Understanding Superpowers: Like gstack but a System Enforcing Engineering Discipline for AI Agents

It Is Not a Skill List, but a Process Constraint

Superpowers appears on the surface as a collection of SKILL.md entries such as brainstorming, writing-plans, subagent-driven-development, test-driven-development, systematic-debugging, verification-before-completion, using-git-worktrees, finishing-a-development-branch and various code‑review related skills.

If you treat it only as an "engineering skill index" you will underestimate its purpose. Its real focus is not to tell an agent “here are some good habits you may follow,” but to enforce a strict sequence: "you must not skip steps; you must do A, then B, then C." This is evident from the repository’s entry point using-superpowers, whose core attitude is:

As long as there is even a 1% chance that a skill applies, you must check the skill before acting.

This rule targets the most common failure mode of agents: jumping straight into implementation without clarification, design, or validation. Many AI‑coding failures stem not from the model’s inability to write code, but from its tendency to act before the problem is fully understood.

Comparing Superpowers with gstack

Placing Superpowers side‑by‑side with gstack clarifies their different emphases. gstack leans toward "delivery division"—assigning specialists to stages such as requirements, review, QA, ship, and retro—while Superpowers concentrates on "engineering discipline." In other words, gstack expands the breadth of where agents can intervene; Superpowers tightens the guardrails around those interventions.

Superpowers enforces constraints at several critical nodes:

Requirements are not solidified; you cannot start implementation.

No executable plan; you cannot write code.

No failing test; you cannot claim TDD.

No root‑cause analysis; you cannot fix a bug.

No latest verification command; you cannot declare completion.

No review gate; you cannot advance the workflow.

Core Workflow: Constrain First, Then Implement

The main line of Superpowers can be expressed as:

using-superpowers → brainstorming → writing-plans → execution → review → verification → finish

In plain language, the steps are:

Determine which process applies to the current task.

Think through what needs to be done.

Translate the design into a detailed execution plan.

Begin implementation.

Review after each segment.

Re‑verify before claiming completion.

Finally decide whether to merge, open a PR, or keep the branch.

The system is organized around preventing loss of control, not around accelerating implementation.

Why the Emphasis on Early Discipline?

Most Vibe‑coding discussions focus on speed, code volume, or hands‑off operation. Superpowers reminds us that the real challenge is maintaining stability: "Running fast is not the hardest part; staying on track without deviation is."

1. AI’s Biggest Problem Is Starting Too Early

Agents often begin work before fully understanding the context, asking questions, comparing alternatives, or obtaining design approval. Superpowers combats this by making the "brainstorming" stage mandatory.

2. Vague Plans Invite Agent Drift

The writing-plans skill demands concrete, test‑driven plans:

Write a failing test.

Run the test to confirm failure.

Write the minimal implementation.

Run the test to confirm success.

Commit the change.

Each step must reference real code, file paths, and commands.

3. Many Engineering Incidents Come From Skipping Verification

The verification-before-completion skill codifies the principle: "Without fresh verification evidence, you cannot claim completion." Real development teams often prematurely announce success under assumptions such as “it should be fixed,” “the logic looks right,” or “the agent says it’s done.” Superpowers isolates this “verification black hole” and forces an evidence chain.

4. Old‑Fashioned Engineering Habits as Default Agent Behaviors

Superpowers does not invent new principles; it encodes classic habits as defaults for agents:

Design before implementation.

Failing test before code.

Root‑cause analysis before bug fix.

Verification before declaring done.

Isolate worktree before touching code.

Code review before proceeding.

These habits become mandatory workflow nodes, ensuring that an AI agent cannot assume it naturally possesses the engineering discipline humans cultivate.

Bottom Line

Superpowers is not a flashy repo that makes agents more capable; it is a repo that makes agents behave like reliable engineers by hard‑coding the engineering process—preventing premature actions, enforcing detailed planning, systematic testing, and rigorous verification.

Future posts will dive deeper into why brainstorming is placed first, how writing-plans achieves the required granularity, and the rationale behind the strong push for TDD, systematic debugging, verification, and review gates.