I am developing skills and workflows and really only want to write them once.
Not one time per IDE - Looking at some of the github downloads, they often have a version per IDE or an installer to target an IDE specific install. Even with AntiGravity and Cursor they have their own directory structures and locations and IDE specific filenames. Some docs say write in .md, others .mdc and others yaml. I have found some that say “/command” is auto-discovered and found from skills and others say from workflows. Then there is the overhead of using plain text that needs to be read, parsed and ingested/digested.
Would you help me out with information and reference to docs that enable writing once and having them work in any IDE or agent?
That’s a bit broader than what this forum covers, since a true “write once, run anywhere” standard for AI skills and workflows doesn’t really exist yet across IDEs.
Each tool (Cursor, Windsurf, Copilot, etc.) has made its own choices about formats and discovery conventions, and I’m not aware of a universal specification that bridges them (at the moment, anyway). Things are evolving fast, so that may change soon!
Speaking for the Cursor side of things:
Cursor uses .mdc files stored in .cursor/rules/ for rules and skills.
If your goal is portability, some people write the core logic in plain markdown files and maintain thin IDE-specific wrappers. But that’s a design tradeoff, not a supported standard.
If you want “write once, use anywhere”, the simplest option in Cursor right now is to put your reusable logic into Agent Skills as SKILL.md folders under .agents/skills/ at the project root, and treat that as your single source of truth. Cursor auto-discovers skills from there (and the other supported */skills paths) using the Agent Skills standard, so those same skills can also be used by other tools that support Agent Skills.
AGENTS.md remains as the other option for rules that can be shared.