opengap

Why we included this project

Most agent frameworks each want their own definition format, so teams supporting several of them end up writing the same agent more than once. OpenGAP sidesteps that by letting you define an agent as plain files inside a git repo: a manifest, an identity document, rules, skills, and tools. One definition then exports to Claude Code, OpenAI, LangChain, CrewAI, or AutoGen through adapters, and because everything lives in git, every change is a commit you can diff, roll back, review in a pull request, and promote across environments like any other code. That also makes it practical for regulated settings, since the spec has first-class support for segregation of duties and audit trails, and the CLI's validator flags role conflicts and compliance problems before you deploy. If you want an agent description that stays portable no matter which framework wins, the standard plus its reference CLI is a solid fit.