What problem does it solve? Standard build, lint, and test checks can pass while real behavior is still broken. UltraQA closes that gap by forcing adversarial dynamic end-to-end verification—malformed input, prompt injection, interruptions, stale state, flaky tests, and misleading success output—before declaring a goal met. ## Core Features & Use Cases - Adversarial Scenario Matrix: Tracks normal, hostile, malformed, interruption, injection, cancel/resume, stale-state, dirty-worktree, hung-command, flaky, and misleading-output scenarios with evidence and cleanup status. - Bounded Test-Diagnose-Fix Cycles: Repeats up to 5 cycles of plan, baseline verification, adversarial e2e execution, architect diagnosis, targeted fixes, and cleanup, with early exit on repeated failures or safety boundaries. - Temporary Harness Guardrails: Generates throwaway test harnesses with absolute imports, safe file writers, sanitized environment variables, and mandatory cleanup so harness debris never pollutes the worktree. - Use Case: After changing a CLI's resume logic, run UltraQA to verify that stale state files, repeated continue commands, and a CLI printing SUCCESS while exiting 1 are all caught, fixed, and reported with evidence. ## Quick Start Ask the agent to run UltraQA against your test suite with a goal such as verifying that the changed behavior survives hostile inputs, interruptions, and misleading success output.
