Hey Product Hunt, I built Harness Starter Kit because I kept seeing the same problem with AI coding agents: every session starts smart, but the important project rules disappear when the chat ends. The idea behind this project is simple: stop putting all the critical context in one-off prompts. Put it in the repo. Harness Starter Kit helps teams add durable agent instructions, lightweight drift checks, failure memory, decision records, adoption reports, and stack-specific harness snippets to real repositories. It is intentionally prompt-first, not installer-first. The target repo stays the source of truth, and the kit pushes agents to inspect, adapt, and verify instead of blindly copying defaults. I’d love feedback from people using Cursor, Claude Code, Codex, Copilot, or other coding agents in production repos: What recurring agent mistakes would you want your repo to remember and prevent?

Harness Starter Kit

Repo guardrails for reliable AI coding agents

Repo guardrails for reliable AI coding agents