Maintainability sensors for coding agents

A practical exploration of using computational sensors to maintain code quality when working with AI coding agents. Focuses on ESLint configuration for catching common AI failure modes like excessive function arguments, file length, cyclomatic complexity, and function length. Introduces the concept of custom ESLint formatters with self-correction guidance injected into lint messages, allowing agents to make judgment calls on warnings rather than blindly suppressing or complying. Also covers a broader harness of sensors including type checkers, Semgrep, dependency-cruiser, mutation testing, and periodic inferential reviews. Key insight: the cost-benefit balance of static analysis has shifted with AI — cheaper to create custom rules, and more valuable since agents produce hygiene issues humans rarely would. Warns about feedback overload and false sense of security from over-relying on linters.