mysuperpowers

mysuperpowers

A structured development workflow plugin for Claude Code.

Introduction

mysuperpowers is a personal development workflow plugin for Claude Code, built on the skills framework pattern from obra/superpowers. It installs a set of skills — structured process guides — that the agent follows automatically when the task matches, covering everything from initial brainstorming through final branch integration.

The main thing it does differently from stock Superpowers: planning and execution run in separate sessions by design. Brainstorming, PRD creation, and milestone planning happen in a single planning session. Each milestone then executes in its own fresh session, started by pasting a self-contained execution prompt from the plan file. This keeps the context window clean and the starting state deterministic. The plugin also adds full update and revision flows across all three planning skills, so when a feature changes mid-development, the design, PRD, and plan can all be updated without starting over.

The execution skills (TDD, debugging, code review, verification, branch completion) are adapted from obra/superpowers, which provided the skills framework pattern and the execution-phase discipline. The PRD creation skill draws on the closedloop-ai prd-creator for its discovery questions, ID conventions (US-###, AC-###, Q-###), and "keep it light" philosophy. Both are MIT licensed; see Credits below.

How it works

Skills are markdown files. Each one describes when to activate and how to proceed. The agent uses the Skill tool to load a skill's full content on demand and then follows it — checklists, decision flows, hard gates and all.

At session start, a SessionStart hook injects the using-mysuperpowers skill directly into the agent's context. This bootstraps the entire framework: the agent immediately knows what skills are available, when to invoke them, and that invocation is mandatory — not optional. When a skill applies to the current task, the agent must invoke it before doing anything else, including asking clarifying questions.

Skills can be rigid or flexible. Rigid skills (TDD, debugging) are followed exactly — no adapting away the discipline. Flexible skills (planning, patterns) apply principles to context. The skill itself tells you which it is.

Installation (development mode)

mysuperpowers is not on the Claude Code plugin marketplace. Install it as a local plugin in dev mode.

1. Clone or download this repository to a local directory.

2. Install it as a local plugin in Claude Code:

   /plugin install /path/to/mysuperpowers
   

   Note: The Claude Code plugin system is still evolving. Verify the exact command against the current Claude Code documentation if the above doesn't work.

3. Restart Claude Code.

4. Start a fresh session. The agent should acknowledge it has mysuperpowers in its first response — that's the SessionStart hook firing.

The Workflow

Development follows two distinct phases. They never run in the same session.

Planning phase

Run in a single session. The output is plan.md — the agent stops there.

1. Describe an idea or task. brainstorming activates in Create Mode.
2. The skill runs a Socratic design conversation: explores project context, asks clarifying questions one at a time, proposes 2–3 approaches with trade-offs, and presents a design for approval.
3. The approved design is saved to docs/features/<feature-name>/<feature-name>-design.md.
4. Brainstorming asks: "Would you like a PRD?" This step is optional.
5. If yes, creating-prd activates. It asks seven discovery questions conversationally, then drafts a 15-section document in two parts — business context (overview, goals, user stories, sequencing) and implementation detail (requirements, acceptance criteria, open questions, risks). Saved to docs/features/<feature-name>/prd.md.
6. creating-prd asks: "Would you like an implementation plan?" Also optional.
7. If yes, milestone-planning activates. It decomposes the work into milestones — each one scoped to be reviewable by a human in a single PR sitting. For each milestone it generates a self-contained execution prompt. The full plan is saved to docs/features/<feature-name>/plan.md.
8. The skill stops. It does not offer to begin execution. The user opens a new session for each milestone.

Each milestone runs in its own session because planning-phase context contaminates execution decisions. A session that starts mid-planning carries assumptions that muddle implementation choices. Starting clean means the agent reasons from the plan and PRD, not from the conversation that produced them.

Execution phase

One fresh session per milestone.