luchangqiu/lrnev-govern

lrnev is a project governance tool designed for workflows where AI assistants help write code. It provides two interfaces: a command-line tool and an MCP server that any MCP-compatible AI client can call. The goal is to give an AI assistant a persistent record of a project's requirements, design decisions, tasks, and known problems, so that opening a new conversation does not mean starting from scratch. When you initialize lrnev in a project, it creates a .lrnev directory containing plain Markdown files. These files hold the project description, architectural constraints, feature specifications (called Specs), task lists, architecture decision records, and an error log. Because everything is ordinary text, you can read it directly, commit it to git, and review it without any special tooling. A Spec is the central unit: a folder with three documents covering what to build (requirements with acceptance criteria), how to build it (design), and the list of individual tasks. The system maintains a state machine for each task, tracking whether it is pending, in progress, or completed. A gate check validates that a Spec's requirements are sufficiently complete before any implementation begins. The tools operate on these files directly without calling any language model. lrnev handles deterministic work: reading files, assigning IDs, updating task states, and checking document structure. When judgment is needed, for example whether a requirement is clear enough, it signals the AI with a follow-up prompt and leaves the decision to the AI and the user. The README describes this as "guide, not enforce" and notes that short one-off scripts do not need Specs at all. The tool is installable globally via npm and works with Claude Code, Cursor, and any other client that supports MCP or CLI access.