Do we need the Superpowers plugin in OpenCode?
Lately, I've been working on our OpenCode setup to see what we can and should automate. We've been using OpenCode for a while now, so we have many agents, skills and plugins that need reviewing. One of those is the Superpowers plugin by Obra.
I did some searching for benchmarks and found this article:
[
Can a Plugin Improve Codex? Benchmarking the Superpowers Plugin
Many plugins claim that they can make agents dramatically better. Add the right plugin, MCP, skill pack, prompt, or workflow and suddenly…
norbert-laszlo.medium.com
](https://norbert-laszlo.medium.com/can-a-plugin-improve-codex-benchmarking-the-superpowers-plugin-05d020066565)Some quotes from the article:
More usage is not automatically bad. If your goal is to maximize a flat-rate subscription and get the agent to spend more effort on each task, extra tokens and runtime may be acceptable. But if you pay per use, that same overhead is harder to justify when the correctness gains are uncertain.
Since GitHub moved Copilot to usage-based billing, with costs tied to token usage, this is relevant for us.
This evaluation does not prove that Superpowers is useless. I would still give the plugin the benefit of the doubt for greenfield development, planning-heavy work, collaborative design, verification discipline, teaching agents a process, or long-running feature development where extra structure is worth the overhead. Those workflows are not what SWE-bench-style issue-resolution benchmarks are designed to test.
Our platform team does not do product engineering. We do a lot of debugging and small code changes to internal tools.
The fair conclusion is not that Superpowers is bad. The fair conclusion is that it is not a silver bullet. It may prove useful in scenarios this benchmark does not cover, but it did not magically make Codex significantly better in the tested areas.
Let's face it: the basic harnesses are pretty good already. So in this case, we've chosen to switch off the plugin in our default setup.