test-and-fix

name	test-and-fix
description	Autonomously run tests, analyze failures, and fix them in the `integration_tests` directory.

Test and Fix

Purpose

This skill provides an autonomous loop to identify, analyze, and fix test failures within the integration_tests directory using the project's make commands.

Loop Logic

1. Identify: Run tests from the integration_tests directory.
   • Agent must make sure what type of test the user wants to run (unit or integration), becaues it takes a long time to run all tests.
   • make test: Runs all unit and integration tests (both dbt-core and dbt-fusion). This is the standard command for a full verification.
   • make run-unit-tests: Runs all unit tests (both core and fusion).
   • make run-integration-tests: Runs all integration tests (both core and fusion).
   • For targeted debugging, use granular sub-targets:
     • make run-unit-tests-core / make run-unit-tests-fusion
     • make run-integration-tests-core / make run-integration-tests-fusion
2. Analyze: Examine the test output and logs to understand which tests are failing.
   • Check the logs/dbt.log in integration_tests/ for detailed dbt execution logs.
   • For macro failures, inspect the generated SQL and compare it with expected BigQuery syntax.
   • Note: BigQuery is the only supported warehouse in this repository.
3. Plan Fix: Before applying any fix, create an MVP scaffold that delivers the smallest valuable slice of the fix.
   • Define acceptance criteria for the fix.
   • Create a plan in memory to track the fix progress.
4. Execute & Test:
   • Implement the MVP scaffold.
   • Run the relevant tests immediately (e.g., cd integration_tests && make run-unit-tests-core).
   • Only add details and enhancements after the scaffold tests pass.
5. Verify: Re-run the tests.
   • If all checks pass: Termination.
   • If new or remaining failures exist: Analyze the failure and repeat the loop.

Termination Criteria

• All relevant tests exit with code 0 (all checks passed).
• Reached max iteration limit of 5 attempts.
• No progress being made (same errors persisting despite fix attempts).

Examples

Scenario: Fixing a failing macro unit test

1. Agent runs make test and identifies a failure in test_macros for dbt-core.
2. Agent analyzes the output and finds that macro_a is returning incorrect SQL for BigQuery.
3. Agent creates a plan in memory.
4. Agent implements a minimal fix for macro_a to address the specific failure.
5. Agent runs make run-unit-tests-core to quickly verify the fix.
6. The test passes. Agent runs make test to ensure no regressions across the entire suite.
7. The test passes. Agent marks the task as complete.