what is cortex fr?

CORTEX is a JavaScript tool that acts as a supervisor for automation pipelines. The project's core argument is that most automation setups degrade silently. A scheduled job finishes successfully, a dashboard shows green, yet nothing actually improves or learns. The tool is built to fix this by checking that each step produced real, verifiable output rather than just confirming the process did not crash. The system works in three layers. First, a preflight scans an existing code repository to detect systems and rank optimization opportunities. This step is read-only and produces a report for a human to approve. Second, a supervisor runs the learning and optimization steps. It verifies that each step generated new output, retries silent failures, and sounds an alarm if a step exits cleanly but produced nothing. Third, an optimizer applies the approved improvements, with every change gated, reversible, and logged. CORTEX is structured as a host with a module registry. Each module handles a specific capability and follows a common contract so the supervisor can schedule it. Modules include reinforcement learning, recursive learning, a flywheel for surfacing opportunities, a distiller for compressing learnings into reusable rules, an adoption module that ships learnings into the live pipeline, and a sentience layer that updates the self-model. The project tracks an adoption rate, measuring how many harvested learnings were actually applied, because a system that harvests but never applies looks healthy while nothing improves. A key technical detail is its locking mechanism. Instead of checking only how old a lock is, CORTEX checks whether the process holding the lock is actually alive. This prevents a crashed run from blocking all future work. The project is designed to be free to run, using whatever models or routers you already have. The full README is longer than what was shown.