A Rave Review of Superpowers (For Claude Code)(emschwartz.me)
emschwartz.me
A Rave Review of Superpowers (For Claude Code)
https://emschwartz.me/a-rave-review-of-superpowers-for-claude-code/
29 comments
I've found [Plannotator]( https://github.com/backnotprop/plannotator) useful for this. It opens the plan on a minimal web page so I can highlight and add comments. This is then passed back to Claude code. Works well enough for now. But ideally this should be natively implemented in cc.
I extensively tried to use this and couldn't get it working. I'm running CC from a docker container and mostly use it from a web interface, but even setting it to just give a url, and using CC from the terminal, it would just not hook in correctly
>It's sure baffling how Anthropic has kept Claude Code's plan mode so linear and inflexible
It's difficult to know what the appropriate process for a model would be without widespread deployment. I can see how they have to strike a fine balance between keeping up with what the feedback shows would be best and changing the way the user interacts with the system. Often it's easy to tell what would be better once something is deployed, but if people are productively using the currently deployed system you always have to weigh the advantage of a new method against the cost of people having to adapt. It is rare to make something universally better, and making things worse for users is bad.
It's difficult to know what the appropriate process for a model would be without widespread deployment. I can see how they have to strike a fine balance between keeping up with what the feedback shows would be best and changing the way the user interacts with the system. Often it's easy to tell what would be better once something is deployed, but if people are productively using the currently deployed system you always have to weigh the advantage of a new method against the cost of people having to adapt. It is rare to make something universally better, and making things worse for users is bad.
In my experience the plan is plastic and responsive to iteration through dialogue. I don't think too much structure here is going to be desired by all users. But the "sounds good, but needs these edits" option can save a small amount of interaction. I often use the tab command that exists to modify the yes and no options. If you tab modify "Yes", I believe the plan will be executed and addendums from your additional prompting will be made afterward. By using tab modified "No", you can be clear about what parts of the plan are to be kept and what is to be immediately adjusted. Tab modified "No", definitely works as the "sounds good" option, but perhaps requires slightly more typing than you would like. "No, use most of this, but change..." I don't think there is a glaring lack here at all.
There's also:
4. Tell claude what to do instead, which will update the plan base on what you say.
5. Add comments to the plan directly - similar to 4 - but you can comment on specific parts.
Note: I use the VSCode extension, not sure if it differs in terminal mode.
4. Tell claude what to do instead, which will update the plan base on what you say.
5. Add comments to the plan directly - similar to 4 - but you can comment on specific parts.
Note: I use the VSCode extension, not sure if it differs in terminal mode.
Maybe they've updated it in the last couple weeks and my comment is out of date - I hope so! Because the 5th option definitely wasn't there when I recently used it, and the 4th you're mentioning looks familiar but I'm pretty sure in that case to it didn't (used to?) write it to a file.
Will try it again.
Will try it again.
The VSCode plugin now (since last week I think) allows you to highlight elements of the plan and add comments. I think it first appeared in Antigravity but now is in Claude code. Unfortunately there is only so much you can do in CLI.
It is also possible to now save plans in a custom location (I have started using ./docs/plans) although the plan mode still frustratingly names plans random file names and struggles with more than. 1 plan file in a session.
But the interface has become much better if you're on vscode
It is also possible to now save plans in a custom location (I have started using ./docs/plans) although the plan mode still frustratingly names plans random file names and struggles with more than. 1 plan file in a session.
But the interface has become much better if you're on vscode
Plans are written to the .Claude folder in your home directory as you discuss them. Just ask for the file name.
You can just tell it what to do. I have a cut and paste handy that lets the tool know to present all judgement calls to me a few at a time in logical groups to give feedback on. I go through that process and then it pulls the plan from that
I accidentally figured out you can go into plan mode then leave, Claude can keep working on the plan, then I use my own custom plan viewer (just a markdown viewer that's watching for the latest for update in the plan dir)
Online reviews indicate that Superpowers is best for people who are not already experienced SW development managers.
Is that true? What is your experience of it?
For me, I am a solid KISS believer, so I have not yet found a better framework than just plain old Claude Code. But happy to move to a better workflow, if it's real.
Is that true? What is your experience of it?
For me, I am a solid KISS believer, so I have not yet found a better framework than just plain old Claude Code. But happy to move to a better workflow, if it's real.
Overall I think it's useful.
Superpowers has several skills. Its core workflow is:
- Brainstorm with you to design a spec
- Use subagents to review its own spec, then get your approval
- Based on the spec, write a plan, use subagents to review before final approval
- Use subagents to implement (using TDD)
I think that the brainstorming skill [1] is great. It helps flesh out a rough early idea. I also like that it uses subagents to adversarially review its own spec/plan; that has caught several things I would've missed. I do not like the separation of spec/plan; IMO the models are good enough to get straight to coding once the spec is written. The plan often ends up being code blocks in a Markdown doc.
[1]: https://github.com/obra/superpowers/tree/main/skills/brainst...
Superpowers has several skills. Its core workflow is:
- Brainstorm with you to design a spec
- Use subagents to review its own spec, then get your approval
- Based on the spec, write a plan, use subagents to review before final approval
- Use subagents to implement (using TDD)
I think that the brainstorming skill [1] is great. It helps flesh out a rough early idea. I also like that it uses subagents to adversarially review its own spec/plan; that has caught several things I would've missed. I do not like the separation of spec/plan; IMO the models are good enough to get straight to coding once the spec is written. The plan often ends up being code blocks in a Markdown doc.
[1]: https://github.com/obra/superpowers/tree/main/skills/brainst...
A recent update got rid of the subagents for the self review and now does the self review in the main agent
Yeah I agree with the implementation plan not being that useful for me to read. I often just tell it to go after reading the design doc.
How is it better than https://github.com/gsd-build/gsd-2 ?
The install mechanism for the superpowers plugin for codex and opencode is .... interesting. From https://github.com/obra/superpowers
Fetch and follow instructions from https://raw.githubusercontent.com/obra/superpowers/refs/head...
it's like curl|bash but with added LLM agents...
Fetch and follow instructions from https://raw.githubusercontent.com/obra/superpowers/refs/head...
it's like curl|bash but with added LLM agents...
I personally don't like superpowers very much. My boss does.
I think Claude makes more mistakes when using superpowers than when not.
but maybe it's my fault.
I recommend trying. it doesn't hurt. Just don't believe it's a silver bullet. It's still the same Claude.
The way I use Claude is quite similar to what super powers does under the hood anyways. Like I always ask things like this: "if I want to do X, what questions do you need to ask me to have all the information you need to make it happen"
but maybe it's my fault.
I recommend trying. it doesn't hurt. Just don't believe it's a silver bullet. It's still the same Claude.
The way I use Claude is quite similar to what super powers does under the hood anyways. Like I always ask things like this: "if I want to do X, what questions do you need to ask me to have all the information you need to make it happen"
>The way I use Claude is quite similar to what super powers does under the hood anyways.
I bake my own bread and solo code, both episodic. No matter how I vary fresh starts, I always end up in the same place. The optimization problem feels to be a giant bowl.
This reminds me of named flavors of management style for teams of programmers. A friend in this role instead prefers deep dives into Apollo flight control, and abstracting.
As a solo programmer I always converge on staying extraordinarily involved in planning and code review, and working in steps of finer granularity than superpowers suggests. One may produce less code this way, but it's the only way I know to discover one needs less code.
I bake my own bread and solo code, both episodic. No matter how I vary fresh starts, I always end up in the same place. The optimization problem feels to be a giant bowl.
This reminds me of named flavors of management style for teams of programmers. A friend in this role instead prefers deep dives into Apollo flight control, and abstracting.
As a solo programmer I always converge on staying extraordinarily involved in planning and code review, and working in steps of finer granularity than superpowers suggests. One may produce less code this way, but it's the only way I know to discover one needs less code.
Is Superpowers skill only for coding? Can I use it to plan other stuff, like to create sales or marketing strategy?
all these orchestrating and giving skills etc. -- are they more or less being developed by trying things out? or is there a better way?
It's sure baffling how Anthropic has kept Claude Code's plan mode so linear and inflexible. It may ask a couple of questions before writing it, but there's always going to be parts that need editing. Yet there's no good "Sounds good, but it needs these edits" option after it presents you the plan. It gives you (paraphrased) "1. Proceed with auto-edits 2. Proceed without auto-edits 3. Cancel the plan". Note that 3 doesn't even write the plan to the file at all, even if it's 95% fine. So your options are either A. Pick 1 or 2, immediately press escape to interrupt, then tell it to make edits or B. Pick 3, tell it to make edits, after which it has to write the entire plan from scratch again.
This is such bad UX that it really feels like either 1. Anthropic employees don't use Claude Code much - this seems incredibly unlikely or 2. It's intended to burn output tokens as it has to write a huge plan again.
IMO Superpowers isn't the ideal solution because it too lacks flexibility, but including the "plan sketch" stage is sure an improvement.