The developer created Crossweft, a tool designed to manage and verify consistency across multi-language software components. The system addresses the issue of 'seams'—points where different modules, such as a Go-based backend and a TypeScript client, must remain synchronized. These connections often fail because compilers cannot validate cross-language dependencies, leaving consistency reliant on human memory.
Development began in January, with over $9,000 spent on AI subscriptions to assist in coding. The author noted that AI agents, such as Claude Code and Codex, frequently struggle with inter-component consistency, often failing to update all sides of a dependency. Crossweft mitigates this by using JSON-based maps that anchor connections to specific code literals, ensuring that if code moves, the verification fails.
The tool employs simple, fast checks like regex, value comparisons, and set matching to ensure synchronization. It requires agents to attest to changes, preventing them from finishing tasks if seams remain misaligned. The system also includes a meta-runner for custom validators, ensuring that security rules actually scan files rather than silently passing due to misconfiguration.
Since implementing Crossweft, the developer has observed a shift in the nature of bugs. Initial implementation involved adding 185 join-guards and 107 set-guards, which identified 15 actual code discrepancies. The tool now supports 50 validators, up from 4 in June, and ensures that maps remain accurate by requiring verification at every agent stop.
