Fix CI

Core Principle

Follow the test philosophy from AGENTS.md — CI failures mean the code is wrong, not the tests or CI config.

Workflow

1. Read the failure output — full error message, stack trace, failing stage
2. Scope to the failing stage only — ignore passing stages entirely
3. Identify the root cause — trace the error to the source in application code
4. Fix the code to satisfy the existing assertions/checks
5. Run linter + type checker locally — ruff check, pyright, etc.
6. Run the failing tests locally — verify the fix before pushing

Config Awareness

Before hardcoding values or adding environment variables:

• Check pyproject.toml for project configuration
• Check settings files (settings.py, .env.example) for existing patterns
• Use config files over env vars when the project already does

Anti-Patterns

Stop immediately if you're about to:

• Change test assertions — the test defines expected behavior
• Investigate passing stages — they're irrelevant to the failure
• Modify CI config (.gitlab-ci.yml, .github/workflows/) — unless explicitly asked
• Add env vars when config files exist — follow existing config patterns
• Broaden scope — "while I'm fixing CI, let me also..." — don't

Scope Rules

• One failing stage at a time
• Only touch files related to the failure
• If the fix requires changes across 3+ files, confirm with the user
• If you can't reproduce locally, say so — don't guess

Integration

• Use systematic-debugging for complex root cause analysis
• Use testing skill patterns when writing reproduction tests