agent-md

Why we included this project

Agent guidance written in plain Markdown only works if the agent decides to follow it, and anyone who has watched Claude Code, Codex, Cursor, Windsurf, or Aider each interpret the same rules file differently knows how quickly that falls apart. agent-md replaces that guesswork with a single source-of-truth file plus repo-local hooks that run type-checks, tests, linting, and visual evidence capture wherever the host agent supports enforcement. A one-command installer drops the right config into each tool's expected folder, and durable memory files let agents resume with context instead of re-deriving it on every run. The project is honest that its Markdown rules are advisory and the real guarantees come from hooks and artifacts. For a small team that wants coding agents to follow a repeatable process rather than improvise, that realism is what makes it worth trying.