You Can't Prompt Your Way Out of a Hard Constraint

An n8n-based content pipeline exposed how easily LLMs can miss hidden formatting rules and follow raw imperative text instead.

After removing a verifier, the author found overlong posts, unsupported markdown, and weak image prompts slipping through.

Prompt-only fixes did not hold; code-based guards like character checks, splitting tweets, and templates solved the issue.

The takeaway: hard constraints in AI workflows should be enforced in code, not left to prompts alone.