2.6Storing and Versioning Templates

Keep your prompt templates in your repository alongside the test code:

tests/
  prompts/
    requirements-to-tests.prompt.md
    api-schema-to-tests.prompt.md
    tdd-prompt.prompt.md
    mutation-aware.prompt.md
  test_users.py
  test_orders.py

Version them in git. When you refine a template based on review feedback, commit the change with a message explaining what improved. This creates institutional knowledge about effective test generation.

Treat Prompts Like Code

By mid-2026 this has hardened from a tip into a team practice: prompts are versioned, reviewed, and tested like any other source file.

• Versioned: templates live in the repo (as above), change via PR, and get meaningful commit messages.
• Reviewed: a prompt change that alters what tests get generated is a behavioral change -- it goes through code review like one.
• Tested: keep a small set of golden inputs (a known spec excerpt) and periodically regenerate against them. If a template edit or a model upgrade suddenly produces weaker tests -- fewer negative cases, vaguer assertions -- you catch the regression before the whole team inherits it. Chapter LLM Evals and AI Test-Suite Quality covers this eval discipline in depth.

If you package team prompts as reusable skills, use the SKILL.md format: as of July 2026 it is portable across Claude Code, Cursor, Gemini CLI, and Codex CLI, so a test-generation skill written once works regardless of which assistant each engineer prefers.