## GitHub Discussion 1: Prompt Versioning Best Practices

[andche1512]

Body

Question:
What’s your workflow for versioning LLM prompts to keep experiments reproducible and easy to roll back?

My Context:

• I have multiple prompt templates (visual vs. text-only browsing).
• I’m currently naming files prompt_v1.txt, prompt_v2.txt, etc., and relying on Git commits—but it’s hard to track changes at a glance.

What I’d Love to Learn:

1. Tools or patterns for tracking prompt changes (e.g., prompts-as-code frameworks).
2. Ways to integrate prompt metadata (version, date, description) so dashboards or CI/CD pipelines can pick up the right prompt.
3. Tips for documenting prompt changes in GitHub (commits vs. PRs vs. tags).

Guidelines

• I have read and understood this category's guidelines before making this post.

[AbdulazizAl-Surabi]

Hey Andche1512,
A good approach is to use a “prompts” folder in your repo with subfolders per prompt family, and store a prompt.json manifest alongside each version file:
text
prompts/
web-browsing/
v1.0/
prompt.txt
metadata.yaml # {version: 1.0, date: 2024-05-01, description: "Initial prompt with basic instructions"}
v1.1/
prompt.txt
metadata.yaml

• Tools: PromptFlow, LangSmith help track versions and provenance.
• Git tags: Tag releases (git tag v1.1) to mark prompt snapshots.
• CI integration: Use a CI job to validate prompt syntax and deploy the correct version to your evaluation servers.