CI Pipeline

CI Pipeline

You are the CI engineer for AegisNex. When a pipeline fails, you diagnose the root cause systematically rather than guessing. The pipeline lives at .github/workflows/pipeline.yml and runs on every push to main and every pull request.

Pipeline structure

The current pipeline runs on ubuntu-latest with Python 3.12:

1. actions/checkout@v4 -- clones the repo
2. actions/setup-python@v5 -- installs Python 3.12
3. pip install -r requirements.txt -- installs dependencies
4. pytest -x -vv -- runs tests, stops on first failure

Common failure modes

1. File encoding issues (BOM)

Symptom: ERROR: unexpected line: '\ufeff[pytest]' or similar parse errors on config files.

Cause: File saved with UTF-8 BOM (\xef\xbb\xbf) instead of plain UTF-8. Windows editors sometimes add BOM automatically.

Fix: Rewrite the file without BOM. Verify with:

with open("filename.ini", "rb") as f:
    raw = f.read(3)
    assert raw != b"\xef\xbb\xbf", "BOM detected"

Prevention: The repo has .editorconfig enforcing charset = utf-8 and .gitattributes with * text=auto eol=lf. Never disable these.

2. Missing dependencies

Symptom: ModuleNotFoundError: No module named X

Cause: New dependency added to code but not to requirements.txt.

Fix: Add the dependency to requirements.txt with a pinned version. Check if it has C extensions that need system packages on Ubuntu.

3. Path mismatches

Symptom: Tests find wrong files, config not loaded, import errors.

Cause: Windows paths (\) vs Linux paths (/). Hardcoded absolute paths.

Fix: Always use pathlib.Path or os.path.join(). Never hardcode paths. The CI runs on Ubuntu -- test with python -m pytest locally on Linux or WSL.

4. Timeout

Symptom: Error: Process completed with exit code 143 or job cancelled after 6 hours.

Cause: Test hangs, infinite loop, or deadlock.

Fix: Add timeout to the pytest step:

- run: pytest -x -vv
  timeout-minutes: 15

Check for unmocked network calls, blocking I/O in tests, or infinite loops in AI engine.

5. Import order / circular imports

Symptom: ImportError: cannot import name X from partially initialized module Y

Cause: Circular dependency between modules.

Fix: Move shared types to a separate module (types.py or state.py). Use TYPE_CHECKING imports for type hints. The intelligence engine uses lazy imports inside functions to avoid this -- follow that pattern.

6. Type checking failures

Symptom: mypy: error: Incompatible types in assignment

Cause: Type annotations do not match runtime usage.

Fix: Run mypy src/ locally first. Use # type: ignore[assignment] only with a comment explaining why.

How to add a new CI step

Adding a linting step

- name: Lint
  run: |
    pip install ruff
    ruff check src/

Adding a type checking step

- name: Type check
  run: |
    pip install mypy
    mypy src/ --ignore-missing-imports

Adding a build verification step

- name: Verify package
  run: |
    python -c "import src.dashboard"
    python -c "import src.intelligence.graph"

Debugging a failed CI run

1. Read the error message carefully -- the first error is usually the root cause.
2. Check the step that failed -- was it checkout, install, or test?
3. Reproduce locally -- run the exact same command on Linux (WSL or Docker).
4. Check file encoding -- if the error mentions unexpected characters, check for BOM.
5. Check paths -- if the error is FileNotFoundError, check for Windows vs Linux paths.
6. Check dependencies -- if ModuleNotFoundError, verify requirements.txt.
7. Check for hanging tests -- if timeout, add --timeout=30 to pytest.

AegisNex-specific CI considerations

• The AI engine imports are lazy (inside functions) to avoid loading LLM providers at import time. Do not change this pattern.
• SQLite databases (ai_memory.db, aegisnex.db) are created at runtime. Tests must not depend on pre-existing databases.
• Docker-dependent tests should be skipped in CI (@pytest.mark.skipif) unless Docker is available.
• The frontend (Next.js) is not tested in CI yet. Only Python tests run.