codejunkie99/agent-harness ? reverse-engineered prompt

Build me a lightweight, file based workflow for AI coding agents that I can drop into any repo and use right away.

I want it to guide the whole process from writing a task, to making a plan, to implementing the work in small chunks, to testing it, and then writing release notes. It should work with common agents like Claude Code, Cursor, Codex, or anything else that can read and write files and run commands.

Set it up so there is a clear main instruction file, reusable skills for different project types like SaaS apps, CLI tools, MCP servers, and mobile apps, plus simple docs for task, plan, QA, and release notes. Include a changelog and a place for test fixtures too.

Make the workflow easy to follow, with guardrails so the agent stops after each chunk and waits for approval when needed. If you need to look up current best practices or docs online, do that.