I open sourced Skillpress
It came out of a very practical irritation.
I’m working on a cluster of CLI tools that consume each other. Runlane needs Remogram. Skillpress needs to understand how those tools publish agent guidance. The whole thing only works if the agents operating inside the codebase are reading the right instructions.
At the same time, I’m developing in both Codex and Cursor. Each surface has its own place for guidance, its own format assumptions, and its own way of going stale.
That sounds like a small problem until you’re debugging behavior that came from an old skill file in one agent root while another agent is following the newer rule somewhere else.
That’s the kind of bug that doesn’t look like a bug at first. It looks like judgment drift.
Skillpress is my answer to that.
Canonical skill source lives in the repo.
Installed provider roots are caches. sync puts the right guidance where each agent can actually load it. status and doctor tell you when things have drifted. The manifest records what was installed, from where, and at what hash.
It supports Agent Skills directories, Codex skill roots, Claude Code skill roots, agent-local roots, and Cursor project rules.
This isn’t orchestration. It isn’t proof. It isn’t trying to become the center of the toolchain.
It’s plumbing for agent guidance.
But if the tools are going to consume each other, and the agents are going to help operate those tools, then the guidance layer has to be boring, explicit, and current.
That’s what Skillpress is for.
Top comments (0)