You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Declarative workflow scope aliases should retain the latest assignment consistently
when state is read, initialized, cleared, or restored.
Description & Review Guide
What are the major changes? Canonicalize managed scope names before persisted
state updates, native reads, initialization, and key enumeration. Reuse the
existing scope-name resolver and add round-trip and state-access coverage.
Migrate unambiguous legacy Topic entries identified by declarative metadata
and delete only those entries from the old scope. Reject overlapping entries
before migration writes, and retry failed
restores without skipping validation. Share the existing portable decoder so
migration preserves native storage types rather than coercing numbers.
Publish restoration updates before post-load hooks or resumed handlers run.
Register interpreter executors for the same shared restoration phase used by
Kit executors. Canonicalize aliases in the plain-context sensitivity helpers,
including state subsequently accessed through expression evaluation.
Without a formula-state restoration callback, plain helper reads fall back
to the requested metadata-marked legacy entry without modifying it; writes
remove only that entry after queuing its canonical value and metadata.
Ad hoc expression loading uses a read-only formula snapshot instead of queuing
migration from each reader. Share the collision preflight between plain and
wrapped assignments, rejecting ambiguous visible same-key Topic entries
before any value, sidecar, or live formula update. Record separate, versioned
canonical ownership with Local assignments and migrated values, so concurrent
raw Topic and canonical Local writes do not misclassify unrelated state.
Use key-level PortableValue probes instead of enumerating scopes for each
point operation. Plain helper null updates remove sensitivity and ownership
metadata together with the value. Preserve both Local and Topic formula names
when binding the shared canonical state.
Document the storage decision in proposed ADR 0048.
If callbacks or restoration publication fail, discard queued updates and
re-import the checkpoint's persisted state before propagating the original
error. Index retrieval and post-load hook failures remain outside that rollback
boundary. Once both restore tasks finish, those later failures discard
unpublished hook updates without reverting successfully published state.
What is the impact of these changes? Equivalent scope names address the same
persisted state. Custom scopes keep their existing behavior; omitted declarative
scopes use Local.
Older checkpoints with an identified declarative key in both physical scopes
and no canonical ownership metadata require explicit reconciliation or a fresh
session instead of an inferred assignment order.
Unmarked Topic entries remain untouched, preserving code-first state in mixed
workflows. Unmarked legacy declarative entries require explicit reconciliation.
Immediate repeated restores do not leave pending migration updates behind.
Failed restoration publication can be retried without retaining partially
applied writes. A later index or hook failure does not undo successful
publication and leave it inconsistent with live executor state.
Queued hook updates from a failed restore do not survive into an immediate retry.
Fan-out expression readers do not queue competing migration writes. Plain
helper clears do not leave metadata that could claim a later code-first key.
New canonical ownership permits independently owned raw and canonical values
to coexist after fan-out publication and restoration; older ambiguous entries
still reject. Point-helper lookup work stays bounded as pending writes grow.
Public API signatures are unchanged.
What do you want reviewers to focus on? Consistency between live formula
state and persisted state, including alias overwrite and clear operations.
Before ownership tracking and key-level probes: failed, all 16 cases.
Eight mixed fan-out cases became unreadable, two migrated/raw-reuse cases
failed restoration, four updates/clears rejected independently owned state,
and two point-operation cases each enumerated full scopes 64 times.
After the correction: passed, all 20 cases, including four added
unsupported-version cases. Coverage includes both execution modes,
plain/wrapped assignments, explicit None/Sensitive metadata, fresh JSON restore,
expression aliases, migration, clear/reuse, and bounded point lookups.
Before the correction: failed, all 14 cases. Four fan-out readers queued
conflicting migration writes, six plain/wrapped collision writes were
accepted, and four clear/key-reuse cases claimed unrelated code-first values.
After the correction: passed, all 14 cases with unchanged assertions.
Previous read-only helper coverage now expects expression-only reads to
retain legacy storage; explicit writes and registered pre-hook restoration
still perform cleanup.
Before targeted compatibility handling: failed, all 10 cases. Four reads
missed saved values, four writes left overlapping entries, and two ambiguous
reads/writes did not reject the overlap.
After the correction: passed, all 10 cases with unchanged assertions.
Before plain-context normalization: failed, all 8 cases. Four read null
after migration and four encountered overlapping entries on another
restoration or expression evaluation.
After the correction: passed, all 8 cases with unchanged assertions.
Before the follow-up correction: failed, with 4 failed and 20 passed cases.
Directly seeded legacy values were not restored or cleaned up, and overlapping
entries were not rejected.
After the follow-up correction: passed, all 24 cases, including the original
alias-assignment regression and repeated conflict rejection.
dotnet test --project tests\Microsoft.Agents.AI.Workflows.Declarative.UnitTests --no-restore --artifacts-path $artifacts -p:CustomBeforeDirectoryBuildProps=$outputProps --no-progress
Passed: 2,498 tests across net10.0 and net472; no failures or skips
in the final run. An earlier run encountered an existing off-thread timing
failure in RestoreCheckpointAsync_ReloadsFormulaSessionOnEveryLoadAsync;
the same test failure was reproduced with implementation files restored to
the published 41207a505 revision before returning to the final implementation.
Before the publication barrier: failed, all 6 cases. Four encountered
conflicting updates at publication, and two observed missing migrated state
in another executor's post-load hook.
After the barrier: passed, all 6 cases with unchanged assertions.
Passed: all 36 cases, including real-state-manager resumes, assignment,
clearing, post-load visibility, immediate repeated restores, and preservation
of unrelated code-first entries.
dotnet test --project tests\Microsoft.Agents.AI.Workflows.UnitTests --no-build --no-restore --artifacts-path $artifacts -p:CustomBeforeDirectoryBuildProps=$outputProps --no-progress
Passed: 1,753 tests across net10.0 and net472; no failures or skips.
Failed on the second baseline repetition: the off-thread resumed writer
changed live state before the test's immediate assertions. Implementation
equality with 41207a505 was verified before testing. The existing test was
not weakened or changed.
Passed: full solution, zero warnings and errors.
After the initial build exhausted the repository drive and was stopped, $artifacts was placed on another drive and $outputProps supplied unique
artifact directory names for projects with duplicate names. These are
validation-only settings; no repository build configuration changed.
Failed: 9,908 passed, 62 failed, 76 skipped. The temporary solution was
generated with the repository's New-FilteredSolution.ps1, selecting net10.0 unit tests and excluding samples. Failures are outside the changed
projects: 61 AgentHooks tests could not load agent_hooks_ffi; one
FileAgentSkillLoader test failed during Windows symlink cleanup with UnauthorizedAccessException. The exact failed test names match the previous
published revision. Both affected projects passed in this run.
Integration tests requiring external services were not run.
Related Issue
N/A - maintainer-directed bugfix; no new issue was opened.
AI Assistance
No material AI assistance was used.
This is an AI-assisted contribution. I reviewed, understood, and verified all submitted content and accept responsibility for it.
AI assistance details: Copilot assisted with implementation, regression tests,
validation, and this pull request description.
Contribution Checklist
The code builds clean without any errors or warnings
All unit tests pass, and I have added new tests where possible
This PR links to an agreed issue with no competing open PR, or the Related Issue section documents a trivial-change or repository-automation exception.
This is not a breaking change. If it is a breaking change, add the breaking change label (or add "[BREAKING]" to the title prefix, before or after any language prefix) — a workflow keeps the label and title prefix in sync automatically.
The reason will be displayed to describe this comment to others. Learn more.
MAF Automated Review — Iteration 1
Result: Findings reported Scope: full PR (1 commit(s)): 0313abbb6155 Model:gpt-5.6-sol
Overview
The change consistently canonicalizes new Topic state access and persistence to Local, with focused coverage for reads, initialization, overwrite, sensitivity, clear, and restore behavior. Custom and null scopes remain pass-through, and managed-scope restrictions remain intact. However, persisted checkpoints created by earlier versions can contain physical Topic entries, and the new read redirection makes those entries unreachable without migration compatibility.
Reviewed the supplied pull-request change set across correctness, security/reliability, architecture, and failure behavior.
1 verified finding remained after source verification (1 medium) across 1 file. Details are attached to the affected lines below.
The change consistently aliases managed Topic access to Local, adds explicit legacy migration and conflict rejection, preserves native portable types, and resets initialization after failed restores. The pre-write conflict check and decode-before-write staging are useful guardrails. Two migration lifecycle gaps remain: the migration assumes ownership of the run-wide shared Topic scope, and its queued updates make an otherwise supported immediate subsequent restore fail.
Reviewed the supplied incremental change set across correctness, security/reliability, architecture, and failure behavior.
2 verified findings remained after source verification (2 medium) across 1 file. Details are attached to the affected lines below.
This incremental revision adds plain-context legacy Topic fallback and cleanup, and discards unpublished post-load writes after a later restore failure. The restore cleanup is narrowly scoped and backed by lockstep/off-thread retry coverage, but the new fallback relies on sensitivity-key presence as an ownership marker. A stale sidecar can therefore reclassify reused code-first Topic state, causing incorrect sensitivity propagation or deletion.
Reviewed the supplied incremental change set across correctness, security/reliability, architecture, and failure behavior.
1 verified finding remained after source verification (1 medium) across 1 file. Details are attached to the affected lines below.
The reason will be displayed to describe this comment to others. Learn more.
🟡 Changes recommended
The collision preflight misses concurrent Topic.K/Local.K writes, allowing conflicting state to be published and causing subsequent reads, expression evaluation, and checkpoint restoration to fail.
Even when Topic is empty, this enumeration scans every pending update: StateManager.ReadKeysAsync calls ApplyUnpublishedUpdates, whose GetUpdatesForScopeStrict filters the entire update dictionary. A plain executor queuing N sensitivity-aware Local assignments now scans 0 + 2 + … + 2(N−1) updates before publication, making the batch quadratic instead of linear. Please use key-level ReadStateAsync<PortableValue> existence probes (including for the metadata and Local checks below), or an indexed lookup, rather than enumerating scopes for each point operation.
The reason will be displayed to describe this comment to others. Learn more.
MAF Automated Review — Iteration 7
Result: No findings Scope: 1 net-new commit(s): 3bdf7411b1ff Model:gpt-5.6-sol
Overview
This increment makes ad hoc expression restoration read-only, adds preflight collision checks before canonical writes, and removes sensitivity metadata when the plain helper clears a value. The new tests cover parallel expression readers, unmarked Topic collisions, and code-first key reuse after helper clears. No Critical, High, or Medium defect introduced by this review range remained after comparison with the previous head and existing feedback.
Reviewed the supplied incremental change set across correctness, security/reliability, architecture, and failure behavior.
No publishable findings remained after source verification for this scope.
Addressed the additional Copilot review-summary feedback about quadratic pending-update scans in 8ca1b34. Point operations now use direct PortableValue probes for Topic, ownership, sensitivity metadata, and Local instead of enumerating scopes. The plain/wrapped regression cases each perform 32 write/read pairs: full-scope enumerations dropped from 64 to zero, with bounded key-probe counts. Whole-state restoration and read-only snapshots retain their necessary scope enumeration.
Acknowledged the latest Copilot review on 8ca1b34: there are no open code findings, and the durable-storage compatibility policy needs maintainer review. Proposed ADR 0048 explicitly covers mixed-runtime limitations and preserving the private metadata scopes in custom checkpoint stores. The requested human decider, westey-m, was added as a reviewer at 2026-10-28T00:07:57Z. Human review remains pending; the ADR will stay proposed until the decider approves it.
The reason will be displayed to describe this comment to others. Learn more.
MAF Automated Review — Iteration 8
Result: No findings Scope: 1 net-new commit(s): 8ca1b347d308 Model:gpt-5.6-sol
Overview
This increment adds versioned canonical ownership metadata, replaces full-scope point probes with key-level reads, and preserves separate raw Topic and canonical Local values across publication and restore. The strongest guardrails are the shared ownership write/read choke points, fail-closed version handling, read-only snapshot restoration, and broad mixed fan-out, migration, clear, and fresh-checkpoint tests. After reconciling the proposed candidates against the previous head, supplied diff, and existing feedback, no Critical, High, or Medium defect was established as introduced by this review range.
Reviewed the supplied incremental change set across correctness, security/reliability, architecture, and failure behavior.
No publishable findings remained after source verification for this scope.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
documentationUsage: [Issues, PRs], Target: documentation in the code base and learn docs.NETUsage: [Issues, PRs], Target: .NetworkflowsUsage: [Issues, PRs], Target: Workflows
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
GitHub Actions
Inactive
agent-framework-automation
Bot
added
.NET
github-actions
Bot
changed the title
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
![github-code-quality[bot]](https://avatars.githubusercontent.com/u/9919?s=60&v=4){kind=small}
Motivation & Context
Declarative workflow scope aliases should retain the latest assignment consistently
when state is read, initialized, cleared, or restored.
Description & Review Guide
state updates, native reads, initialization, and key enumeration. Reuse the
existing scope-name resolver and add round-trip and state-access coverage.
Migrate unambiguous legacy
Topicentries identified by declarative metadataand delete only those entries from the old scope. Reject overlapping entries
before migration writes, and retry failed
restores without skipping validation. Share the existing portable decoder so
migration preserves native storage types rather than coercing numbers.
Publish restoration updates before post-load hooks or resumed handlers run.
Register interpreter executors for the same shared restoration phase used by
Kit executors. Canonicalize aliases in the plain-context sensitivity helpers,
including state subsequently accessed through expression evaluation.
Without a formula-state restoration callback, plain helper reads fall back
to the requested metadata-marked legacy entry without modifying it; writes
remove only that entry after queuing its canonical value and metadata.
Ad hoc expression loading uses a read-only formula snapshot instead of queuing
migration from each reader. Share the collision preflight between plain and
wrapped assignments, rejecting ambiguous visible same-key
Topicentriesbefore any value, sidecar, or live formula update. Record separate, versioned
canonical ownership with Local assignments and migrated values, so concurrent
raw Topic and canonical Local writes do not misclassify unrelated state.
Use key-level PortableValue probes instead of enumerating scopes for each
point operation. Plain helper null updates remove sensitivity and ownership
metadata together with the value. Preserve both Local and Topic formula names
when binding the shared canonical state.
Document the storage decision in proposed ADR 0048.
If callbacks or restoration publication fail, discard queued updates and
re-import the checkpoint's persisted state before propagating the original
error. Index retrieval and post-load hook failures remain outside that rollback
boundary. Once both restore tasks finish, those later failures discard
unpublished hook updates without reverting successfully published state.
persisted state. Custom scopes keep their existing behavior; omitted declarative
scopes use
Local.Older checkpoints with an identified declarative key in both physical scopes
and no canonical ownership metadata require explicit reconciliation or a fresh
session instead of an inferred assignment order.
Unmarked
Topicentries remain untouched, preserving code-first state in mixedworkflows. Unmarked legacy declarative entries require explicit reconciliation.
Immediate repeated restores do not leave pending migration updates behind.
Failed restoration publication can be retried without retaining partially
applied writes. A later index or hook failure does not undo successful
publication and leave it inconsistent with live executor state.
Queued hook updates from a failed restore do not survive into an immediate retry.
Fan-out expression readers do not queue competing migration writes. Plain
helper clears do not leave metadata that could claim a later code-first key.
New canonical ownership permits independently owned raw and canonical values
to coexist after fan-out publication and restoration; older ambiguous entries
still reject. Point-helper lookup work stays bounded as pending writes grow.
Public API signatures are unchanged.
state and persisted state, including alias overwrite and clear operations.
Commands were run from
dotnet/.dotnet test --project tests\Microsoft.Agents.AI.Workflows.Declarative.UnitTests -f net10.0 --no-restore --artifacts-path $artifacts -p:CustomBeforeDirectoryBuildProps=$outputProps --filter-method '*CanonicalOwnership*' --no-progressEight mixed fan-out cases became unreadable, two migrated/raw-reuse cases
failed restoration, four updates/clears rejected independently owned state,
and two point-operation cases each enumerated full scopes 64 times.
unsupported-version cases. Coverage includes both execution modes,
plain/wrapped assignments, explicit None/Sensitive metadata, fresh JSON restore,
expression aliases, migration, clear/reuse, and bounded point lookups.
dotnet test --project tests\Microsoft.Agents.AI.Workflows.Declarative.UnitTests -f net10.0 --no-restore --artifacts-path $artifacts -p:CustomBeforeDirectoryBuildProps=$outputProps --no-progressdotnet test --project tests\Microsoft.Agents.AI.Workflows.Declarative.UnitTests -f net10.0 --no-restore --artifacts-path $artifacts -p:CustomBeforeDirectoryBuildProps=$outputProps --filter-method '*LegacyBoundary*' --no-progressconflicting migration writes, six plain/wrapped collision writes were
accepted, and four clear/key-reuse cases claimed unrelated code-first values.
Previous read-only helper coverage now expects expression-only reads to
retain legacy storage; explicit writes and registered pre-hook restoration
still perform cleanup.
dotnet test --project tests\Microsoft.Agents.AI.Workflows.Declarative.UnitTests -f net10.0 --no-restore --artifacts-path $artifacts -p:CustomBeforeDirectoryBuildProps=$outputProps --filter-method '*PlainLegacyHelpers*' --no-progressmissed saved values, four writes left overlapping entries, and two ambiguous
reads/writes did not reject the overlap.
dotnet test --project tests\Microsoft.Agents.AI.Workflows.UnitTests -f net10.0 --no-restore --artifacts-path $artifacts -p:CustomBeforeDirectoryBuildProps=$outputProps --filter-method '*Checkpoint_Restore_LaterFailureDiscardsHookUpdatesAsync' --no-progressdotnet test --project tests\Microsoft.Agents.AI.Workflows.Declarative.UnitTests -f net10.0 --no-restore --artifacts-path $artifacts -p:CustomBeforeDirectoryBuildProps=$outputProps --filter-method '*PlainSensitivityAliasesRemainConsistentAsync' --no-progressafter migration and four encountered overlapping entries on another
restoration or expression evaluation.
dotnet test --project tests\Microsoft.Agents.AI.Workflows.UnitTests -f net10.0 --no-restore --artifacts-path $artifacts -p:CustomBeforeDirectoryBuildProps=$outputProps --filter-method '*Checkpoint_Restore_LaterFailurePreservesPublishedStateAsync' --no-progressfailures reverted persisted values while live executor state stayed restored.
dotnet test --project tests\Microsoft.Agents.AI.Workflows.Declarative.UnitTests -f net10.0 --no-restore --artifacts-path $artifacts -p:CustomBeforeDirectoryBuildProps=$outputProps --filter-method '*ResumeAsync_LegacyCheckpoint*' --no-progressassignments encountered conflicting updates; two immediate repeated restores
rejected pending migration writes.
dotnet test --project tests\Microsoft.Agents.AI.Workflows.UnitTests -f net10.0 --no-restore --artifacts-path $artifacts -p:CustomBeforeDirectoryBuildProps=$outputProps --filter-method '*Checkpoint_Restore_FailedPublicationCanBeRetriedAsync' --no-progresswrites after failure; two subsequent restores rejected queued updates.
dotnet test --project tests\Microsoft.Agents.AI.Workflows.Declarative.UnitTests -f net10.0 --filter-method '*RestoreAsync_PreservesLatestAliasAssignmentAsync' --no-progressrestored state returned the earlier value instead of the latest assignment.
dotnet test --project tests\Microsoft.Agents.AI.Workflows.Declarative.UnitTests -f net10.0 --no-restore --artifacts-path $artifacts -p:CustomBeforeDirectoryBuildProps=$outputProps --filter-class '*WorkflowFormulaStateTests' --no-progressDirectly seeded legacy values were not restored or cleaned up, and overlapping
entries were not rejected.
alias-assignment regression and repeated conflict rejection.
dotnet test --project tests\Microsoft.Agents.AI.Workflows.Declarative.UnitTests --no-restore --artifacts-path $artifacts -p:CustomBeforeDirectoryBuildProps=$outputProps --no-progressnet10.0andnet472; no failures or skipsin the final run. An earlier run encountered an existing off-thread timing
failure in
RestoreCheckpointAsync_ReloadsFormulaSessionOnEveryLoadAsync;the same test failure was reproduced with implementation files restored to
the published
41207a505revision before returning to the final implementation.dotnet test --project tests\Microsoft.Agents.AI.Workflows.Declarative.UnitTests -f net10.0 --no-restore --artifacts-path $artifacts -p:CustomBeforeDirectoryBuildProps=$outputProps --filter-method '*RestoreAsync_LegacyMigrationPreservesNativeTypesAsync' --no-progresspassed case because migrated numeric storage types changed.
dotnet format src\Microsoft.Agents.AI.Workflows.Declarative\Microsoft.Agents.AI.Workflows.Declarative.csproj --no-restore --verbosity quietdotnet format src\Microsoft.Agents.AI.Workflows\Microsoft.Agents.AI.Workflows.csproj --no-restore --verbosity quietdotnet format tests\Microsoft.Agents.AI.Workflows.Declarative.UnitTests\Microsoft.Agents.AI.Workflows.Declarative.UnitTests.csproj --no-restore --verbosity quietdotnet format tests\Microsoft.Agents.AI.Workflows.UnitTests\Microsoft.Agents.AI.Workflows.UnitTests.csproj --no-restore --verbosity quietdotnet test --project tests\Microsoft.Agents.AI.Workflows.Declarative.UnitTests -f net10.0 --no-restore --artifacts-path $artifacts -p:CustomBeforeDirectoryBuildProps=$outputProps --filter-method '*ResumeAsync_PublishesLegacyMigration*' --no-progressconflicting updates at publication, and two observed missing migrated state
in another executor's post-load hook.
dotnet test --project tests\Microsoft.Agents.AI.Workflows.Declarative.UnitTests -f net10.0 --no-restore --artifacts-path $artifacts -p:CustomBeforeDirectoryBuildProps=$outputProps --filter-method '*ResumeAsync_PreservesUnmarkedTopicEntriesAsync' --no-progressunrelated unmarked entries as conflicting aliases.
dotnet test --project tests\Microsoft.Agents.AI.Workflows.Declarative.UnitTests -f net10.0 --no-restore --artifacts-path $artifacts -p:CustomBeforeDirectoryBuildProps=$outputProps --filter-class '*WorkflowFormulaStateTests' --no-progressclearing, post-load visibility, immediate repeated restores, and preservation
of unrelated code-first entries.
dotnet test --project tests\Microsoft.Agents.AI.Workflows.UnitTests --no-build --no-restore --artifacts-path $artifacts -p:CustomBeforeDirectoryBuildProps=$outputProps --no-progressnet10.0andnet472; no failures or skips.dotnet test --project tests\Microsoft.Agents.AI.Workflows.Declarative.UnitTests -f net472 --no-build --no-restore --artifacts-path $artifacts -p:CustomBeforeDirectoryBuildProps=$outputProps --filter-method '*RestoreCheckpointAsync_ReloadsFormulaSessionOnEveryLoadAsync' --no-progresschanged live state before the test's immediate assertions. Implementation
equality with
41207a505was verified before testing. The existing test wasnot weakened or changed.
dotnet build agent-framework-dotnet.slnx --disable-build-servers -nr:false --artifacts-path $artifacts -p:CustomBeforeDirectoryBuildProps=$outputProps --no-restore --tl:off --verbosity minimal -m:2After the initial build exhausted the repository drive and was stopped,
$artifactswas placed on another drive and$outputPropssupplied uniqueartifact directory names for projects with duplicate names. These are
validation-only settings; no repository build configuration changed.
dotnet test --solution alias-unit-validation.slnx -f net10.0 --no-build --no-restore --artifacts-path $artifacts -p:CustomBeforeDirectoryBuildProps=$outputProps --no-progress --max-parallel-test-modules 2 --ignore-exit-code 8generated with the repository's
New-FilteredSolution.ps1, selectingnet10.0unit tests and excluding samples. Failures are outside the changedprojects: 61 AgentHooks tests could not load
agent_hooks_ffi; oneFileAgentSkillLoader test failed during Windows symlink cleanup with
UnauthorizedAccessException. The exact failed test names match the previouspublished revision. Both affected projects passed in this run.
Integration tests requiring external services were not run.
Related Issue
N/A - maintainer-directed bugfix; no new issue was opened.
AI Assistance
AI assistance details: Copilot assisted with implementation, regression tests,
validation, and this pull request description.
Contribution Checklist
breaking changelabel (or add "[BREAKING]" to the title prefix, before or after any language prefix) — a workflow keeps the label and title prefix in sync automatically.