compound-engineering_skill

Overview

This skill implements the Compound Engineering workflow for AI-assisted development. It guides teams through a Plan → Work → Review → Compound loop so each unit of engineering makes subsequent work easier. Use it to plan features, execute code, run reviews, and codify learnings so knowledge compounds across the codebase.

How this skill works

The skill inspects the codebase, commit history, and documentation to produce a detailed implementation plan tailored to existing patterns. During execution it enforces systematic work: isolated branches, incremental commits, and continuous validation. The review step runs a checklist for quality, security, performance, and tests. Finally it generates concrete artifacts (patterns, decisions, lessons) and updates project guidance so future work is faster and safer.

When to use it

• Planning a new feature that should follow existing patterns
• Implementing a change where long-term maintainability matters
• Reviewing pull requests to ensure consistency and robust tests
• Codifying post-mortems, decisions, and patterns for future contributors
• Creating acceptance criteria and testing strategies before coding

Best practices

• Invest 60–80% effort in planning and review; execution should be disciplined and incremental
• Search codebase and git history for priors before inventing new patterns
• Keep commits small, focused, and accompanied by clear messages and tests
• Prefer simple duplication over premature abstraction; document any new abstraction with rationale
• Turn every bug or design choice into a documented lesson, test, or AGENTS.md entry

Example use cases

• Generate a comprehensive plan for a multi-day feature with acceptance criteria and risks
• Run a pre-merge review checklist on a PR and produce an actionable feedback list
• Execute a plan: create a feature branch, break tasks into TODOs, and validate after each change
• Compound learnings after a bug fix: write a lesson, add a regression test, and update subsystem AGENTS.md
• Convert architectural decisions from a design discussion into a Decision artifact with trade-offs