
Late last year I installed Superpowers in Claude Code, used it a few times and quit. I liked it. I may also have been holding it wrong, which is the least flattering reason to stop using anything.
Superpowers is a set of Markdown skills for coding agents from Jesse Vincent and the team at Prime Radiant. Install it and your agent loses the urge to start typing code the moment you finish a sentence. It asks what you actually want, writes a design, turns the design into a plan and works the plan with subagents, tests first. It started as a Claude Code plugin in October 2025.
The good part
Brainstorming before code. An agent that asks questions before it touches files finds out the request is wrong while that is still cheap.
Test-driven development with no bargaining. The README says the skill deletes code written before its tests. Harsh. I approve.
Written plans handed to subagents, with a review between tasks. Somebody checks the work before the next task starts, even when every somebody is a model.
And it’s all plain Markdown, so you can read every rule before it gets anywhere near your code.
The paperwork
Every design and every plan became a dated file in my repository. At the time, the brainstorming skill saved designs to docs/plans/ and committed them, and the planning skill filed plans right beside them. A few uses in, I had more documentation than I could keep track of. Progress lived in the session’s todo list, not in the files.
Nothing tied them to an issue tracker either. The plan knew what to build. Nothing outside the chat knew whether it had been built.
It kept going without me
Last reviewed: 2026-09-26.
Superpowers has moved on since I left. It now installs into more than a dozen agent tools. In March 2026, specs and plans moved into docs/superpowers/specs/ and docs/superpowers/plans/, and both skills now say your preference for where they go overrides the default. Its entry skill also ranks your instructions, in CLAUDE.md, AGENTS.md or a direct request, above its own. Whether that fixes my problem, I can’t say. I haven’t tried it.
I didn’t find a built-in connection to an issue tracker. For something that installs into that many agents, skipping one is a reasonable call.
Plans next to code, or next to status
My disagreement is narrow: where the plan lives. Superpowers keeps it next to the code. That’s a good home if your repository is your entire system of record. You work alone or in a small team, you don’t use a tracker, and you’d rather review everything in git than sign in to one more hosted thing. Hydrant needs a tracker account, and in that setup it’s weight you didn’t ask for.
We went the other way. Our workflow puts the scope, the plan and the evidence on the issue, next to its status. The repository keeps the code and the decisions that outlive a single task. Steal our workflow has the steps.
If you like Superpowers’ discipline and ended up with the same pile of files I did, point its plans somewhere you’ll actually look. If you haven’t tried it, read the brainstorming skill before you install anything.
Filed under PrinciplesHenry Kobutra with Claude
We post on X too. Mostly shipping, occasionally opinions. @usehydrant (opens in a new tab)