test: add Microsoft Agent Framework VCR coverage

[sipercai]

Description

This PR adds a test-first, replay-only VCR contract for the Microsoft Agent Framework instrumentation and fixes an oldest-supported-version compatibility gap.

• exercises the real MAF framework and OpenAI-compatible client across content-disabled/content-enabled non-streaming, fully consumed streaming, deferred finalization, tool round-trip, provider error, and concurrent calls;
• includes five bounded DashScope provider recordings as the real-provider baseline and two deterministic local OpenAI-compatible supplements for tool and error behavior;
• defaults committed replay to record_mode=none, fake credentials, strict request matching, and sanitized cassette headers/identifiers;
• adds reproducible oldest/latest dependency lanes and the generated GitHub Actions jobs;
• calls MAF's enable_instrumentation(force=True) only when the installed MAF version exposes the force parameter.

The framework/client packages are test-only dependencies; runtime package metadata is unchanged. Commercial probe packaging and Robin-only behavior are outside this PR.

Fixes # (N/A)

Type of change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• This change requires a documentation update

How Has This Been Tested?

• tox -e precommit
• uvx --from 'tox>=4,<5' tox -e generate-workflows
• uvx --from 'tox>=4,<5' tox -c tox-loongsuite.ini -e py312-test-loongsuite-instrumentation-microsoft-agent-framework-oldest,py312-test-loongsuite-instrumentation-microsoft-agent-framework-latest -- --vcr-record=none -q

Does This PR Require a Core Repo Change?

• Yes. - Link to PR:
• No.

Checklist:

See contributing.md for styleguide, changelog guidelines, and more.

• Followed the style guidelines of this project
• Changelogs have been updated
• Unit tests have been added
• Documentation has been updated

Validation Evidence

Spec and Scope

• Linked issue/spec: MAF-VCR-TDD Spec v1; no GitHub issue.
• Approved spec/comment: approved before implementation.
• Changed surface: MAF instrumentation startup compatibility, real-framework replay tests/cassettes, test dependency lanes, and generated CI matrix.

Local Checks

Check	Command	Result	Notes
Static readiness	LoongSuite check_loongsuite_pr_readiness.py --repo . --base origin/main --committed-only	pass	No errors or warnings.
Precommit	tox -e precommit with a fresh hook cache	pass	No files modified.
Focused tests	uvx --from 'tox>=4,<5' tox -c tox-loongsuite.ini -e py312-test-loongsuite-instrumentation-microsoft-agent-framework-oldest,py312-test-loongsuite-instrumentation-microsoft-agent-framework-latest -- --vcr-record=none -q	pass	79 tests passed in each lane.
Test-first red	Same replay tests against fd2cd94^	pass	Collection and cassette replay succeeded; all seven scenarios reached targeted telemetry assertions and failed on the pre-fix source.
VCR replay-only	Focused tox command above with --vcr-record=none	pass	Seven scenarios, fake credentials, no cassette misses or network fallback.
Independent review	Two qoder review rounds	pass	Final verdict LGTM; no blocking findings. Serializer versions were pinned after review.
VCR/privacy scan	Repository cassette privacy scan	pass	No credentials, local paths, usernames, auth/cookie/trace/request headers, or private keys found.

Business Isolation

Scenario	Status	Evidence
prepare/start fault	N/A	No business-call wrapper or prepare/start callback behavior changed.
response mapping/stop fault	N/A	No response-mapping or stop callback implementation changed.
fail callback fault	N/A	No failure callback implementation changed.
stream chunk/final callback fault	N/A	No stream callback wrapper changed; replay validates framework-owned finalization only.
early close/cancellation/GeneratorExit	N/A	MAF exposes deferred final-response retrieval, not provider transport cancellation, on this tested surface.
cross-task or cross-Context stream	N/A	No context token reset implementation changed.
clean sibling after fault	N/A	No instrumentation callback fault path changed.

Real E2E Matrix

Scenario	Status	Command or Demo	Evidence
non-streaming	pass	Real MAF + provider-recorded VCR replay	Content-disabled and content-enabled calls produce a closed AGENT/LLM trace.
streaming	pass	Real MAF + provider-recorded SSE replay	Fully consumed and deferred-finalization flows complete with expected content and TTFT where supported.
concurrency	pass	Two concurrent real-framework calls	Trace IDs, parents, and captured content remain isolated.
agent/tool/ReAct	pass	Deterministic local OpenAI-compatible tool replay	One AGENT, two LLM, and one TOOL span form one closed trace.
tool-heavy	N/A	Not part of the changed surface	The PR changes replay coverage and startup compatibility, not multi-tool orchestration.
error path	pass	Deterministic provider-500 replay	AGENT and LLM spans retain one trace and export ERROR status.

Telemetry and Weaver

Check	Status	Command or Artifact	Notes
Span tree / span kinds	pass	Seven replay scenarios	Closed AGENT roots with expected LLM/TOOL counts and no cross-trace parents.
Content capture modes	pass	Content-disabled and content-enabled replays	Sensitive message/system attributes are absent when disabled and present when enabled.
Concurrency isolation	pass	Two-call replay	No trace, parent, or content leakage.
Weaver live-check	pass	Actual in-memory-exported AGENT and LLM sample checked with the loongsuite-genai advice profile	No violations; only registry development-stability and provider-enum advisories.

CI

• GitHub checks: pending PR creation.
• Known unrelated failures: none.
• Follow-up needed: monitor the generated oldest/latest MAF jobs after publication.

[ralf0131]

⚠️ Merge conflict detected

This PR has conflicts with the main branch and cannot be merged. Please rebase or merge main into your branch and resolve the conflicts:

git fetch origin
git checkout test/maf-vcr-contract
git rebase origin/main
# resolve conflicts, then:
git push --force-with-lease

This is a one-time reminder. Feel free to @mention me for a re-review after conflicts are resolved.

---

Automated notification by github-manager-bot

[ralf0131]

Summary

This PR adds comprehensive VCR-based integration tests for the Microsoft Agent Framework instrumentation, covering 8 scenarios (non-streaming, streaming, tools, errors, concurrency, content modes). It also includes a clean backwards-compatibility fix for MAF 1.0 by using inspect.signature() to conditionally pass the force parameter to enable_instrumentation().

The code is well-structured, test coverage is thorough, and the compatibility fix is minimal and safe. LGTM.

Findings

• [Info] conftest.py — record_mode: "none" is the right default for CI safety. Contributors re-recording cassettes locally will need to explicitly set record_mode="once" — consider adding a brief comment in the file header for discoverability.
• [Info] test_integration_vcr.py — The _assert_closed_maf_trace helper is a solid abstraction. It correctly verifies root span kind, child span count, and that all spans are ended. Consider also asserting status_code != StatusCode.UNSET for error-path tests to catch accidental status regressions.
• [Info] __init__.py — The inspect.signature() approach with try/except (TypeError, ValueError) fallback is robust. Clean fix for MAF 1.0 compatibility.

---

Automated review by github-manager-bot