fix(tests): repair v0.32.0 release validation contracts

[Daniel Meppiel (danielmeppiel)]

fix(tests): repair v0.32.0 release validation contracts

TL;DR

Repair the two test contracts that blocked the unpublished v0.32.0 release after #3080.
Plugin uninstall now checks removed files/lock ownership and the unchanged retained skill instead of obsolete cleanup wording.
The live Windows-marker collection check gets a finite 300-second budget with explicit environment and fail-closed timeout coverage.

Important

Test-only release repair: no runtime, workflow, version, dependency, README, or changelog changes.
This PR does not merge, move the v0.32.0 tag, or publish.
Protected-branch review, required checks, and merge queue still apply.

Problem (WHY)

• The v0.32.0 tag run failed plugin sequential-install coverage on Linux ARM and macOS ARM. The install/uninstall succeeded, but the test required Cleaned up and integrated while actual cleanup reported Cleaned 2 stale files. Linux x64 was cancelled by matrix fail-fast, not a separately confirmed assertion failure.
• The same run's macOS Intel unit job timed out after 120 seconds during live nested collection. Its four-worker suite otherwise reported 22,768 passed, 4 skipped, and 21 xfailed. The parent also diagnosed this timeout in main run 36155951461.
• [!] -m windows_compat deselects only after importing and collecting both roots. A small selected family does not imply a small collection workload.

The scope follows the observed failures rather than changing production behavior to fit a test.
Validation is execution-backed, following the Agent Skills instruction to
"do the work, run a validator (a script, a reference checklist, or a self-check), fix any issues, and repeat until validation passes."

Approach (WHAT)

• Replace the cleanup-message check with one shared filesystem/lockfile assertion used by local and network-backed sequential-install cases.
• Extend the existing local fixture case to cover both an independent skill and a separately installed skill from the plugin.
• Keep real pytest collection, both collection roots, Unix-import blocking, plugin-autoload isolation, and strict non-empty subset selection; change only the finite subprocess budget.

Implementation (HOW)

File	Intent
tests/integration/test_plugin_e2e.py	Snapshot retained skill bytes, require non-vacuous plugin deployment cleanup, remove plugin lock ownership, and preserve the complete surviving skill entry and files. Reuse these assertions in the existing network case; require deps tree success.
tests/unit/test_windows_compat_gate_workflow.py	Increase nested collection from 120 to 300 seconds, matching the existing repository taxonomy collection budget. Pin timeout/cwd/interpreter/environment arguments and propagation of TimeoutExpired.
docs/src/content/docs/contributing/integration-testing.md	Document the collection budget and state-based plugin lifecycle assertions. No user-facing release entry for internal test machinery.

Diagrams

Omitted under the template's omission provision: these are independent test-contract repairs, with no runtime architecture or workflow-topology change.

Trade-offs

• Finite headroom, not a timeout bypass. A stuck collector can now consume up to five minutes instead of two. The existing taxonomy guard already uses 300 seconds; the collector still raises on timeout and never returns a success-shaped fallback.
• Hermetic local evidence, retained hosted coverage. Local real-CLI fixtures cover standalone and plugin-origin skills. The pinned external network-backed case remains intact, but was not run locally; hosted integration/release validation remains necessary.
• No collection optimization refactor. Avoid per-file allowlists, skipping the live selector, or changing production output. The measured workload and existing slower-runner failures support a bounded budget change; they do not prove a future hosted run will pass.

Benefits

1. Both hermetic skill-origin variants prove plugin files disappear while the survivor's SKILL.md, other deployed file bytes, and dependency entry stay unchanged.
2. The real network-backed lifecycle case shares the same state assertions instead of coupling cleanup correctness to prose.
3. Full-root marker selection remains live, with explicit regression coverage that a real timeout fails closed.

Validation

Local validation passed; hosted PR/release checks are not claimed green.
All commands used the editable 0.32.0 checkout and its locked dependencies.
Local dependency setup used the ambient Microsoft proxy without committing proxy configuration.
UV_NO_SYNC=1 prevented lockfile rewrites.

Collection evidence and targeted results

Before the change, the entire Windows-gate contract file:

23 passed in 32.66s

Its nested collection call took 30.36 seconds cold.
The exact helper on a warm cache:

wall time: 9.45s; exit: 0
566/32601 tests collected (32035 deselected) in 6.94s

A separate cProfile collection imported 1,156 test modules:

566/32601 tests collected (32035 deselected) in 12.58s

UV_NO_SYNC=1 uv run --extra dev pytest -p no:cacheprovider -q tests/unit/test_windows_compat_gate_workflow.py tests/integration/test_plugin_e2e.py::TestPluginHeroScenarios::test_local_plugin_uninstall_after_sequential_skill_install_cleans_deployed_files --durations=5

26 passed in 15.29s

UV_NO_SYNC=1 uv run --extra dev pytest -p no:cacheprovider -q tests/integration/test_plugin_e2e.py -m 'not requires_e2e_mode and not requires_network_integration' --durations=5

15 passed, 12 skipped in 10.55s

The skipped cases are the existing network opt-in class; no skip was added.

UV_NO_SYNC=1 uv run --extra dev pytest -p no:cacheprovider -q tests/unit/test_windows_compat_gate_workflow.py -n 4 --dist worksteal --durations=3

24 passed in 11.30s

Full lint, strict test-quality ratchets, and scope checks

After fetching and merging origin/main, the canonical lint chain used UV_NO_SYNC=1:

uv run --extra dev ruff check src/ tests/ scripts/lint_architecture_boundaries.py scripts/architecture_linter/
uv run --extra dev ruff format --check src/ tests/ scripts/lint_architecture_boundaries.py scripts/architecture_linter/
uv run --extra dev python -m pylint --disable=all --enable=R0801 --min-similarity-lines=10 --fail-on=R0801 src/apm_cli/ scripts/lint_architecture_boundaries.py scripts/architecture_linter/
bash scripts/lint-auth-signals.sh
bash scripts/lint-architecture-boundaries.sh

Output excerpts:

All checks passed!
1881 files already formatted
Your code has been rated at 10.00/10 (previous run: 10.00/10, +0.00)
[+] auth-signal lint clean

The architecture-boundary script exited 0 without output.
The remaining YAML I/O, 2100-line, and portable-path guards also passed using Python equivalents of the CI grep/awk predicates because ripgrep is unavailable in the local shell.

UV_NO_SYNC=1 uv run --frozen python scripts/check_test_assertions.py and UV_NO_SYNC=1 uv run --frozen python scripts/check_exact_test_duplicates.py:

[+] assertion-quality ratchet clean: AQ001=4, AQ002=12
[+] exact test duplicate ratchet clean: 1241 files, 0 allowed duplicate group(s)

git diff --check and git diff --exit-code origin/main -- uv.lock pyproject.toml CHANGELOG.md README.md exited 0.
The docs-impact classifier returned schema-valid no_change, high confidence, at deterministic L0: only tests and contributor documentation changed.

Scenario Evidence

#	Scenario (user promise)	Principle(s)	Test(s) proving it	Type
1	Install a plugin, then an independent skill; removing the plugin must preserve that skill.	DevX (pragmatic as npm), Governed by policy	tests/integration/test_plugin_e2e.py::TestPluginHeroScenarios::test_local_plugin_uninstall_after_sequential_skill_install_cleans_deployed_files[standalone]	e2e
2	Keep a separately installed plugin skill while removing the plugin's other files and lock ownership.	DevX (pragmatic as npm), Governed by policy	Same local test with [plugin-skill]; TestPluginNetworkE2E::test_lockfile_preserved_on_sequential_install retains hosted coverage (regression contract for run 36156463741).	e2e
3	New Windows-marked regressions remain discoverable without maintaining a file list.	Portability by manifest, OSS / community-driven	tests/unit/test_windows_compat_gate_workflow.py::test_windows_compat_gate_marker_selects_nonempty_subset	integration
4	A hung validation collector must fail, with isolated plugin loading and a finite budget.	OSS / community-driven	tests/unit/test_windows_compat_gate_workflow.py::test_nested_collection_preserves_environment_and_bounded_timeout and ::test_nested_collection_timeout_fails_closed	unit

How to test

• Use a frozen development environment, or the documented local proxy-safe bootstrap; keep UV_NO_SYNC=1 when dependencies are already installed.
• Run the combined 26-case command above; expect both local skill variants and the live collection contract to pass.
• Run the four-worker Windows-gate contract command; expect 24 passing tests, including fail-closed timeout coverage.
• Run the canonical lint and strict test-quality commands above; expect exit 0 and no uv.lock diff.
• Let protected PR checks and hosted integration validation complete before merge. The parent release session owns subsequent tag correction and publication follow-through.

Co-authored-by: Copilot 223556219+Copilot@users.noreply.github.com

[Copilot]

Copilot review overview

🟡 Changes recommended

Exclude shared retained-skill paths from the plugin-only removal assertion.

Get a fresh assessment by requesting another Copilot review.

Review effort: Lite
Findings: 1

Open (1)

• Exclude shared skill paths from plugin removal assertions · New

What changed in this PR

Test-only repair for plugin lifecycle validation and Windows compatibility-gate collection.

Changes:

• Adds filesystem and lockfile cleanup assertions.
• Adds bounded, fail-closed collection timeout coverage.
• Documents the revised test contracts.

File	Summary
tests/​unit/​test_windows_compat_gate_workflow.py	Adds bounded timeout and environment validation.
tests/​integration/​test_plugin_e2e.py	Strengthens plugin cleanup and retained-skill checks.
docs/​src/​content/​docs/​contributing/​integration-testing.md	Documents updated test contracts.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.