dbt-autofix: Instructions file for Codex

AGENTS.md

Goals

1. Make a pre-existing dbt project compatible with the new authoring layer, so it can parse and compile successfully on the Fusion engine.
2. Preserve existing functionality as much as possible.
3. Flag when the dbt project is using functionality that's not yet supported in Fusion, and ask users so they can choose whether to wait for feature support or remove use of unsupported features.

Migrating to the new authoring layer and the dbt Fusion engine

if a user says "migrate my dbt project to the new authoring layer" or "make my dbt project compatible with the Fusion engine" do the following steps and the migration is only successfully completed once dbtf compile finishes with 0 errors. Create a changes_made.md file documenting and summarizing all of the code changes made in the user's dbt project, including what the error was, the fix applied, and a summary of why the fix was chosen.

Before you begin, please read and understand the resources section below. You should strictly use the resources below to update the project.

1. Run dbtf debug in the terminal to check their data platform connections. Proceed to step 2 if there are no errors. If there are errors, please summarize the error succinctly so the user knows how to debug on their own.
2. Run dbtf parse --show-all-deprecations in the terminal to check for compatibility errors in their current project. Summarize the log output by specifying how many errors were found and group the errors in a way that's easily understandable.
3. Install dbt-autofix and run autofix to try to fix errors. Prefer uv/uvx to install (uv tool install dbt-autofix) and run the package (uvx dbt-autofix deprecations), but fall back to pip and other methods if needed. Summarize the results of the autofix run and include how many errors were resolved. Run dbtf parse again to check for remaining errors and summarize with how many errors were found and a brief summary of the types of errors.
4. For remaining errors, please ONLY use the resources below to attempt to resolve them. If you can't figure out a fix from the resources below, notify the user and break out of the flow. Attempt the fixes error by error, grouping similar errors based on the error code and message. You should also summarize which error you're working on in the chat to give users context.