Fix branch CI

Get the current branch green on GitHub Actions. Work in a loop: read pipeline → reproduce locally → fix → verify → re-check CI.

Read root AGENTS.md and CLAUDE.md before changing code — especially the code style rules, the async conventions, and the exact lint/typecheck/test commands.

Do not weaken CI workflows or skip checks to get green. Do not fix unrelated pre-existing failures unless merging main into the branch resolves them.

Phase 1 — Read the pipeline

1. Establish git context

Run in parallel when possible:

git branch --show-current
git status -sb
gh pr view --json url,headRefName,baseRefName,statusCheckRollup 2>/dev/null || true

• No open PR: use gh run list --branch "$(git branch --show-current)" --limit 15
• Open PR: prefer statusCheckRollup from gh pr view for a single summary

2. List failing checks

From statusCheckRollup, collect every item with conclusion: FAILURE or state: FAILURE. Note:

• Workflow name (e.g. @wandelbots/wandelbots-nova: Typecheck, Lint and Format)
• Job name (e.g. test, yamllint, validate-title)
• detailsUrl (link to the run)

For each failure, fetch logs:

gh run view <run-id> --log-failed

If the URL is a job link, extract run-id from the path or use:

gh run list --branch "$(git branch --show-current)" --workflow "<workflow-file-name>" --limit 3

See references/gh-commands.md for more patterns.

3. Build a failure table (show the user)

Before editing code, summarize:

Check	Job	Likely scope	Local repro command
…	…	paths touched on branch	uv run …

Map workflow → local command using references/nova-ci.md.

Prioritize: the fast Typecheck, Lint and Format gate first (format → import order → lint → typecheck → unit tests), then integration/example jobs.

Phase 2 — Reproduce locally

Run the same command CI runs. Make sure the environment matches first:

uv sync --extra "nova-rerun-bridge" --extra "wandelscript" --extra "novax"

Common repro commands (from nova-dev.yaml):

uv run ruff format --check .          # formatting gate
uv run ruff check --select I          # import order
uv run ruff check .                   # lint
uv run ty check                       # typecheck
PYTHONPATH=. LOG_LEVEL=WARNING uv run pytest -rs -v -m "not integration"   # unit tests

Single test file / test:

PYTHONPATH=. uv run pytest -rs -v path/to/test_file.py
PYTHONPATH=. uv run pytest -rs -v path/to/test_file.py::test_name

If local passes but CI fails: check Python version (3.11 in the workflows), missing extras in uv sync, PYTHONPATH=., or a required env var (CELL_NAME, NOVA_API, NOVA_ACCESS_TOKEN).

Phase 3 — Fix (surgical)

Failure patterns in this repo

CI signal	Usual cause	Fix direction
ruff format --check fails	Unformatted code	uv run ruff format .
ruff check --select I fails	Import order	uv run ruff check --select I --fix
ruff check . fails	Lint violations	Fix the reported rule; avoid blanket # noqa
ty check errors	Type drift / bad annotation	Fix types (modern list[T] syntax); do not add # type: ignore to silence real issues
Unit test failure	Real bug or stale test	Fix code or test logic; keep tests async-correct (await, async with)
Integration test failure (-m integration)	Needs live NOVA instance	Usually CI-infra; only debug locally if you have NOVA_API + NOVA_ACCESS_TOKEN
Example run failed (python examples/…)	API/SDK change broke an example	Update the example to the current SDK, or fix the regression
PR title check	Title not Conventional Commits	Retitle PR to chore|feat|fix[(scope)][!]: Description
YAML Lint	.github/workflow YAML style	Fix per .yamllint; run yamllint . locally if installed
uv audit (SAST)	Vulnerable dependency	Bump the dependency in pyproject.toml; escalate if no fix available

Fix rules

1. Match the failure — only change what the failing check implies.
2. Prefer local verify before telling the user to push.
3. Never disable a check (no removing ty errors with ignores, no --no-verify, no deleting/relaxing tests) to force green without understanding the diff.
4. Async: everything in this SDK is async — keep await / async with correct when touching runtime code or tests.
5. Escalate when: failure is on main too, needs a running NOVA instance or secrets you don't have, or the fix would weaken a check.

Phase 4 — Verify and loop

After fixes, re-run the exact failing command(s):

uv run ruff format --check . && uv run ruff check --select I && uv run ruff check . && uv run ty check
PYTHONPATH=. LOG_LEVEL=WARNING uv run pytest -rs -v -m "not integration"

Optional — re-query CI (user must push first for new runs):

gh pr view --json statusCheckRollup
gh run list --branch "$(git branch --show-current)" --limit 5

Tell the user:

• What failed and why (one sentence per check)
• What you changed
• What you ran locally
• Whether they need to push for CI to re-run
• Anything still red or outside scope

Phase 5 — Multi-failure ordering

When several checks fail:

1. Format (ruff format --check) — cheapest
2. Import order (ruff check --select I)
3. Lint (ruff check .)
4. Typecheck (ty check)
5. Unit tests (pytest -m "not integration")
6. Integration tests / examples (need a live NOVA instance; slow, often infra)

Fix one cluster at a time; re-run the local check before moving on.

Reference files

• references/gh-commands.md — gh CLI for runs, logs, PR checks
• references/nova-ci.md — workflow → local command mapping and env notes

Output format

End with a short status block:

## CI status
- Branch: …
- PR: … (if any)

## Fixed
- [workflow/job]: cause → change → verified with …

## Still failing / needs you
- …

## Next step
Push and re-watch CI, or run: …