met.
• Evidence capture — screenshots organized per test case, per run, with the result embedded in the folder name for at-a-glance review.
• Safety controls — a hard cap on agent loop iterations prevents a stuck run from executing indefinitely, and strict spend limits are enforced on the underlying API usage.

Features

• Organize test cases by project
• Structured test case authoring: name, preconditions, steps, and expected result
• Run any saved test case on demand