What problem does it solve? Code review and documentation alone cannot reveal how software actually behaves at runtime. This Skill structures investigations into real runtime behavior—edge cases, undocumented behavior, regressions, and failure modes—using planned probe scripts, validation matrices, and evidence-based reporting instead of guesswork. ## Core Features & Use Cases - Validation Matrix Planning: Design a case matrix covering baseline success, controls, boundary inputs, invalid inputs, transient failures, retries, and concurrency before running anything. - Disposable Probe Scripts: Generate temporary, observable Python probe scripts that capture runtime context such as git commit, Python version, and package versions, with support for repeat-N and warm-up execution modes. - Safe Live Probing: Apply destination, intent, and data gates before live API probes, including explicit approval for environment variables like OPENAI_API_KEY and cleanup or rollback plans for mutating cases. - Findings-First Reporting: Report unexpected or negative findings first, with scope, confidence, comparison parity notes, and artifact status. - Use Case: When you suspect a regression in the OpenAI Agents SDK after a change, use this Skill to run a controlled probe against a known-good baseline like origin/main, compare behavior across repeat runs, and produce an evidence-backed report. ## Quick Start Use the runtime-behavior-probe skill to verify whether streaming retries behave differently after the latest commit, with a control run against origin/main.
