fix(ci): reconcile the cross-repo sparse-scope mirror and the sandboxed exit test

[dieterolson]

Three tests fail on main at edfeae684 today, before this PR. They are invisible because neither file lands in main's changed-file set, so delta CI never runs them. The practical effect is that the first branch to touch either file inherits the failures and appears to have caused them — which is what happened to #4447.

tests/ops/test_cross_repo_python_integration.py
  ::test_each_downstream_declares_its_required_sparse_scope
  ::test_upstream_scope_includes_every_release_build_package_root
tests/rotation_converter/test_scripting_env.py
  ::TestConsoleEnvironment::test_refresh_user_functions_system_exit_propagates

Reproduce on a clean checkout of main:

git worktree add --detach /tmp/probe origin/main
cd /tmp/probe && pytest tests/ops/test_cross_repo_python_integration.py \
  tests/rotation_converter/test_scripting_env.py
# 3 failed

1-2. The cross-repo sparse scope and its mirror test drifted apart

The workflow declares UpstreamDrift's scope as src/chat, src/contracts.py, src/python/src/utils, src/shared, src/sidekick, tests/shared_contracts, tests/support. The test asserts the pre-src/ names — chat, contracts.py, python/src/utils, shared, sidekick — plus src/shared/python. Both are relics of an UpstreamDrift reorganization that neither side fully followed.

I checked every candidate against UpstreamDrift's default branch via the API, not a local clone:

path	exists on UD main?
src/shared, tests/shared_contracts, tests/support	yes
shared, src/shared/python	yes
chat, sidekick, contracts.py, python/src/utils	no — in neither form

Four declared paths match nothing at all. Removed rather than replaced with a guess — a sparse entry that resolves to nothing is the same vacuous-gate family as #4477.

Which side should move was the real question, and the test answers it itself:

assert "src" not in upstream_scope
assert "ui" not in upstream_scope

Narrowness is a deliberate constraint, not drift. So declaring bare src was not available to me, even though UpstreamDrift's hatchling config says packages = ["src"] and that would have been the simplest way to make the mirror agree. src/shared is the narrow root that actually carries the packages this repo provides — UD's own test_tools_vendoring.py resolves from src/shared/python, and cone-mode sparse checkout populates src/shared without pulling all of src. The workflow was the closer side, so the test's expectation is what moved.

Resulting scope:

D-sorganization/Gasification_Model -> src, tests/shared_contracts
D-sorganization/UpstreamDrift      -> src/shared, tests/shared_contracts, tests/support

Left out deliberately, so this stays reviewable and does not collide with work in flight:

• vendor/ud-tools and the pinned-submodule init step are added by Variation / Morris Sensitivity Suite: consolidate 34 draft PRs (#4142, #4433, #4120, #4206) #4447.
• tests/fixtures exists on UD and its vendoring test seeds it onto sys.path, so it may be a further gap — flagged, not fixed here.

Context a reviewer should have: the Cross-Repo Python Integration workflow's last completed run on main was a failure (2026-08-06). So the previously declared path set cannot be assumed to have been the working configuration either. This PR makes the declaration honest about what exists; it does not claim the cross-repo job is green end to end.

3. The sandboxed SystemExit test

The scripting sandbox now blocks import sys — Error loading user library: import of 'sys' is blocked in the scripting sandbox — so the test's import sys; sys.exit(1) never reaches sys.exit and no SystemExit is raised.

The sandbox is correct behaviour; the test is stale against a good security change. The payload is now raise SystemExit(1): SystemExit is a builtin, needs no import, and the test goes back to checking what its name says — that it propagates rather than being caught and logged.

Why this is a separate PR

These are main's defects, not any one branch's. Fixing them here unblocks #4447, #4446 and every future branch that touches these files, instead of one; and it keeps a 34-PR consolidation from absorbing unrelated repairs.

Verification

• 15 passed across both test files on Python 3.11 (was 3 failed).
• ruff check and ruff format --check clean on both changed test files.
• Changed-test assertion gate passes.
• Workflow still parses as valid YAML; both jobs intact.
• All three files keep their existing line endings (the workflow is CRLF on main and stays CRLF, so the diff is content only).

Part of #4460
Part of #4481

🤖 Generated with Claude Code