What problem does it solve? Prompts that ship to users are often edited ad hoc, stored outside version control, and changed without testing, causing silent behavior regressions. This Skill treats system prompts as production artifacts with enforced structure, versioning, and regression gates. ## Core Features & Use Cases - Seven-Section Prompt Anatomy: Enforces a load-bearing structure covering role, negative charter, injected facts, field-level register specs, numbered behavior flows, tie-break rules, and enumerated off-limits topics. - Versioning Discipline: Requires prompts in source control, single-source-of-truth fact injection, dated model pinning with env revert levers, and reviewed diffs with stated reasons. - Regression Gates: Mandates golden routing sets, 100% off-limits suites, register lint checks, and latency measurement before any prompt change merges. - Use Case: When a teammate says "the model keeps saying X" or proposes switching models, use this Skill to diagnose whether the fix is structural or prompt-level, write the change as a reviewed diff, and gate it behind a full regression run. ## Quick Start Review my system prompt for the customer support agent and tell me whether it is production-grade, then rewrite the weak sections.
