A simple mental model for choosing between Claude Code's built-in plan mode and the superpowers workflow
When I started using Claude Code, I used plan mode for everything. Then I discovered the superpowers workflow and used that for everything.
Both were wrong.
The truth is simpler. These are different tools for different situations. Once you know the signals, picking the right one takes about two seconds.
Before diving into when to use each, let's clarify what we're talking about.
Plan mode is built into Claude Code. Hit Shift+Tab twice or run /plan to enter it.
When you're in plan mode, Claude operates in read-only. It can explore your codebase, search files, read code, and run research agents. But it can't edit files or run commands until you approve the plan and exit.
This creates a natural two-phase workflow. First Claude researches and asks questions to understand what you need. Then it presents a plan. You review it, give feedback, and when you're ready, Claude executes.
Plans get saved to ~/.claude/plans/ so you can reference them later. But they're separate from your codebase.
The key thing plan mode gives you is a permission boundary. Claude can't accidentally change anything while it's figuring out what to do.
Superpowers is a skill-based system built by Jesse Vincent that adds a different kind of structure. It has three distinct phases.
Brainstorming → Writing Plans → Executing Plans
The brainstorming phase is explicitly Socratic. Claude asks you questions one at a time, forcing you to articulate requirements before any planning happens. It's not just "what do you want to build" but a methodical exploration of constraints, trade-offs, and edge cases.
Plans get written to docs/plans/ as permanent documentation inside your codebase. Six months later, you can read 2025-01-15-auth-refactor-design.md and understand why decisions were made, not just what the code does.
You can also spin up git worktrees for isolation, launch parallel agents to explore different parts of the codebase simultaneously, and work with structured checkpoints.
More overhead, but more structure.
Both modes help you think before you code. Both involve Claude asking questions. So what's actually different?
Plan mode is about permissions. It's a safety mechanism that separates research from execution. Claude can explore freely without risk of changing anything until you're ready.
Superpowers is about process. It's a structured workflow that produces artifacts. The Socratic brainstorming is methodical. The plans become permanent documentation in your repo.
Think of it this way. Plan mode is like putting on safety glasses before you start working. Superpowers is like following a detailed checklist that documents everything you do.
Here's the mental model that works for me.
The quick test. Is this a task where you just need to be careful? Plan mode. Is this a task where you need to think through the design and want documentation? Superpowers.
Plan mode is your daily driver for most work in codebases you understand.
Use it when
What you get
Plan mode works best when you have context already. If you've been working in this codebase for weeks, you probably know the architecture. You just want Claude to be careful while implementing.
Superpowers earns its overhead when you need to figure things out and document the thinking.
Use it when
What you get
docs/plans/ that become part of your codebasePay attention to Claude's behavior during the conversation. Is it losing track of things you've discussed? Asking repetitive questions? Missing connections between parts of the codebase?
That's context pressure. The structured phases in superpowers help manage sprawling context by breaking work into documented stages instead of one long conversation.
File count or lines of code. These are arbitrary. File sizes vary wildly. I'm not counting files before deciding which workflow to use.
Risk level. Both approaches help you be careful. Plan mode through permissions, superpowers through process.
| Aspect | Plan Mode | Superpowers |
|---|---|---|
| Core idea | Permission boundary | Structured process |
| Overhead | Low, quick to start | Higher. Brainstorming, docs, worktrees |
| Best when | You need to be careful | You need to think it through |
| Artifacts | Plans in ~/.claude/plans/ | Plans in your repo at docs/plans/ |
| Questions | Ad-hoc as needed | Methodical Socratic method |
| Codebase fit | Any size, clear tasks | Complex systems, fuzzy requirements |
| Documentation | Separate from codebase | Built into codebase |
Neither is "better". It's about matching the tool to the task.
Using superpowers for a one-line fix is overkill. Using plan mode when you haven't figured out what you're building yet means you'll probably need to redo the plan.
Plan mode. You know what you want. You just want Claude to be careful while doing it.
Superpowers. You need to figure out what you want. You want the thinking documented.
This becomes intuitive with practice. The signals start to register automatically. You stop thinking about which mode to use. You just know.
Try this. On your next few tasks, consciously pick one approach and pay attention. When does it feel like overkill? When does it feel insufficient? That's your tipping point.
Find your own threshold, and stop overthinking it.
Superpowers is an open source project by Jesse Vincent. If you're using Claude Code seriously, it's worth checking out.