Why reinvent what already exists, or what will be native tomorrow

I see many teams spend time and money copying or reproducing generic skills for their agents: reading a repository, opening a pull request, querying a database, summarising a ticket, browsing a documentation site. These capabilities already exist, as plugins, MCP servers or integrations maintained by someone else. Is that really the right investment?

Copying is the most telling case. You take an existing skill, paste it into your own repository to adjust it at the margin, and you become its owner without having chosen to: every fix to the original has to be ported by hand, every change in the underlying model or tool has to be tracked, and nobody on the team holds the knowledge that produced the first version.

You bought a maintenance debt for one evening of savings

There is a second reason to hold back. Part of these capabilities will probably end up built into the models themselves. What needs a dedicated skill, a careful prompt and three tools today will be a native capability tomorrow, called without any configuration. Code written to fill that gap does not only have a cost: it has an expiry date, and it is close.

What remains, and what deserves the investment, is what exists nowhere else: domain knowledge, company rules, the system’s invariants, the conventions the team has agreed on. A skill that encodes those will never ship as a plugin or be absorbed by a model, because nobody else knows them.

So the question to ask before writing a skill is not “can we do it?” but “are we the only ones who can?”.

If the answer is no, install it. If it is yes, build it, and keep it small