Engineering Loop Standard

An open, agent-agnostic specification for disciplined, AI-assisted software engineering. AI coding tools change every few months; the methodology that makes them produce trustworthy code does not.

engineering-loop.json is a small, versioned config. A CLI generates each tool’s native rules file from it — CLAUDE.md (Claude Code), .cursorrules (Cursor), CONVENTIONS.md (Aider), AGENTS.md (Codex/Gemini), .windsurfrules (Windsurf). Write the spec once, generate an adapter for every tool.

npx engineering-loop init

Read the specification View on GitHub

The manifesto

Full text and rationale in MANIFESTO.md on GitHub. Six principles govern every adapter this project generates:

  1. Understand before acting.
  2. Research before building.
  3. Plan before execution.
  4. Verify before completing.
  5. Document as you go.
  6. Leave the codebase better than you found it.

Where to go next

  • Specification — the full standard: the config schema, the 10-phase engineering loop, the persistent-memory convention, and how adapters work.
  • Writing an Adapter — the interface a new AI-tool adapter must implement, for contributing support beyond the five tools covered today.
  • engineering-loop.schema.json — the formal JSON Schema (draft 2020-12), pinned to the current spec tag.

Want to see it applied to a real, working project instead of a bare scaffold? See the todo-api example.

Engineering Loop Standard is MIT licensed. Source and issues on GitHub.

This site uses Just the Docs, a documentation theme for Jekyll.