AI Literacy Superpowers: A Complete Framework for Building AI-Native Development Workflows

Core Idea: AI Literacy Superpowers is a plugin system for Claude Code and GitHub Copilot that encodes best practices like Constraints Engineering, Literate Programming, and CUPID code reviews into executable agent workflows, enabling systematic governance of AI-assisted development. This project aims to solve the problem of upgrading AI coding assistants from personal tools to team infrastructure, making the way AI participates in development predictable, repeatable, and improvable.

When AI coding assistants like Claude Code and GitHub Copilot evolve from experimental toys to daily tools, developers face new challenges: scattered tips and prompts cannot form team capabilities. The AI Literacy Superpowers project emerged to encode mature software engineering principles into executable agent workflows, pushing AI-assisted development from personal magic to team infrastructure.

AI Literacy Superpowers is a plugin system created and maintained by Russ Miles, supporting both Claude Code and GitHub Copilot CLI. Based on the AI Literacy framework, it transforms concepts like Constraints Engineering, Agent Orchestration, and Literate Programming into specific skills, agents, hooks, and commands. The core innovation lies in the 'executable development philosophy'—instead of instilling theory, it enables AI assistants to perform checks, reviews, and improvements at the right time. Run superpowers-init after installation to set up a complete AI-assisted development environment.

Skill System: The basic unit that encapsulates domain-specific knowledge, divided into four categories: code quality (Literate Programming, CUPID review), security and supply chain (GitHub Actions hardening, dependency vulnerability audit, Docker Scout audit), engineering practices (Constraints Engineering, Context Engineering, etc.), and governance and evaluation (AI literacy assessment, constraint observability, etc.), totaling 14 skills.

Agent Team: Covers the entire development lifecycle, including orchestrators (pipeline coordination), spec writers (updating specifications), TDD agents (writing failing tests), code reviewers (from CUPID/Literate Programming perspectives), integration agents (handling PR/CI, etc.), as well as constraint discoverers and enforcers. Different agents have different permissions, reflecting the principle of trust boundaries.

Constraints Engineering: The core concept—constraints are verifiable and executable rules, using a verification slot model to unify deterministic checks (e.g., lint) and agent checks (e.g., code review). Execution timing is divided into three layers: edit time (PreToolUse hook warns without blocking), pre-commit (Git hook enforces), and CI (continuous integration verification).

Model Routing: Guides cost-sensitive model selection via MODEL_ROUTING.md, assigning levels (strongest/standard/fast) based on agent roles. The orchestrator references the configuration and sets token budgets to balance quality and cost.

Composite Learning: Accumulates team knowledge via AGENTS.md and REFLECTION_LOG.md—humans curate conventions, and agents propose improvements. The reflect command captures learning points after tasks and appends them to the log as future context,沉淀 team experience.

Hook System: Registers five auto-activated hooks: PreToolUse constraint gate (warns of violations during editing), stop drift check (detects configuration changes and prompts audits), stop snapshot staleness check (prompts constraint updates), stop reflection prompt (suggests reflection on commit), etc., ensuring constraints are integrated into the development process.

Installation Steps: For Claude Code users: Add plugin marketplace claude plugin marketplace add russmiles/ai-literacy-superpowers → Install claude plugin install ai-literacy-superpowers; For GitHub Copilot CLI users: plugin install ai-literacy-superpowers.

Generate Files: Running superpowers-init generates CLAUDE.md (conventions), HARNESS.md (constraint documents), AGENTS.md (composite learning), MODEL_ROUTING.md (model guide), REFLECTION_LOG.md (reflection log), and CI templates, etc.

AI Literacy Superpowers marks the transition of AI-assisted development from personal skills to team engineering practices, converting abstract principles into executable, verifiable, and improvable workflows, making AI a reliable partner. For teams looking to systematically introduce AI-assisted development, it provides a complete implementation path: from skill learning to agent configuration, from constraint design to composite learning. As AI coding assistants become more popular, such frameworks will become an important part of team engineering capabilities.