Show HN: Mindlas – catch your coding agent drifting before the bad code lands (github.com)

Mindlas (Mind-Atlas) is a new tool designed to enhance the reliability of coding agents by actively monitoring their performance in real-time sessions. It identifies "drift," where the agent's output diverges from the expected results, particularly during lengthy coding sessions. By capturing session signals live, Mindlas can issue alerts before errors compound, providing concrete corrective actions that can be verified through before-and-after measurements. With a focus on local processing—ensuring no data leaves the machine—Mindlas operates seamlessly while ensuring coding accuracy and integrity. This tool is significant for the AI/ML community as it addresses a common issue faced by developers: the deterioration of context over extended interactions with coding agents, which often remain undetected until significant errors arise. Mindlas utilizes a deterministic approach without relying on complex models or network calls for corrections, making it efficient and reproducible. It includes multiple corrective measures, each paired with specific triggering gauges and provides detailed exportable scorecards to evaluate session performance. This structured framework not only boosts coding accuracy but also advocates for better practices in AI-assisted development, ultimately promoting more reliable software production.