DoneSpec – deterministic completion checks for AI coding agents (github.com)

DoneSpec, now officially launched as version 1.0.0, introduces a deterministic completion layer for AI coding agents, ensuring that tasks claimed as "done" are genuinely complete and validated against a defined contract. This innovative tool transforms the concept of task completion into a machine-checkable format through a local JSON file called done.json. This explicit verification process requires that all specified conditions—such as the presence of necessary files and the passing of tests—are met before tasks are officially considered complete, making it particularly significant for enhancing accountability in AI workflows. The core functionality of DoneSpec includes strict mode validations, allowing developers to enforce a rigorous standard for task completion that integrates seamlessly into CI pipelines and local workflows without dependency on cloud services or complex frameworks. With various check types available, including file existence and command execution, DoneSpec helps developers and teams maintain clarity on what "done" really means. Its emphasis on deterministic checks and repeatable outputs positions DoneSpec as a crucial tool for improving the reliability and transparency of automated coding processes within the AI/ML community.