testsprite/testsprite-cli

TestSprite is a cloud-based testing platform, and this repository is its official command-line tool. The core problem it addresses is that AI coding tools (agents that write code automatically) are fast at generating features but cannot verify their own work against a real running application. TestSprite closes that loop: the agent ships code, then calls this CLI to open the live app, click through it like a real user, and get back a structured report of exactly what broke. The agent can then read that report and fix its own mistakes before a human sees them. The workflow the README describes has three steps. First, the agent creates a test run by pointing the CLI at a project and a plan file that describes the behavior to check. Second, if the run fails, the agent downloads a single "failure bundle", a self-contained file that describes what went wrong, including screenshots and DOM state. Third, the agent makes code changes and replays the test to confirm the fix passed. Passing tests stay in a durable suite for future runs. Installation requires Node.js version 20 or higher. The init command handles authentication, confirms the API key works, and installs a small skill file that teaches the agent (Claude, Cursor, Codex, and others are named) how to call TestSprite in its workflow. For automated pipelines, environment variables replace the interactive prompts. The CLI covers a wide command surface: listing projects and tests, fetching individual test results and their run history, downloading generated test code, replacing that code manually, running tests, and deleting them. The README includes a note about a public leaderboard where models were compared while using this verification loop, and the result was that the cheapest model ranked first by correctness, which the project uses to argue that verification quality matters more than raw model capability. The license is Apache 2.0. A free-tier API key is available at testsprite.com to get started.