[functional-test-alignment] Test supported documentation tooling behavior #239

Context and request

Align MSXOrg/docs tests with the existing Testing standard. Keep supported tool behavior and legitimate package quality. Remove assertions that freeze source structure, fixed skill inventories, or implementation-only call counts. Deliver one independent pull request without policy changes, merge, or auto-merge.

#240 now meets the review-readiness gates. It is not merged. The Task stays open for its normal merge closure. The user-approved narrow heading defect fix also resolves #241.

Acceptance criteria

• Every discovered suite, assertion group, runner, and validation entry point has a source-bound disposition, real entry point, observed result, and policy reason.
• Remove source-function extraction, fixed catalog-name/count assertions, and the pin updater's exact transport-call count. Do not add equivalent snapshots, regex source guards, copied schemas, or inventory tests.
• Preserve real checker, generator, updater, and PowerShell dependency behavior on controlled inputs, with independent expected results and demonstrated red-green checks.
• Retain format/static-analysis/link/package/build/pin/security controls. Do not invent a plugin validator, runtime, framework, or production feature to retain a source assertion.
• Keep production behavior unchanged except for the user-approved narrow heading-rendering fix. Preserve supported formatting and GitHub's literal underscore/attribute text.
• Keep the accepted removal of documentation-content suites and checked-in index-drift enforcement intact. Complete exact-head CI and full review, then hand off without merge automation.

Source identity and technical decisions

Both complete recursive tree responses report truncated=false and the same six suite files. The original 64 tests pass at the initial baseline. #237 merged into default-branch commit 1b0d489 during the audit and added two fixed catalog assertions. This branch was refreshed from that merged main commit with an ordinary merge; no open PR branch was imported. Its writing/skill/manifest changes remain baseline context, not this PR's delta.

The scoped candidate diff changes exactly four files: .github/scripts/Test-CrossRepositoryLink.ps1, tests/PluginMarketplace.Tests.ps1, tests/Test-CrossRepositoryLink.Tests.ps1, and tests/Update-GitHubActionPin.Tests.ps1. The executable production change stays inside the existing heading-rendering helper; related help and source-provenance text match that change. No workflow, pin, security setting, policy page, unrelated script, or repository notes file changed. The worktree is clean.

The user approved the narrow heading fix after the real command exposed an underscore defect hidden by the AST-loaded unit cases. The final helper reuses PowerShell's bundled Markdig.Signed assembly with CommonMark plus GitHub autolinks/strikethrough, not site-only generic attributes. It does not install a parser or use generated renderer IDs. The existing GitHub slug algorithm and all 14 original expected anchors stay unchanged. PowerShell v7.0.0 ships Markdig.Signed 0.18.1; the selected public APIs were verified at immutable renderer source dd6418d108c6cb66f2196ad04fad5d59a231e72e. The PowerShell 7.0 minimum is unchanged. This is source/API compatibility evidence, not a claim that PowerShell 7.0 was executed.

The repository owns no plugin installer or separate production manifest validator. Retain four existing static package-quality checks, not a claim of runtime or complete schema coverage. Their public format basis is the package profile, marketplace contract, Agent Plugins manifest contract, and Agent Skills specification. A fixed skill-directory inventory does not exercise client discovery, so remove both such groups and their setup.

Implementation plan

• Verify repository instructions, live source identities, policy, complete tree inventory, and focused baselines.
• Remove the pin-updater call-count assertion, repeat the 4-test run, commit, and open the linked draft immediately.
• Audit every suite group and invoked implementation; classify package quality separately from behavioral coverage.
• Rewrite all 14 internal slug cases through the real checker command, preserve independent expected results, and prove the replacement checks fail.
• Apply the user-approved narrow heading fix with command-level regression coverage.
• Refresh from merged default-branch source and remove its two fixed catalog assertions without changing the product catalog.
• Read both non-clean review summaries fully; fix all 5 valid findings with red-green cases, specific replies, and resolved threads.
• Run focused and combined local tests, independent oracles, negative package controls, static analysis, and checkout link validation.
• Verify all final-head CI checks and meaningful executed-unit counts.
• Read the full exact-head approving review, complete the standards pass and scoped issue-convergence sweep, and verify all threads and native closing references.
• Publish the review-ready PR and deliver the final handoff without merge automation.

Evidence inventory

Validation and independent evidence

The actual github-slugger 2.0.0 implementation at immutable source 3461c4350868329c8530904d170358bca1d31448 independently verifies all 31 positive anchor expectations. Existing Node v26.12.1 was used; no package was installed and the oracle is not a new dependency. GitHub Markdown API mode=gfm separately confirmed visible heading text, including all review examples and literal generic attributes. Production renderer output is not used to compute expected anchors.

Red-green evidence is retained in the Task audit trail and #241. All 64 original tests passed, while the complete command rejected the valid underscore anchor. The first expanded command run exposed 6 failures in 39 cases. The first review's 7 examples fail before correction (42 pass/7 fail), then pass. Both generic-attribute acceptance/rejection cases fail before the syntax correction (50 pass/2 fail), then pass (52 pass/0 fail). An attempted name filter selected 0 cases and is explicitly excluded from evidence.

Anonymous package controls use the unchanged quality suite on temporary roots: a positive retained-field fixture gives exit 0 and 4 passes; duplicate identities give exit 1 with 1 expected failure. An empty catalog gives exit 1 through both the runner and direct Pester. No count guard or inventory assertion was added. Session-owned experiment directories were removed after verification; evidence remains in GitHub.

Full review disposition

The clean result is based on the full summary, exact source identity, and resolved threads, not only the runner exit or new-comment count. The native PR closing references were verified for this Task and the convergent user-approved Bug.

Standards, documentation, and convergence

The final alignment pass re-read the process, testing, PowerShell, and applicable format guidance. The changed surfaces are aligned with Testing, PowerShell Testing, Functions/reuse-before-build, Code Layout, Documentation, Natural Language, the package profile, and the actual checker/updater contracts. The PR contains one alignment row per changed file. .github/CONTRIBUTING.md and the optional generator -Check guidance already describe the retained commands correctly. Only help tied to the changed heading helper required an update. No policy prose or repository notes were added. The route builds and deploys Pages; no Release Management block or labels apply.

The scoped open-issue sweep covers tests, tooling, plugins, and anchors. It finds this Task, the fully delivered user-approved #241, and the unrelated policy PBI #105. No other issue converges. Published-URL, AGENTS-routing, navigation, and writing-campaign work remain outside this diff. No aggregate is closed.

Coverage limits and safety

Marketplace checks are partial static package checks, not complete schema validation, client installation, activation, routing, or trigger-reliability evidence. Release/API mocks and loopback responses prove caller behavior, not live authentication or every vendor failure mode. They permit safe deterministic quota exhaustion, missing targets, and unavailable responses. No live pin-updater API check or plugin-client installation test was run.

The existing read-only live cross-link CI job is separate evidence: Ubuntu 24.04, repository-scoped workflow token, public repository reads, no remote writes or resources to clean up. Tool pins, checks, security settings, and production behavior outside the approved helper remain unchanged. GitHub auto-merge and app Agent merge are off. No merge or release publication was performed.

Campaign qualification supplied by the coordinator

This is coordinator-observed evidence, not this Task's audit of those repositories: complete recursive Git trees (truncated=false) for MSXOrg/.github main at 0c67251812ea306543f3a14ab473ae357cfb76ac and PSModule/.github main at 4e0092de8f208b11e0d4607ced93471607aa748e contain community/agent guidance, issue forms, attributes/ignore configuration, and media, but no test code, runners, scripts, dependency manifests, hooks, or validation workflows. No empty PR is warranted for those snapshots. Their open branches, including PSModule/.github#89, were not imported. PSModule/Process-PSModule has a separate scoped audit; this Task covers only MSXOrg/docs.
The review-ready PR card is published and the final handoff is saved with the owning coordinator. GitHub confirms the PR is open, ready for review, unmerged, at ee3c17855cb282cdf236fc764df3f45502f4c057, with auto-merge unset. App Agent merge is disabled and its merge action is false. No scoped blocker remains. This Task and the Bug stay open for their normal PR merge closure; no merge is authorized by this handoff.