What problem does it solve?
The harness-init skill eliminates fragile, one-shot agent workflows by bootstrapping a persistent project harness that preserves intent, enforces quality gates, and restores context across multi-session AI development.
Core Features & Use Cases
- Scaffolding: Creates a .harness directory with harness.json, features.json, context_summary.md, claude-progress log, and an init.sh tuned to the detected stack.
- Quality gates & hooks: Installs PreToolUse, PostToolUse, TaskCompleted, TeammateIdle, and PostCompact hooks under .claude to enforce scope, git identity, and automated test verification.
- Team & task coordination: Suggests Agent Teams structures, proposes initial features with scopes and dependencies, and records git identity to prevent push/pull mistakes.
- Use Case: When starting a long-running project with Claude Code, run this initializer to enforce TDD, prevent context loss, and automate sanity checks each session.
Quick Start
Run the harness initializer, confirm the detected git identity and stack, and approve installing the .claude hooks to create the project scaffolding.