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. For details on individual stages, see the sub-pages: complete pipeline (6.1), brainstorming (6.2), git worktrees (6.3), writing plans (6.4), subagent-driven development (6.5), batch execution (6.6), code review (6.7), and branch finishing (6.8).
For information on how skills are loaded and invoked at a technical level, see Skills Discovery and Resolution and Finding and Invoking Skills.
Superpowers does not suggest workflows — it enforces them. The using-superpowers meta-skill (skills/using-superpowers/SKILL.md1-96) requires that agents check for applicable skills before any response or action. This means the workflow stages described here are mandatory gates, not optional steps.
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, and one before implementation execution begins.
Each stage is implemented as a named skill. The skill name maps directly to a directory under skills/ containing a SKILL.md file.
| Stage | Skill Name | Skill File | Human Gate? |
|---|---|---|---|
| Design exploration | brainstorming | skills/brainstorming/SKILL.md | ✅ Approve design |
| Isolated workspace | using-git-worktrees | skills/using-git-worktrees/SKILL.md | — |
| Task breakdown | writing-plans | skills/writing-plans/SKILL.md | ✅ Approve plan |
| Subagent execution | subagent-driven-development | skills/subagent-driven-development/SKILL.md | — |
| Batch execution (alt) | executing-plans | skills/executing-plans/SKILL.md | Between batches |
| Per-task testing | test-driven-development | skills/test-driven-development/SKILL.md | — |
| Inter-task review | requesting-code-review | skills/requesting-code-review/SKILL.md | — |
| Branch completion | finishing-a-development-branch | skills/finishing-a-development-branch/SKILL.md | ✅ Choose outcome |
Sources: README.md80-93 skills/brainstorming/SKILL.md1-4
Workflow Pipeline: Skill Sequence
Sources: README.md80-93 skills/brainstorming/SKILL.md24-55
The pipeline contains two mandatory human approval gates that prevent forward progress without explicit confirmation.
Design approval gate — enforced by brainstorming. The skill contains a <HARD-GATE> directive skills/brainstorming/SKILL.md14-16 that explicitly forbids invoking any implementation skill, writing any code, or taking any implementation action until the user has approved the design.
Plan approval gate — enforced by writing-plans. After the implementation plan is written to docs/plans/, execution does not begin automatically. The agent waits for explicit direction from the user to proceed with either subagent-driven-development or executing-plans.
Three skills have corresponding slash commands defined in commands/ that can be invoked directly by the user:
Each command file sets disable-model-invocation: true so that the command handler delegates entirely to the skill rather than generating a model response directly. See commands/brainstorm.md1-7 commands/write-plan.md1-7 commands/execute-plan.md1-7
Sources: commands/brainstorm.md1-7 commands/write-plan.md1-7 commands/execute-plan.md1-7
Process skills govern the overall approach and must be invoked before implementation skills. The brainstorming and writing-plans skills are process skills.
Implementation skills guide execution within the approach set by a process skill. test-driven-development and requesting-code-review are implementation skills that run inside each task cycle.
The using-superpowers skill skills/using-superpowers/SKILL.md79-82 specifies this priority ordering explicitly:
- Process skills first (brainstorming, debugging) — these determine HOW to approach the task
- Implementation skills second — these guide execution
After a plan is approved, the agent selects between two execution modes:
| Mode | Skill | Behavior |
|---|---|---|
subagent-driven-development | skills/subagent-driven-development/SKILL.md | Dispatches a fresh subagent per task; each subagent runs an implementer, a spec-reviewer, and a code-quality-reviewer in sequence |
executing-plans | skills/executing-plans/SKILL.md | Runs tasks in batches of 3 (default) within the same agent session; human review occurs between batches |
Both modes run test-driven-development and requesting-code-review as inner loops per task.
Sources: README.md84-86
The using-superpowers skill is the entry point for every session. Its protocol (skills/using-superpowers/SKILL.md24-54) specifies that before any response — including clarifying questions — the agent must check whether a skill applies. This is what causes the workflow pipeline to activate automatically without the user needing to invoke it explicitly.
For a full description of the mandatory skill check protocol, see The Mandatory Skill Check Protocol.
For a full description of the complete end-to-end workflow including the task loop internals and the two-stage review gate, see Complete Workflow Pipeline.
Refresh this wiki
This wiki was recently refreshed. Please wait 3 days to refresh again.