Add management SDK breaking-change attribution

[Yuchao Yan (msyyc)]

Summary

Extend the management SDK PR-review agentic workflow to explain newly added or modified changelog breaking-change entries using direct TypeSpec/API evidence, leaving entries without clear evidence to human reviewers. Emitter/toolchain attribution is out of scope.

• Add a deterministic Python collector and focused tests for changelog deltas, generation provenance, inferred release baselines, and incomplete evidence.
• Use one collector, fetching executable code from the trusted PR base SHA, never from the PR head.
• Add an advisory attribution table with cause, evidence, and confidence, separate from blocking review findings.
• Preserve first-versus-latest PR commit API-version drift semantics and the single self-updating review comment.

Validation Performed

• 27 focused tests passed, including compiler/setup-action version alignment, versioned Unreleased headings, and date-only release corrections, missing changelog baselines, the 250-commit endpoint cap, request-budget handoffs, missing/invalid apiVersion diagnostics, and source/generated bootstrap syntax checks.
• Workflow compilation using the checksum-verified gh-aw v0.87.1 binary (matching the PR base), with compile mgmt-sdk-pr-review --action-mode action --action-tag v0.87.1 --validate, passed.
• Exercised the updated collector against test PR [TEST ONLY] Duplicate azure-mgmt-deviceregistry generation #48959 #48993: complete discovery, six introduced breaking-change entries, and unchanged first-versus-latest API version.
• git diff --check passed.

Known Issues and Limitations

• The duplicate collector and bootstrap indentation defect are fixed. The single trusted bootstrap and generated lock are covered by syntax regression tests. No end-to-end GitHub Actions run has been performed.
• Previous-release selection is inferred from changelog headings, not confirmed generator comparison metadata.
• Agent investigation limits are prompt-enforced; no controlled SDK regeneration is performed.
• Trusted workflow/collector changes must be deployed to the appropriate trusted branch before end-to-end testing with pull_request_target.

PR Source

Replacement for fork-based draft #48991, with the same implementation published from Azure/azure-sdk-for-python:copilot/mgmt-breaking-change-attribution. The earlier PR remains open.

• Verified all 56 generated runtime-script references exist in pinned gh-aw-actions setup v0.87.1 (423b3dc04bbf1b1797194a4a75aa5cf5d0d4f5b3).

[azure-pipelines]

Azure Pipelines:
Successfully started running 1 pipeline(s).
9 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.

[azure-pipelines]

Azure Pipelines:
Successfully started running 1 pipeline(s).
9 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.

[Copilot]

🟡 Changes recommended

The workflow currently has fatal Python indentation errors, redundant collection, a toolchain downgrade, and attribution scope inconsistent with its description.

Get a fresh assessment by requesting another Copilot review.

Pull request overview

Extends the management SDK reviewer with breaking-change attribution and immutable provenance collection.

Changes:

• Adds changelog and provenance collection.
• Adds focused unit tests.
• Adds an advisory attribution table to review comments.

File summaries

File	Description
.github/workflows/scripts/mgmt_sdk_review_context.py	Collects changelog and provenance evidence.
.github/workflows/tests/test_mgmt_sdk_review_context.py	Tests parsing and failure handling.
.github/workflows/mgmt-sdk-pr-review.md	Adds collection and attribution instructions.
.github/workflows/mgmt-sdk-pr-review.lock.yml	Contains the generated workflow.

Review details

Suppressed comments (2)

.github/workflows/mgmt-sdk-pr-review.md:291

• The inline Python cannot execute because these lines have inconsistent top-level indentation (with, encoded_content, and content are unexpectedly indented, while payload is outside the with block). This makes every workflow run fail before the trusted collector is launched; fix the indentation and regenerate the lock file.

            with urllib.request.urlopen(request, timeout=30) as response:
          payload = json.load(response)
            encoded_content = re.sub(r"\s+", "", payload["content"])
            content = base64.b64decode(encoded_content, validate=True)

.github/workflows/mgmt-sdk-pr-review.lock.yml:474

• The generated workflow embeds the same malformed indentation as the source bootstrap, so the executable lock file raises IndentationError before creating review-context.json. After correcting the markdown source, regenerate this lock file so the deployed workflow contains valid Python.

        run: "python - <<'PY'\nimport base64\nimport json\nimport os\nimport pathlib\nimport re\nimport urllib.parse\nimport urllib.request\n\nrepository = os.environ[\"GH_REPOSITORY\"]\nrevision = os.environ[\"TRUSTED_BASE_SHA\"]\nif not re.fullmatch(r\"[A-Za-z0-9_.-]+/[A-Za-z0-9_.-]+\", repository):\n    raise SystemExit(\"Invalid repository reference\")\nif not re.fullmatch(r\"[0-9a-f]{40}\", revision):\n    raise SystemExit(\"Invalid trusted base revision\")\npath = \".github/workflows/scripts/mgmt_sdk_review_context.py\"\nurl = (\n    f\"https://api.github.com/repos/{repository}/contents/\"\n    f\"{urllib.parse.quote(path, safe='/')}?ref={revision}\"\n)\nrequest = urllib.request.Request(\n    url,\n    headers={\n        \"Accept\": \"application/vnd.github+json\",\n        \"Authorization\": f\"Bearer {os.environ['GH_TOKEN']}\",\n        \"User-Agent\": \"azure-sdk-python-mgmt-review\",\n        \"X-GitHub-Api-Version\": \"2022-11-28\",\n    },\n)\n      with urllib.request.urlopen(request, timeout=30) as response:\n    payload = json.load(response)\n      encoded_content = re.sub(r\"\\s+\", \"\", payload[\"content\"])\n      content = base64.b64decode(encoded_content, validate=True)\nif len(content) > 128 * 1024:\n    raise SystemExit(\"Trusted collector exceeded the size limit\")\nscript = pathlib.Path(\"mgmt_sdk_review_context.py\")\nscript.write_bytes(content)\nPY\npython mgmt_sdk_review_context.py\n"

• Files reviewed: 4/4 changed files
• Comments generated: 4
• Review effort level: Balanced

[Copilot]

🟡 Changes recommended

The collector can misclassify changelog history and fail to report incomplete evidence accurately.

Get a fresh assessment by requesting another Copilot review.

Review details

• Files reviewed: 4/4 changed files
• Comments generated: 4
• Review effort level: Balanced

[Copilot]

🟡 Changes recommended

The lock pins an incompatible setup action, and versioned Unreleased headings select the wrong release baseline.

Get a fresh assessment by requesting another Copilot review.

Review details

• Files reviewed: 4/4 changed files
• Comments generated: 2
• Review effort level: Balanced

[Copilot]

🔵 Needs a closer look

Malformed PR-controlled repository metadata can crash the collector before it produces a review context.

Review details

Suppressed comments (2)

Previously missed (1) — in code that hasn't changed since the last review.

.github/workflows/scripts/mgmt_sdk_review_context.py:372

• repository_url comes from PR-controlled _metadata.json. When it is a truthy non-string (for example, an object), re.fullmatch raises TypeError, aborting the collector before it writes review-context.json. Validate the value and record invalid provenance instead of executing the regex.

This issue also appears on line 459 of the same file.

.github/workflows/scripts/mgmt_sdk_review_context.py:463

• This second regex path also accepts the PR-controlled repository_url without checking its type. If tsp-location.yaml is absent, the earlier conflict comparison does not run, so a truthy non-string reaches this call and crashes attribution collection. Return unverified for a non-string value before matching it.

    if repository_url:
        match = re.fullmatch(
            r"https://github\.com/([A-Za-z0-9_.-]+/[A-Za-z0-9_.-]+?)(?:\.git)?/?", repository_url
        )
        repository = match.group(1) if match else None

• Files reviewed: 5/5 changed files
• Comments generated: 0 new
• Review effort level: Balanced