Programming as Theory Building Skill
A Claude Code skill that applies Peter Naur's 1985 paper 'Programming as Theory Building' to coding-agent workflows, asking the agent to state the domain rule, place changes beside the closest existing facility, and verify behavior before writing code.
Programming as Theory Building Skill is a Claude Code plugin and reusable coding-agent skill built on Peter Naur's 1985 paper of the same name, whose central claim is that a program's durable asset is not just its text but the programmer's theory of how it maps real-world affairs into behavior. The skill turns that idea into operational checks that a coding agent runs through before non-trivial code changes: rebuilding the theory behind the change, placing new code beside the closest existing facility by similarity, keeping changes surgical, avoiding speculative flexibility, and verifying the theory rather than only the syntax.
It targets a specific failure mode in coding-agent output: code that satisfies the prompt's surface but misses the invariant it should protect, producing misplaced helpers, tests that prove only the happy path, and speculative abstractions added before they are needed. The plugin packages one Claude Code skill together with a project-level CLAUDE.md guideline file, and an accompanying benchmark compares commerce-backend code generation across isolated arms with and without the skill.
It is written in Python and MIT licensed.