AI outputs keep getting better, yet I kept losing the process that produced a good result. A great session can yield a plan, a draft, a decision, even a prototype. A week later, I can’t reconstruct the exact steps, tools, checks, and judgment calls that made it work.
Most “workflow sharing” online still collapses into prompt sharing. Prompts copy well, but they rarely carry: - human judgment calls and handoffs - tool hookups and intermediate artifacts - quality gates and constraints that prevent confident wrongness
I built Epismo Skills to make workflows the unit of reuse. A “skill” packages an explicit process (human vs agent steps, expected artifacts, checks), and stays runnable via MCP so it can plug into the agent environment you already use.
What this enables for the community: - Share best practices as executable workflows: inspect, fork, adapt, and publish improved variants back - Import and adapt a workflow to a new project instead of reinventing it from scratch - Turn existing knowledge (project context or an external doc/link) into a reusable workflow, then publish it
Repo (MIT): https://github.com/epismoai/skills Deeper write-up: https://hiroki.blog/posts/everything-your-agent-needs-to-run...
What workflow would you want in a community library that you could import and run today?