Superpowers Plugin Enforces Claude Code Discipline
Superpowers adds 14 skills to Claude Code for clarify-design-plan-code-verify phases, cutting tokens 14% and boosting quality on medium/complex tasks via automatic dispatching and human-in-loop visuals.
Master Dispatcher Automates 14 Skills Across Phases
Superpowers installs a 'using superpowers' master skill that fires at every Claude Code conversation start, scanning and dispatching from 14 specialized skills without manual prompts. Core phases enforce developer discipline: clarify via brainstorming that asks 4-5 targeted questions to fill spec gaps; design generates visual companions like local HTML dashboards showing 3 UI options (e.g., force graphs vs. card grids for knowledge explorers) with pros/cons for user selection; plan outputs hyper-detailed checklists with 2-5 minute tasks, exact file paths, and inline code tests savable for reuse; code via 'executing plans' with safety stops on blockers, sub-agent-driven development (fresh agents per task with reviews), and parallel agents for independent subtasks; verify through test-driven development (write failing tests first, then minimal passing code), four-phase systematic debugging (investigate root cause, analyze, hypothesize, fix), and completion verification.
Skills invoke automatically—brainstorming and planning nearly always trigger on new requests, execution/quality ones contextually (e.g., TDD or debugging on demand). Append 'use relevant superpower skills' to prompts for insurance. Meta-skill 'writing skills' lets Claude extend the system test-driven: write failing test scenario, code skill to pass it, close loopholes.
This beats Claude's Ultra Plan (planning-only) by guiding full implementation, preventing misalignment before token-heavy coding.
Visual Dashboards Align Before Building
Brainstorming skill spins up localhost dashboards for interactive previews, e.g., website heroes (cinematic full-bleed vs. split-screen vs. centered video) or mind-map nodes with hover/click interactions, filters, and search bars. Users select options (e.g., 'graph hero + card details'), refining via feedback like 'make it modern, dark mode, polished'—avoids 4-5 revisions by confirming vision early. For reports/automations, clarifying questions extract unstated needs, ensuring spec compliance without over-prompting.
Use terminal over VS Code extension for better visibility with status lines; visuals are token-intensive but save via fewer retries.
Experiments Prove Token Savings and Quality Lift
12 automated runs (6 with/without Superpowers, Opus model, $2/run cap, no human loop) across simple/medium/complex tasks yielded 9% cost savings, 14% fewer total tokens overall, fewer API turns. Simple tasks used 8% more tokens (skip Superpowers here—no need for over-engineering). Medium/complex saw savings as planning prevents backtracking; with-Superpowers variance was 2-3x tighter.
Quality radar: Superpowers excelled in correctness, code structure, test coverage, error handling (larger green pentagon area); robustness slightly worse (subjective). Domain knowledge/spec compliance unchanged—model-limited. Value: consistency reduces expensive retries; automate experiments via Claude scripts calling sessions with/without plugin.
Install Globally in Seconds
In Claude Code terminal/VS Code: plugin install gh:obra/superpowers --user for global use across projects (avoids per-project setup). Set-and-forget; read full doc free in author's Skool community.