Superpowers
Properties
created
16.07.2026, 11:03
modified
05.09.2026, 11:36
published
Empty
sources
claude.com/plugins/superpowers · obra/superpowers · Superpowers: How I’m using coding agents in October 2025
topics
Superpowers, Claude Code, Skills Framework, TDD
authors
Jesse Vincent
ai-assisted
Yes
- By Jesse Vincent (obra), October 2025
- obra/superpowers
- Positioning: complete software development methodology for coding agents — strict pipeline, activates automatically when building starts
- Philosophy: deep upfront reasoning, then hands-off execution
# The Six-Stage Pipeline
- Brainstorming — no jump to code; ask questions, explore alternatives, produce design doc
Teasing a spec out of the conversation.
- Worktree isolation — git worktree per task; parallel tasks cannot clobber each other
- Writing plans — design broken into 2–5 min tasks with exact file paths, complete code, verification steps
Clear enough for an enthusiastic junior engineer with poor taste, no judgement, no project context, and an aversion to testing to follow.
- Subagent-driven development — fresh subagent per task; task reviewer checks spec compliance + code quality before closing; critical issues block progress
- Test-driven development — strict RED-GREEN-REFACTOR; code before tests gets deleted
- Code review — between tasks, review against plan, report by severity
# Key Distinguishers
- Autonomy — designed for hand-off of large work chunks, return to reviewed result
- Subagent architecture — main agent manages workers, does not implement itself
- Composability — skills are building blocks; drop brainstorming subagent into custom skill, use worktree isolation without TDD enforcement
# Trade-off
- Full pipeline feels heavy for one-line bug fixes
# Invocation
/brainstorming— explore requirements and design before implementation/execute-plan— run batched implementation plans with review checkpoints- Code-reviewer subagent evaluates against plans, coding standards, architectural principles
- Debugging methodology: root cause → pattern analysis → hypothesis testing → implementation
- Safeguard: architectural review triggers after 3 failed fix attempts
writing-skillsmodule teaches authoring new skills via TDD applied to documentation