orvixx-labs/orvix-provepr is a tool that actually verifies change claims before a coding agent says it has finished a task. The latest release is v0.3.0. When the agent writes a verification plan, provepr runs the checks listed in the plan via browsers, HTTP requests, and shell commands. The check results and evidence are passed back to Claude Code and Codex. This repository is written in Python and uses the MIT license.
provepr sorts each check into pass, fail, or unverified by fixed rules, and keeps screenshots, errors, and request results as evidence. It also provides checks that detect changes weakening the plan, as well as continuous integration (CI) integration. The way this tool operates is by not calling any separate language model. So instead of merely re-evaluating the agent's statements, the focus is on a flow that verifies what the plan lists using actual browsers and commands.
However, the provided FACTS do not list anything newly added or changed in v0.3.0, so the specific changes in this release cannot be confirmed. The check results should also be read against the claims included in the plan and the evidence obtained in actual execution. Items classified as unverified must be distinguished from confirmed results, and the scope of this tool is running the checks written in the plan and returning the evidence.