Name: superpowers-plus
Author: bordenet

Stats

Actions

Available In

Tags

superpowers-plus

Skills for AI coding assistants that enforce the practices AI would otherwise skip. Built on bordenet/superpowers, a maintained fork of Jesse Vincent's obra/superpowers (MIT). Can be used for non-coding workloads, too!

Platform Support

Platform

Status

Claude Code

✅ Full support — skills, lifecycle hooks (SessionStart, PreCompact, PreToolUse), commit gates, pre-push hooks, and red-autonomy guardrails all install and run cleanly.

Augment Code

✅ Full support — skills, routing, commit gates, pre-push hooks, and MCP integrations install and run cleanly.

What This Is

AI coding assistants skip the practices that catch bugs before production: they implement the first idea without evaluating alternatives and claim "done" without verification.

Skills are structured procedures that AI agents follow automatically. obra/superpowers by Jesse Vincent is a framework for teaching AI agents reusable procedures; superpowers-plus builds on it via a maintained fork (bordenet/superpowers) for governance stability. superpowers-plus adds skills across 9 domains. Start debugging and systematic-debugging enforces root-cause investigation before fixes. Commit code and a gate chain blocks the commit until lint, type checks, and security scans pass.

Each skill exists because it caught a real problem.

Standout Skills

Skill

What it does

code-review-battery

Dispatches up to 6 specialist reviewers in parallel (Defect Finder, Design Critic, Guardian, Standards Enforcer, Performance Analyst; AttackerPersona on security-sensitive diffs). BugPath Verifier activates in bug-fix mode. Slash command: /sp-cr-battery [min-score] (optional 1.0–10.0 quality threshold, default 7.0).

debate

Generates 3+ decision options, builds a comparison matrix, then red-teams the winner. Requires adversarial review before committing to an approach.

progressive-harsh-review

Three escalating critic personas score non-code deliverables (plans, docs, designs) on 5 dimensions. Score below 6 = rejected.

systematic-debugging

Enforces root-cause-first investigation: reproduce, hypothesize, isolate, fix. No fixes without completing Phase 1.

feature-development

Full lifecycle orchestrator: brainstorm, debate, plan, TDD, review, verify.

think-twice

Detects when the AI is stuck in a loop and dispatches a fresh sub-agent with zero shared context. Auto-triggers on circular reasoning.

context-ferry

Generates a self-contained resume prompt before context compaction fires. Updates in-progress plan docs, captures pending questions and queued tasks verbatim. Slash command: /context-ferry. Auto-fires via PreCompact hook in Claude Code.

detecting-ai-slop

Scores text 0-100 for machine-generated patterns across lexical, structural, semantic, and stylometric dimensions.

wiki-orchestrator

Pipeline for bulk documentation: de-dup, content, coherence, links, secrets, slop detection, fact-check, publish.

evolution-loop

Self-improvement cycle: scans failures for recurring patterns, generates skill updates, tracks metrics over time.

unified-commit-gate

Runs all 5 commit gates in sequence (lint/build/test → style → code review → language → IP audit). Slash command: /sp-commit. Deep-dive into any gate via its individual skill.

Quick Start

git clone https://github.com/bordenet/superpowers-plus.git && cd superpowers-plus && bash install.sh

Enable pre-commit gates: bash tools/install-hooks.sh

These hooks are required if you want the full commit-gate chain to run locally before git commit and git push.

Then tell your AI assistant what you're doing: