token-harness

Token Harness is a local CLI tool designed to improve the efficiency and usability of coding agents like Claude Code and Codex. It runs entirely on-device, never sends data to external services, and avoids modifying configurations unless explicitly approved. The tool detects installed agents and optimization providers (like RTK or HarnessTrim), verifies their integration, and recommends safe, evidence-based improvements. Users start with `token-harness setup`, which diagnoses current state and suggests one next step — often requiring no change. After setup, the dashboard (`token-harness ui`) provides real-time status: active tools, observable token allowances, and actionable guidance. Optimizations run automatically through existing agent commands (e.g., `claude` or `codex`), so no special launcher is needed. Advanced features include metrics tracking, provider updates, rollback, and native policy application via plan/apply workflows. Safety is prioritized: changes are transactional, backed up, and require explicit `--yes` flags. Token Harness supports read-only inspection of usage history (ccusage) and quota companions (cclimits) but never installs them automatically. It’s ideal for developers seeking transparency, control, and minimal friction when working with AI coding assistants.