claude-code-harness

Claude Code Harness is a disciplined delivery loop for AI coding agents like Claude Code, offering bounded paths for Codex and OpenCode. It transforms raw agent work into a repeatable operating path: Spec, Plan, Work, Review, and Release. Harness uses a source-of-truth loop via 5 verb skills. It drafts specific contracts for user approval, implements only the approved slice with TDD via '/harness-work', and ensures independent verification. Its Go-native core prevents plan drifting, enforcing structured task execution and evidence packaging.

• Disciplined Agent Delivery Loop (Plan, Work, Review, Ship)
• Source-of-truth Contract Execution (spec.md & Plans.md)
• Independent Review & TDD Verification
• Multi-host Compatibility (Claude Code, Codex, OpenCode, Cursor)
• Go-native Guardrail Engine without Node.js dependency