[Amaury Levé (Evangelink)]

Pull request

Summary

Fixes #12131. Adds opt-in failed-test attachments to Microsoft.Testing.Extensions.GitHubActionsReport:

• --report-gh-attachments <absolute-directory> stages declared regular files with bounded size/counts and per-session test/module/retry manifests.
• Primary failure details retain archive-relative paths. The documented official upload-artifact workflow adds real download links in supplementary per-test details after upload; credentials remain outside the test process.
• Metadata is checked before opening a source stream. FIFOs, links to FIFOs, devices, directories and other non-regular entries are warned and recorded as non-regular with no staged path. This prevents a no-writer Linux FIFO from blocking session completion and the primary summary.
• Missing inputs, cancellation and reporting errors do not hide test results. Session-level dumps are intentionally out of scope.

Native issue type: Feature. Next open milestone: MSTest 5.0 / MTP 3.0, preserving the maintainer's concurrent update. Current build/package versions remain 4.6/2.6 below.

Design and decision record

Maintained research and decisions.

GPT-6.1 Sol and Claude Opus 5.5 reviewed the plan, followed by three independent code-review/fix rounds. In-process artifact credentials, a separately shipped companion action, and deferred primary reporting were rejected: upload failure must not hide normal diagnostics.

FIFO follow-up: the filesystem abstraction now exposes metadata-only regular-file classification. Unix uses the runtime's fixed-width, normalized stat ABI, not architecture-specific libc stat layouts. File.Exists, FileAttributes and size-zero checks cannot distinguish FIFOs from empty regular files. Producers continue to own path lifetime and must not retarget a declared path concurrently with consumption.

Demonstration

Final-head hosted validation passed on ubuntu-latest at 3936f06c13a65d2354cbcc30d528cde062dc038d: standard pack, both real Linux FIFO regressions, composed smoke, official upload, the exact documented post-upload script, real download, and byte/test/file/URL verification.

Linux no-writer FIFO regressions: total 2; succeeded 2; skipped 0.
AttachmentsA/disabled: host exit 2; executed 2; failed 1.
AttachmentsA/passing: host exit 0; executed 2; failed 0.
AttachmentsA/failed: host exit 2; executed 2; failed 1.
AttachmentsB/failed: host exit 2; executed 2; failed 1.
Native multi-module dotnet test: parent exit 2; executed 4; failed 2.
Verified downloaded artifact: 8 files, including both module manifests.

Actual attachment archive · reproducible validation evidence. Four staged files plus four manifests were downloaded and checked. Archive digest: sha256:3699c1290fd0b5468743e71dbaa6d8be02c4760c746fba6f068fff0c336f0608.

Reproducer: .\.github\scripts\test-gh-attachments.ps1 -WorkDirectory <fresh-task-owned-directory> — wrapper exit 0, with 12 actual consumer executions and five intentional failed results. Fresh MSTest packages: 4.6.0-dev; platform/MSBuild/TRX/GitHubActionsReport: 2.6.0-dev; pinned SDK: 12.0.100-alpha.1.26509.101. The script checks nupkg SHA256, resolved runtime-DLL hashes, final associations and merged TRX without double-counting.

The public browser showed the successful job and artifacts. Authenticated expanded summary rendering, wrapping and controls were not visually inspected because the available browser is signed out. Saved real Markdown/console transcripts are the demonstration; no substitute rendering or fabricated screenshot is claimed.

Validation and regression proof

Current FIFO fix: 3936f06c13a65d2354cbcc30d528cde062dc038d. Pinned SDK compilation of net8.0 and netstandard2.0 with TreatWarningsAsErrors=true: exit 0, no warnings. Pinned Linux .NET 8.0.33 executed all 272 focused cases: 272 passed, 0 failed/skipped. Windows executed 270 passed, with the two explicitly Linux-only cases skipped there; those skips are not Linux proof.

The hosted job executes this exact focused regression command after packing — exit 0, 2 executed/passed, 0 skipped:

./.dotnet/dotnet test --project test/UnitTests/Microsoft.Testing.Extensions.UnitTests/Microsoft.Testing.Extensions.UnitTests.csproj --configuration Release --framework net8.0 --no-build --no-restore -p:EnableCodeCoverage=false -bl:artifacts/log/Release/fifo-regression-{}.binlog --filter FullyQualifiedName~GitHubActionsSummaryReporterTests.SessionFinishing_LinuxFifo_RejectsBeforeOpenAndPreservesPrimarySummary --minimum-expected-tests 2 --report-trx --report-trx-filename fifo-regression.trx --results-directory artifacts/TestResults/fifo-regression

Requirement	Evidence
Reject non-regular sources before stream creation	CaptureAsync_NonRegularSource_IsRejectedBeforeOpening — passed; verifies no open, warning, null path and manifest
Actual FIFO/no-writer behavior and primary reporting	SessionFinishing_LinuxFifo_RejectsBeforeOpenAndPreservesPrimarySummary — both direct-FIFO and symlink cases passed locally and in hosted Linux CI
Staging, limits, failure isolation and retries	CaptureAsync_StagesExactBytesAndWritesManifest, CaptureAsync_ModuleByteBudget_AllowsExactLimitButRejectsNextByte, ConsumeAsync_AttachmentFailure_DoesNotDropFailedRecordOrHistory, ReadAndAggregate_ProcessRetries_RenderLastFailureAndKeepEarlierArchive — included in 272-case Linux pass
Real packaged MSTest and native multi-module composition	Final-head hosted smoke: 12 executions, five expected failures, four staged files/manifests, wrapper exit 0
Real GitHub publishing	Actual upload/download, eight files checked and four supplementary per-test details matched to the returned URL

Earlier feature validation remains applicable history: 269 cases passed on net8.0 and 269 on net462/net481 (netstandard2.0), plus 11 packed acceptance/help cases, with actual package/cache/DLL provenance verified. TRX/JUnit/CTRF exports were reconciled. The current Linux run additionally covers the FIFO fix.

Guards: check_source_bom.py, check_action_pins.py, git diff --check, and markdownlint-cli2 on changed docs — passed. Resource catalogs were generated with owning-project UpdateXlf; the original localization check passed 41 guard cases and 364 catalogs. The FIFO fix adds no resource strings.

Local limitations: the full Windows package verifier expects legacy UAP payloads omitted by local SkipUapTargetFramework=true; it was not weakened or called successful. Standard Linux build.sh -pack passed on the current head. An optional standalone NativeAOT probe stopped during restore because the shared drive filled; this is not NativeAOT execution proof. The import uses libSystem.Native, matching the runtime's NativeAOT direct-P/Invoke library name. Only owned optional SDK/comparison resources were removed; no global caches were cleared.

Test-value evidence

FIFO red/green: retained the same two Linux cases/assertions and removed only the collector's regular-file guard in a task-owned comparison. Both executed and failed Times.Never assertions because one forbidden source open was attempted per case (exit 2, 2 failed, 0 skipped). The stream-open trap fails immediately rather than opening a FIFO, so no hung worker/writer or infrastructure timeout is counted as proof. Rebuilt restored production libraries were hash-verified; the identical tests then passed (exit 0, 2 passed, 0 skipped).

Contract / test case	Without behavior	Correct implementation
SessionFinishing_LinuxFifo_RejectsBeforeOpenAndPreservesPrimarySummary (FIFO and symlink)	Executed: detected — one forbidden read-stream open per case	Both passed; original failed verdict, summary/history and manifest preserved
CaptureAsync_StagesExactBytesAndWritesManifest	Earlier buildable collection ablation detected missing staged metadata	Passed before and after restoration
SessionFinishing_DeferredFragment_RetainsAttachmentsForAggregateRendering	Earlier buildable collection ablation detected missing fragment metadata	Passed before and after restoration

Original feature-ablation base: 6b9bbb4b0f956a32f1920e2b8e53410fa85e8da2; support patch SHA256 2DF6A32E9C2267B4B1B60A0C9265070280AB88AD375655F47F7F91E1C0A620C5; production ablation SHA256 95A07BC3991A486811747B55BFBD51A89074FCB107C78DECA3E74F0224EA470B. Initial null-reference and setup/disk failures were rejected as behavioral evidence. Retained test bytes, comparison patches, actual reports and logs were preserved before owned comparison cleanup.

[github-actions]

Note

🤖 Automated review by GitHub Copilot. Generated by the Expert Code Review workflow. To request a follow-up action, reply by tagging @copilot directly.

Summary

I read the diff for the new GitHub Actions failed-test attachment feature. I focused on GitHubActionsAttachmentCollector, the reporter integration, and the new test-gh-attachments workflow and script. I did not build or run anything. I left one inline comment and have no blocking findings.

Collector and reporter

• The size and count limits, CreateNew plus cleanup of partial files, atomic manifest write through a .tmp file, filename sanitization, and cancellation propagation all look sound.
• Failures in capture or manifest writing are reported as warnings and do not fail the test run.
• Only failed, non-superseded results are captured, which matches the stated scope.
• Open item: the module file counter also counts the synthetic count-limit entry (see the inline comment).

Build, dependencies & scripts

• No dependency version changes.
• The workflow uses minimal permissions (contents: read) and persist-credentials: false. All actions are pinned by SHA.
• Please confirm that cancel-in-progress on push to main is intended, because it can cancel a post-merge validation run. Consider limiting it to pull requests.

Localization and API

• The .resx and .xlf files are consistent. The new internal API is declared in InternalAPI.Unshipped.txt.
• I did not verify that the .xlf files were regenerated through UpdateXlf, or that the --help and --info expectations match the new CLI option. HelpInfoAllExtensionsTests is updated in this PR, which suggests they do.

Tests

• The unit and acceptance coverage is broad. It includes byte-exact staging, limits, cancellation, and a round trip through the real uploaded archive.
• The test-value evidence in the decision record looks adequate. I did not re-execute it.

[github-actions]

🧪 Test quality review — PR #12184

Compared merge base 3a95626573dd6932505a860cbd059dae4f79a81d with head dd69b2f47dc2ab1f4b9df23c8d2d3451f21ec1cd. Evidence applies to this head only.

Scope note: this was a bounded, static triage pass. I did not complete per-test rubric grading or run the specialist parallel-safety subtask, and no tests were executed.

Test-value evidence

• New attachment collector contract: GitHubActionsAttachmentCollectorTests has 16 test methods covering staged bytes and manifest, duplicate basenames, long names, byte and count budgets, growth, cancellation and warning-callback failure. They look relevant to the changed contract. This is a static prediction only, with no executed red/green records.
• Summary reporter and command-line option changes: GitHubActionsSummaryReporterAttachmentTests (17 tests), the updated GitHubActionsCommandLineProviderTests and GitHubActionsSummaryReporterTests, plus the acceptance tests in GitHubActionsAttachmentTests (5) and HelpInfoAllExtensionsTests, appear to cover the new behavior. Their exact assertions were not individually traced, so the evidence is unverified.
• No executed paired results were available for this head.

🧵 Parallel safety

STATUS: PARTIAL. The specialist task was not run. A spot check found the new unit tests use a unique per-instance temp directory (Guid-based), and grep found no DoNotParallelize, Thread.Sleep or Environment.SetEnvironmentVariable use in the new test files. The acceptance test ConcurrentModulesSharingRoot_KeepDistinctManifestsAndBytes deliberately shares a root and was not reviewed for isolation. This is not a clean result.

Per-test assessment

Per-test grades were not produced in this pass, so every changed test method is Uncertain pending the full rubric. There are no inline findings.

_{This advisory review is heuristic and non-blocking. Re-run with /review-tests, or /parallel-audit for the parallel-safety specialist.}

🤖 Automated content by GitHub Copilot. Generated by the Test Quality Review on PR (on open / sync) workflow. · copilot · auto · 26.7 AIC · ⌖ 0.816 AIC · ⊞ 7.8K · ◷

[github-actions]

🧪 Test quality review — PR #12184

Compared merge base 3a95626573dd6932505a860cbd059dae4f79a81d with head 41c23d5301d4ac807c0f4530bfe1d93b22ba1404. Evidence applies to this head, not every intermediate commit. This was a static, read-only pass: nothing was built or executed, and no inline findings are published.

Test-value evidence

• Attachment staging, limits and cancellation (GitHubActionsAttachmentCollectorTests, ~16 methods). Static prediction: the byte-exact staging assertion, exact-limit versus limit+1 byte boundaries, the count limits, cancellation cleanup and the manifest-retention cases should catch an implementation that stages nothing or ignores a limit. The PR description reports an executed ablation for CaptureAsync_StagesExactBytesAndWritesManifest. I could not access the paired TRX/binlog records, so this is unverified here.
• Reporter integration (GitHubActionsSummaryReporterAttachmentTests and the changed GitHubActionsSummaryReporterTests). Static prediction: the failure-kind data rows read the staged bytes back and compare them exactly. The author reports an executed ablation of the deferred-fragment case. This is likewise unverified here.
• CLI option (GitHubActionsCommandLineProviderTests). Static prediction: the option arity and visibility, the invalid-path data rows, and the check that a directory is not created or an existing file overwritten are all asserted exactly. The HelpInfoAllExtensionsTests expectation edits cover the new help and info output.
• Acceptance (GitHubActionsAttachmentTests, 5 methods). These use the packed package. Static prediction: they cover opt-in, disabled and outside-CI behavior, unavailable files, and concurrent modules sharing a root. I did not trace each one end to end.
• Non-test contracts (workflow, PowerShell script, docs). These are not applicable to unit-level evidence. The hosted upload and download run is pending, per the author.
• No executed paired results were available to me, so no red-green claim is made.

🧵 Parallel safety

STATUS: CLEAN. The specialist did not run analyzers, so this rests on source reading.

Assembly	Scope
Microsoft.Testing.Extensions.UnitTests	MethodLevel
Microsoft.Testing.Platform.Acceptance.IntegrationTests	assembly-level method parallelization

• The changed tests use Guid-unique temp directories.
• No static mutable state or process-wide environment-variable mutation was found.
• The acceptance GH_ATTACHMENT_* variables are only read in the child test asset, and each run gets its own root under TestRunDirectory.
• Not reviewed: the HelpInfoAllExtensionsTests.cs expectation-string edits.
• Counts: Critical=0, High=0, Warning=0, Info=0.

Per-test assessment

Per-test rubric grading was not completed method by method. Methods are grouped by class and graded from the source read. The rows therefore do not list each method.

Pseudo-mutation evidence is not included, and any survivor claim would be a static prediction rather than an executed result.

Result	Quality	Test	Notes	How to improve
Pass	A (90–100)	new GitHubActionsAttachmentCollectorTests. CaptureAsync_* / WriteManifestAsync_*	Exact bytes, boundaries and cancellation outcomes are asserted.	None
Pass	A (90–100)	new GitHubActionsSummaryReporterTests. ConsumeAsync_* (attachment partial)	Staged bytes and the failure-kind data rows are compared exactly.	None
Pass	A (90–100)	new/mod GitHubActionsCommandLineProviderTests. *Attachments*	Option shape, rejected paths, and no-side-effect behavior are asserted.	None
Pass	B (80–89)	new GitHubActionsAttachmentTests. *	Process-level tests are long, but their assertions are on observable outputs.	None
Uncertain	—	mod GitHubActionsSummaryReporterTests. (existing methods with changed lines)	These edits were not traced individually.	Trace each edited existing method to confirm its assertions are unchanged.

_{This advisory review is heuristic and non-blocking.}

Re-run the combined review with /review-tests, or only the parallel-safety specialist with /parallel-audit.

🤖 Automated content by GitHub Copilot. Generated by the Test Quality Review on PR (on open / sync) workflow. · copilot · auto · 56.7 AIC · ⌖ 0.834 AIC · ⊞ 7.8K · ◷

[Copilot]

🟡 Changes recommended

A FIFO attachment with no writer can block synchronously on Linux, preventing session shutdown and summary generation; the new byte-count suffix also bypasses localization resources.

2 open issues

• Prevent FIFO attachments from blocking session shutdown · New
• Localize the attachment size text · New

What changed in this PR

Adds opt-in failed-test attachment support to Microsoft.Testing.Extensions.GitHubActionsReport for #12131, keeping artifact upload and authentication in the workflow rather than the test process.

Changes:

• Stages declared failed-test files with size/count limits and per-session manifests.
• Preserves attachment associations through summaries, aggregation, and retries.
• Adds workflow documentation, upload/download validation, localization resources, and unit and acceptance coverage.

Test-value evidence: The PR reports two-test green/red/green feature-ablation evidence against 6b9bbb4b0f956a32f1920e2b8e53410fa85e8da2 with retained assertion changes. Execution artifacts were not independently verified here; hosted upload/download and rendered UI validation remain pending.

File	Description
test/​UnitTests/​Microsoft.Testing.Extensions.UnitTests/​GitHubActionsSummaryReporterTests.cs	Extends the shared reporter fixture.
test/​UnitTests/​Microsoft.Testing.Extensions.UnitTests/​GitHubActionsSummaryReporterAttachmentTests.cs	Tests attachment reporting, retries, and compatibility.
test/​UnitTests/​Microsoft.Testing.Extensions.UnitTests/​GitHubActionsCommandLineProviderTests.cs	Tests attachment-option validation.
test/​UnitTests/​Microsoft.Testing.Extensions.UnitTests/​GitHubActionsAttachmentCollectorTests.cs	Tests staging limits, manifests, and failure handling.
test/​IntegrationTests/​Microsoft.Testing.Platform.Acceptance.IntegrationTests/​HelpInfoAllExtensionsTests.cs	Updates help and info expectations.
test/​IntegrationTests/​Microsoft.Testing.Platform.Acceptance.IntegrationTests/​GitHubActionsAttachmentTests.cs	Adds packed-consumer attachment scenarios.
src/​Platform/​SharedExtensionHelpers/​SummaryReporterHelpers.cs	Adds attachment metadata to failure details.
src/​Platform/​SharedExtensionHelpers/​CiRunSummaryModels.cs	Adds optional attachment serialization models.
src/​Platform/​SharedExtensionHelpers/​CiRunSummaryAggregation.cs	Carries attachments into failure summaries.
src/​Platform/​Microsoft.Testing.Platform/​CommandLine/​CommandLineOptionsValidator.UnknownAndBootstrapValidation.cs	Recognizes the attachment option.
src/​Platform/​Microsoft.Testing.Extensions.GitHubActionsReport/​Resources/​xlf/​GitHubActionsResources.zh-Hant.xlf	Adds Traditional Chinese catalog entries.
src/​Platform/​Microsoft.Testing.Extensions.GitHubActionsReport/​Resources/​xlf/​GitHubActionsResources.zh-Hans.xlf	Adds Simplified Chinese catalog entries.
src/​Platform/​Microsoft.Testing.Extensions.GitHubActionsReport/​Resources/​xlf/​GitHubActionsResources.tr.xlf	Adds Turkish catalog entries.
src/​Platform/​Microsoft.Testing.Extensions.GitHubActionsReport/​Resources/​xlf/​GitHubActionsResources.ru.xlf	Adds Russian catalog entries.
src/​Platform/​Microsoft.Testing.Extensions.GitHubActionsReport/​Resources/​xlf/​GitHubActionsResources.pt-BR.xlf	Adds Brazilian Portuguese catalog entries.
src/​Platform/​Microsoft.Testing.Extensions.GitHubActionsReport/​Resources/​xlf/​GitHubActionsResources.pl.xlf	Adds Polish catalog entries.
src/​Platform/​Microsoft.Testing.Extensions.GitHubActionsReport/​Resources/​xlf/​GitHubActionsResources.ko.xlf	Adds Korean catalog entries.
src/​Platform/​Microsoft.Testing.Extensions.GitHubActionsReport/​Resources/​xlf/​GitHubActionsResources.ja.xlf	Adds Japanese catalog entries.
src/​Platform/​Microsoft.Testing.Extensions.GitHubActionsReport/​Resources/​xlf/​GitHubActionsResources.it.xlf	Adds Italian catalog entries.
src/​Platform/​Microsoft.Testing.Extensions.GitHubActionsReport/​Resources/​xlf/​GitHubActionsResources.fr.xlf	Adds French catalog entries.
src/​Platform/​Microsoft.Testing.Extensions.GitHubActionsReport/​Resources/​xlf/​GitHubActionsResources.es.xlf	Adds Spanish catalog entries.
src/​Platform/​Microsoft.Testing.Extensions.GitHubActionsReport/​Resources/​xlf/​GitHubActionsResources.de.xlf	Adds German catalog entries.
src/​Platform/​Microsoft.Testing.Extensions.GitHubActionsReport/​Resources/​xlf/​GitHubActionsResources.cs.xlf	Adds Czech catalog entries.
src/​Platform/​Microsoft.Testing.Extensions.GitHubActionsReport/​Resources/​GitHubActionsResources.resx	Defines attachment help, labels, and warnings.
src/​Platform/​Microsoft.Testing.Extensions.GitHubActionsReport/​PACKAGE.md	Documents staging and workflow-owned upload.
src/​Platform/​Microsoft.Testing.Extensions.GitHubActionsReport/​InternalAPI/​InternalAPI.Unshipped.txt	Tracks new internal attachment APIs.
src/​Platform/​Microsoft.Testing.Extensions.GitHubActionsReport/​GitHubActionsSummaryReporter.SummaryWriting.cs	Writes manifests and isolates reporting errors.
src/​Platform/​Microsoft.Testing.Extensions.GitHubActionsReport/​GitHubActionsSummaryReporter.Snapshot.cs	Retains attachments in failure snapshots.
src/​Platform/​Microsoft.Testing.Extensions.GitHubActionsReport/​GitHubActionsSummaryReporter.Markdown.PerProject.cs	Passes attachments to Markdown rendering.
src/​Platform/​Microsoft.Testing.Extensions.GitHubActionsReport/​GitHubActionsSummaryReporter.cs	Collects attachments from eligible failed results.
src/​Platform/​Microsoft.Testing.Extensions.GitHubActionsReport/​GitHubActionsFailureDetails.cs	Renders archive-relative attachment paths.
src/​Platform/​Microsoft.Testing.Extensions.GitHubActionsReport/​GitHubActionsCommandLineProvider.cs	Registers and validates the staging directory.
src/​Platform/​Microsoft.Testing.Extensions.GitHubActionsReport/​GitHubActionsCommandLineOptions.cs	Defines --report-gh-attachments.
src/​Platform/​Microsoft.Testing.Extensions.GitHubActionsReport/​GitHubActionsAttachmentCollector.cs	Implements bounded copying and manifest generation.
src/​Platform/​Microsoft.Testing.Extensions.AzureDevOpsReport/​InternalAPI/​InternalAPI.Unshipped.txt	Tracks shared failure-model additions.
docs/​decisions/​github-actions-test-attachments.md	Records design choices and validation evidence.
.github/​workflows/​test-gh-attachments.yml	Adds hosted artifact upload/download validation.
.github/​scripts/​test-gh-attachments.ps1	Verifies packed consumers, provenance, and file associations.

_{Review effort: Balanced}

[github-actions]

🧪 Test quality review — PR #12184

Compared merge base d3ce600610baaebac245acdbd4162615f5dfcaae with head cc156a9267f76a2d5c1f8712e4b1ae64b18a3db7. Evidence applies to this head only.

Test-value evidence

This is a static review. No tests, builds or mutation tools were run, and no paired execution records were available. Green CI or author claims do not count as red/green proof.

• Attachment staging and manifest (GitHubActionsAttachmentCollector): GitHubActionsAttachmentCollectorTests has about 17 tests. They cover exact bytes, duplicate basenames, long names, size/byte/count boundaries, unavailable files, growing files, cancellation and manifest failures. The decisive assertions look independent of the implementation. Static prediction: a wrong limit, overwrite or leaked partial file would likely be caught. Not executed.
• Summary reporter attachment output (GitHubActionsSummaryReporterAttachmentTests, about 40 cases): static prediction only, not deeply traced.
• CLI option, help/info and acceptance (GitHubActionsCommandLineProviderTests, HelpInfoAllExtensionsTests, GitHubActionsAttachmentTests): changes are expected to be protected, but not traced in detail.
• Unassessed: I did not trace IFileSystem/SystemFileSystem changes, the shared summary-aggregation changes, the workflow/script/docs files, or the small edits in AzureDevOpsLivePublishingTests, TrxResultStreamingStoreTests and GitHubActionsSummaryReporterTests. This was due to review budget. No gaps are claimed for them.

🧵 Parallel safety

Status: CLEAN, but coverage was partial.

Assembly	Scope
Microsoft.Testing.Extensions.UnitTests	MethodLevel
Microsoft.Testing.Platform.Acceptance.IntegrationTests	assembly-level method parallelization

• The specialist read only the three focus files: the new collector test file, the new summary reporter attachment test file and the new acceptance test file.
• Each new test uses its own Guid-named directory and per-instance mocks, with no process-wide state.
• The other 5 changed test files were not read in detail.

Per-test assessment

Per-test grading is not published. About 90 new test methods were in scope and each could not be graded individually in this run. This is an evidence gap, not a clean result. No inline findings met the publication gate. Pseudo-mutation evidence is static prediction only.

_{This advisory review is heuristic and non-blocking. Re-run with /review-tests, or /parallel-audit for parallel safety only.}

🤖 Automated content by GitHub Copilot. Generated by the Test Quality Review on PR (on open / sync) workflow. · copilot · auto · ~50 AIC · ⌖ ~1 AIC · ⊞ 8K

[Copilot]

🟡 Changes recommended

SizeBytes is recorded before stream disposal succeeds, so a disposal failure can leave the manifest reporting staged bytes for an unavailable attachment whose destination was deleted.

1 open issue

• Reset staged byte count when destination cleanup fails · New

1 resolved since last review

• Localize the attachment size text

_{Review effort: Balanced}

[github-actions]

🧪 Test quality review — PR #12184

Compared merge base d3ce600610baaebac245acdbd4162615f5dfcaae with head 682446fb54465e4e5f543ddf631583fa2ae0171b. Evidence applies to this head only.

Scope limitation: this PR is very large (about 4.3k added lines, 8 changed test files, 33 source files). This run did a bounded static review and did not complete a full per-method grading pass. No inline findings were verified, so none are published.

Test-value evidence

• Executed: none from this workflow, which does not run code. The PR description claims a red/green comparison for SessionFinishing_LinuxFifo_RejectsBeforeOpenAndPreservesPrimarySummary and CaptureAsync_NonRegularSource_IsRejectedBeforeOpening (guard removed, Times.Never open assertions fail, then pass). I could not match that claim to accessible run records for this exact head, so treat it as an author claim.
• Static prediction: GitHubActionsAttachmentCollectorTests covers the main collector contracts with boundary data cases: file-size limit, byte budget, per-test and module counts, metadata budget, unavailable sources, cancellation, and manifest write failure. These look sensitive to the matching production limits, but I did not trace each assertion.
• Not assessed: the reporter integration tests (GitHubActionsSummaryReporterAttachmentTests, GitHubActionsSummaryReporterTests), the acceptance tests (GitHubActionsAttachmentTests), help/info expectations, GitHubActionsCommandLineProviderTests, and the workflow/script. The --info and --help expectation updates should be confirmed by running the acceptance tests after -pack.

🧵 Parallel safety

STATUS: PARTIAL. The parallel-safety specialist was not run in this budget, so coverage is incomplete and must not be read as clean. The collector tests use per-instance temporary directories and a Dispose cleanup, which looks favorable. The acceptance test ConcurrentModulesSharingRoot_KeepDistinctManifestsAndBytes deliberately shares a root and should be checked for shared generated assets.

Per-test assessment

<table>
<thead>
<tr><th>Result</th><th>Quality</th><th>Test</th><th>Notes</th><th>How to improve</th></tr>
</thead>
<tbody>
<tr>
<td>Uncertain</td>
<td>—</td>
<td>new tests in the 8 changed test files</td>
<td>Per-method grading was not completed for this large diff.</td>
<td>Re-run with <code>/review-tests</code>, or review by file.</td>
</tr>
</tbody>
</table>

_{This advisory review is heuristic and non-blocking. Existing-thread deduplication was not completed, so no inline comments were published. Re-run the combined review with /review-tests, or only the parallel-safety specialist with /parallel-audit.}

🤖 Automated content by GitHub Copilot. Generated by the Test Quality Review on PR (on open / sync) workflow. · copilot · auto · ~29 AIC · ⌖ ~1 AIC · ⊞ 8K

[Copilot]

🔵 Needs a closer look

Maintainer validation is needed because the new Unix native interop lacks NativeAOT execution evidence and authenticated expanded GitHub summary rendering remains uninspected.

No open issues 🎉

1 resolved since last review

• Reset staged byte count when destination cleanup fails

_{Review effort: Balanced}

[github-actions]

🧪 Test quality review — PR #12184

Compared merge base d3ce600610baaebac245acdbd4162615f5dfcaae with head 6f9de78dc0e276c91b3b6ab889cdaffca5d0fe3e. Evidence applies to this head, not every intermediate commit.

Test-value evidence

Static, budget-limited assessment. Nothing was executed in this workflow.

• FIFO and non-regular source rejection: CaptureAsync_NonRegularSource_IsRejectedBeforeOpening and SessionFinishing_LinuxFifo_RejectsBeforeOpenAndPreservesPrimarySummary assert that no stream is opened. Static prediction: removing the regular-file guard would be detected. The PR description reports an executed red/green result (2 failed, then 2 passed), but I did not have matching records to verify it.
• Staging, limits, manifest: CaptureAsync_StagesExactBytesAndWritesManifest, CaptureAsync_ModuleByteBudget_AllowsExactLimitButRejectsNextByte, CaptureAsync_PerTestCountLimit_StagesTenAndRecordsSkippedFile and CaptureAsync_ModuleCountLimit_StopsAtOneThousandAndMarksManifestTruncated assert exact bytes, names, boundary values and manifest fields. Static prediction: likely detected.
• Command-line option and help/info: GitHubActionsCommandLineProviderTests and HelpInfoAllExtensionsTests cover the option. Static prediction: likely detected.
• Not assessed or not applicable: the workflow, script, documentation and .xlf files are not applicable. Reporter rendering of the attachment links is only covered by static reading, and the rendered GitHub UI is a validation gap that the PR already discloses.

🧵 Parallel safety

STATUS: PARTIAL. The specialist task was not launched. A manual scan of the new test files found no changed parallelization configuration, no Environment.SetEnvironmentVariable, no Directory.SetCurrentDirectory and no mutable statics. The unit-test classes use per-instance GUID temp directories. This is not a clean verdict for the whole assembly.

Per-test assessment

Pseudo-mutation evidence is not enumerated below. Any such evidence would be a static prediction, not an executed mutation result. The new test files contain about 78 test methods across the changed test files (GitHubActionsAttachmentCollectorTests 19, GitHubActionsSummaryReporterAttachmentTests 19, GitHubActionsCommandLineProviderTests 34, acceptance 5, plus modified existing tests). I did not do a per-method grade pass, so I am not publishing per-row results.

<table>
<thead>
<tr><th>Result</th><th>Quality</th><th>Test</th><th>Notes</th><th>How to improve</th></tr>
</thead>
<tbody>
<tr>
<td>Uncertain</td>
<td>—</td>
<td>new <code>GitHubActionsAttachment*Tests</code> (all methods)</td>
<td>Per-method grading was not completed within this run's budget.</td>
<td>Re-run with <code>/review-tests</code> for a per-method grading pass.</td>
</tr>
</tbody>
</table>

No verified, actionable inline findings were identified, and dedup context from prior reviews did not show an existing report for this head.

<sub>This advisory review is heuristic and non-blocking.

Re-run the combined review with /review-tests, or only the parallel-safety specialist with /parallel-audit.</sub>

🤖 Automated content by GitHub Copilot. Generated by the Test Quality Review on PR (on open / sync) workflow. · copilot · auto · ~32 AIC · ⌖ ~1 AIC · ⊞ 8K

[Copilot]

🔵 Needs a closer look

Human assessment is needed because a critical target-framework mismatch prevents both FIFO regressions from running and leaves downstream consumer and download verification skipped.

No open issues 🎉

_{Review effort: Balanced}

[github-actions]

Final test-triage resolution — Azure Pipelines build 1662929

Resolution: inconclusive. Evidence is incomplete (1 artifact download failure and the download budget was exhausted). Historical comparison is also incomplete, so I can't claim these failures have no prior occurrences. No issue was created.

Build 1662929 (microsoft.testfx) ran against head 6f9de78d / merge d02cb0ec. These are the failure signatures I found:

Signature	Legs	Affected tests	Assessment
NU1903 (warning as error): Microsoft.DotNet.ILCompiler 11.0.0-rc.2.26504.105 has a known high-severity advisory, so restore of the generated NativeAOT test assets fails	Linux Release	NativeAotTests_*, NativeAot_Smoke_Test, AssemblyFixtureProvider_WhenDynamicCodeUnsupported_IsSkipped, PublishAotNonNative, SourceGenerationNonAotTests, and MSTest.Acceptance / MTP.Acceptance NativeAot tests	Likely a dependency or advisory problem, not tied to this PR's changes. Medium confidence.
Exit code 150: the test asset targets net8.0 but only 10.0, 11.0-rc.3 and 12.0-alpha runtimes are installed (Microsoft.NETCore.App 8.0.0 is missing)	Linux Release	GitHubActionsAttachmentTests.* (AttachmentsOptIn_StagesFailedFilesAndPreservesExistingOutput, AttachmentsAbsentOrOutsideGitHubActions_DoNotCreateStagingOutput, and others). The MSTestHostingInjection* tests also failed on assertions, with their output truncated.	Failed on all 3 attempts. The runtime gap is the probable cause. If this PR touched those tests, the target framework or roll-forward should be fixed there. Medium confidence.
Building the real UWP asset failed (x86 and ARM64 ModernUwp_BuildsMtpPackageForArchitecture, plus both ModernUwp_ConsumesUwpAssets_AndRunsPlainAndUiTestsThroughMtp cases)	Windows App Model Diagnostics	ModernUwpTests	Failed on both attempts. The cause is not shown in the retained output. Low confidence.
Assert.IsFalse(fail) ("expected host failure")	Windows App Model Diagnostics	NativeTests.Outcome in NativeSidecar TRX files	Likely intentional fixture output (the test asset deliberately fails). Not treated as a defect.
RetryRecovery_OpenDenied / Orchestrate_ManifestIoFailuresWarnAndPreserveAvailableArtifacts: expected message sequences differ	Windows Debug, net472	Microsoft.Testing.Extensions.UnitTests	Failed on attempt 2, with the attempt 1 outcome unclear. This is the most plausibly PR-related signal if the PR touches retry-artifact recovery. Check the manifest-read warning text and path normalization on net472.

No crash or hang diagnostics were collected, and there were no slow-test regressions. None of these failures shows a fail-then-pass recovery, so I'm not calling any of them flaky.

Next steps:

1. Check the NU1903 advisory suppression or ILCompiler version for the NativeAOT assets.
2. Check the net8.0 target for the GitHub Actions attachment test asset.
3. Review the RetryArtifactRecoveryProtocolTests expectation on net472.
4. Re-run the build to refresh the incomplete evidence.

🤖 Automated content by GitHub Copilot. Generated by the Pipeline Test Triage workflow. · copilot · auto · ~36 AIC · ⌖ ~1 AIC · ⊞ 10.1K

[github-actions]

🧪 Test quality review — PR #12184

Compared merge base d3ce600610baaebac245acdbd4162615f5dfcaae with head b485c8b0653c7807712d40b6b2284eceabf377ec. Evidence applies to this head, not every intermediate commit.

Test-value evidence

This review is static. It did not run code or tests.

• Attachment staging and manifests (GitHubActionsAttachmentCollectorTests, e.g. CaptureAsync_StagesExactBytesAndWritesManifest): static prediction that removing or corrupting staging or manifest output would fail the exact-bytes and manifest assertions. This is unverified.

• Non-regular source rejection (CaptureAsync_NonRegularSource_IsRejectedBeforeOpening, SessionFinishing_LinuxFifo_RejectsBeforeOpenAndPreservesPrimarySummary): static prediction that removing the regular-file guard trips the Times.Never open assertions. The PR description claims an executed red/green comparison (2 failed without the guard, 2 passed with it). I did not verify it, and no matching run record was accessible to this review.

• Reporter integration (GitHubActionsSummaryReporterAttachmentTests): static prediction that dropping attachment metadata would fail the rendering and aggregation assertions. This is unverified.

• CLI option and help text (GitHubActionsCommandLineProviderTests, HelpInfoAllExtensionsTests): static prediction that option validation and help expectations would fail if the option changed. This is unverified.

• Not applicable or not assessed:

  • The workflow, the script and the docs are not executable test contracts.
  • Resource .xlf files and the InternalAPI text files are not assessed.
  • The small edits to AzureDevOpsLivePublishingTests and TrxResultStreamingStoreTests are not assessed.
  • The new SystemFileSystem.Unix.cs stat P/Invoke is covered only by the two Linux FIFO tests. Those tests are not run on other platforms.

🧵 Parallel safety

STATUS: PARTIAL. The parallel-safety-reviewer specialist was not run in this session, so this is not a clean result.

A quick manual scan of the new unit-test classes found per-instance Guid temp directories and no static mutable state. GitHubActionsAttachmentCollectorTests implements IDisposable for cleanup. I found no cross-test hazard. The acceptance tests in GitHubActionsAttachmentTests.cs were not analyzed.

Per-test assessment

Per-test grading was not performed. About 40 new test methods were identified across GitHubActionsAttachmentCollectorTests, GitHubActionsSummaryReporterAttachmentTests and GitHubActionsAttachmentTests. I did not grade or pseudo-mutate them individually, so I cannot assign a result or quality to any of them.

Coverage gaps:

• No per-method grades.
• Existing-thread deduplication was only partly done. I read the earlier automated review and found no prior report from this workflow for this head.
• No inline findings were verified, so none are published.

<sub>This advisory review is heuristic and non-blocking.

Re-run the combined review with /review-tests, or only the parallel-safety specialist with /parallel-audit.</sub>

🤖 Automated content by GitHub Copilot. Generated by the Test Quality Review on PR (on open / sync) workflow. · copilot · auto · ~31 AIC · ⌖ ~1 AIC · ⊞ 8K