๐ข๐ฝ๐ฒ๐ป-๐ฆ๐ผ๐๐ฟ๐ฐ๐ฒ ๐๐ฒ๐๐ ๐ก๐ฒ๐ฒ๐ฑ ๐ง๐ผ ๐ฆ๐ต๐ถ๐ฝ ๐๐ถ๐๐๐ฟ๐ถ๐ฏ๐๐๐ถ๐ผ๐ป
Publishing a package is not the finish line.
Developers use AI agents now. Tools like Cursor and Claude Code search for skills. They run shell commands. They read manifests.
If you only have a library, you are invisible. You need a distribution layer. This is the wiring around your code. It lets agents find and use your tool. You do not want humans to paste your README into a chat.
Build these four things:
- A CLI with clear flags.
- Skill files like SKILL.md.
- Plugin bundles.
- Marketplace manifests.
A CLI is a stable API. Agents run it to see the output. Skill files tell agents when to use the tool. Marketplace files let the agent install it.
Do not build an MCP server for everything. MCP is for stateful sessions. Use it for things like a live browser session. For most tools, a CLI is better. It is simpler to debug.
Distribution is an interface. Treat it like your API. Test it with a fresh agent session. If the agent fails to set it up, your distribution is broken.
Stop treating agent support as a future task. Early adopters use agents today. If they do not find your tool, they use a famous one.
Source: https://dev.to/prithwish_nath/open-source-devs-need-to-ship-distribution-not-just-code-3j6a Optional learning community: https://t.me/GyaanSetuAi