Beyond CLAUDE.md: deterministic CI validation of architecture rules for AI coding agents

razxrr · reddit · 2026-09-11

A Reddit post argues coding agents (Claude Code, Cursor, Codex, Copilot) can generate code that compiles and passes tests yet violates architectural decisions. The author proposes a "governance layer": store normative rules (decisions, business rules, architecture) in Git under a structured .context/ directory, expose them to agents via MCP, and let CI validate changes deterministically without calling an LLM. Core thesis: memory tells an agent what happened; governance tells it what's allowed. The author is building a project around this and soliciting discussion on whether tests + review + ADRs already suffice.

Original post →

More from coding & agent

coding & agent channel →