This page provides an overview of the structured development pipeline that Superpowers enforces through its skill system. It describes the sequence of skills that activate during a development session and how they connect together.
Superpowers is built on a set of composable "skills" and initial instructions that ensure your agent uses them README.md3-5 The system enforces a step-back approach: before writing code, the agent teases out a spec, presents it in digestible chunks, creates an implementation plan, and then launches a subagent-driven development process README.md18-24
For details on individual stages, see the sub-pages:
The full pipeline covers a project from initial idea to merged branch. There are two explicit human approval gates: one after the design is presented in brainstorming, and one before implementation execution begins after writing-plans skills/brainstorming/SKILL.md12-14 skills/writing-plans/SKILL.md158-166
| Stage | Skill Name | Human Gate? | Purpose |
|---|---|---|---|
| Design exploration | brainstorming | ✅ Approve design | Refine rough ideas, explore alternatives, save design document. |
| Isolated workspace | using-git-worktrees | — | Create isolated branch, run setup, verify clean test baseline. |
| Task breakdown | writing-plans | ✅ Approve plan | Break work into 2-5 minute tasks with exact file paths and code. |
| Subagent execution | subagent-driven-development | — | Dispatch fresh subagent per task with two-stage review. |
| Batch execution (alt) | executing-plans | Between batches | Execute tasks in current session with checkpoints. |
| Inter-task review | requesting-code-review | — | Review against plan; report issues by severity. |
| Branch completion | finishing-a-development-branch | ✅ Choose outcome | Verify tests, merge/PR, and clean up worktree. |
Sources: README.md190-202 skills/brainstorming/SKILL.md20-33 skills/writing-plans/SKILL.md158-175
The following diagram bridges the high-level workflow concepts to the specific skill files and human interaction points.
Workflow Pipeline: Skill Sequence
Sources: README.md190-202 skills/brainstorming/SKILL.md34-59 skills/writing-plans/SKILL.md158-175
The pipeline contains two mandatory human approval gates that prevent forward progress without explicit confirmation.
1. Design approval gate — enforced by the brainstorming skill via a <HARD-GATE>. The agent is prohibited from jumping into code or implementation skills until the user has approved the presented design skills/brainstorming/SKILL.md12-14
2. Plan approval gate — enforced by writing-plans. Once the design is approved and a plan is written to docs/superpowers/plans/, the agent offers an execution choice. Execution only begins once the user selects an execution mode skills/writing-plans/SKILL.md158-166
After a plan is approved, the agent selects between two primary execution modes skills/writing-plans/SKILL.md160-166:
| Mode | Skill | Behavior |
|---|---|---|
| Subagent-Driven | subagent-driven-development | Dispatches a fresh subagent per task with a two-stage review (spec compliance, then code quality). Recommended for fast iteration skills/writing-plans/SKILL.md162 |
| Batch Execution | executing-plans | Executes tasks in the current session in batches with checkpoints for human review skills/writing-plans/SKILL.md164 |
Both modes are expected to follow the task granularity defined in the plan, which enforces a RED-GREEN-REFACTOR cycle: write failing test, verify failure, write minimal code to pass, verify pass, and commit skills/writing-plans/SKILL.md95-125 README.md198
The workflow emphasizes rigorous verification through automated and manual review processes.
.worktrees/) to ensure a clean baseline README.md192 .gitignore1requesting-code-review skill activates between tasks to review against the plan and report issues by severity README.md200finishing-a-development-branch skill verifies tests and handles cleanup of the worktree before merging or creating a PR README.md202Sources: README.md190-202 skills/writing-plans/SKILL.md38-53 skills/writing-plans/SKILL.md95-125