agent-rules-books: classic software books as rules for agents
Agents write code that runs yet tend to add shallow abstractions, blur layers and mix refactoring with new features. This repo offers 14 rule sets drawn from well-known books on design, architecture, DDD and legacy code, each in three sizes: mini for daily tasks, nano for tight context budgets and full as the complete reference.
Clean Code and Code Complete rules on naming, small functions, responsibilities, defensive programming and tests.
Architecture sets for Clean Architecture, Patterns of Enterprise Application Architecture and three books on domain-driven design.
Legacy work after Michael Feathers: characterization tests, seams, dependency breaking, sprout method and wrap method.
The release-it set after Michael Nygard covers timeouts, retries, circuit breakers, bulkheads, backpressure and observability.
Every SKILL.md loads the mini rule set by default and keeps the full file at hand as deeper reference.
When to use it
Before a large refactor, when the agent must move in small steps and keep observable behaviour intact.
A DDD codebase where the agent should reason in bounded contexts and aggregates, not tables and DTOs.
Reviewing a service with queues and third-party APIs, where timeouts and retries decide uptime.
Ask your agent
Refactor the billing module following the refactoring rules: small steps, run tests after each one
Review this service against the release-it rules and flag calls with no timeout or retry limit
Split the legacy OrderManager class, writing characterization tests first
Questions
How do I install agent-rules-books?
Install every set with npx skills add ciembor/agent-rules-books --all, or one with --skill and its name, for example --skill refactoring. Add --list to see what is available without installing.
Does agent-rules-books work in Cursor and Codex?
Yes. The rules are tool-agnostic Markdown for AGENTS.md or skills, and docs/USAGE.md explains the setup for Codex, Claude Code and Cursor.
What is the difference between mini, nano and full?
full is the canonical complete rule text, mini is the recommended shorter version for real tasks, and nano is the compact fallback for very tight context.
