Hey everyone! I've been tasked with helping to standardize how our engineering team uses our coding assistant (think Copilot, Cline, etc.). We're about 25 data engineers and analysts, mostly working in Python with Airflow, some dbt, and our data warehouse is Snowflake.
The goal is to create a shared set of custom instructions or a base prompt that everyone can use as a starting point. The idea is to improve consistency, especially for common patterns like writing idempotent data pipelines, adding proper logging, and following our Snowflake connection standards.
But honestly, I'm terrified 😅. My attempts so far have been... messy. I tried a simple markdown file in our internal wiki with a "copy-paste these instructions," but:
* Two people immediately overwrote the Airflow task template with their own shortcuts.
* Someone's prompt had a typo that made the assistant generate code with a weird indent bug, and it took us a day to trace it back to the shared instructions.
* The feedback is all over the place: "too restrictive," "not specific enough," "doesn't work for my dbt models."
I'm looking for concrete "recipes" or strategies from teams who've done this at scale. How do you manage it?
* **Versioning:** Do you keep the master prompt in a git repo? How do you notify people of updates?
* **Customization:** Do you have a mandatory "base" everyone uses, plus personal additions? How do you avoid the base being disabled accidentally?
* **Testing:** How do you even *test* a prompt change before rolling it out to 20+ people?
* **Buy-in:** How did you get engineers to actually use it instead of just doing their own thing?
Any real-world examples, especially from data pipeline folks, would be a lifesaver. Even screenshots of how you structured the instructions file would be amazing. I feel like I'm building a pipeline without a DAG definition – it's just chaos waiting to happen!
null