Add consumer documentation extensions and runtime DLL helper

[sean-parent]

Summary

• Add additive DOCS_INPUTS and DOCS_OPTIONS configuration while retaining custom Doxyfile compatibility.
• Expose cpp_library_copy_runtime_dlls for custom executables and share it with toolkit tests/examples.
• Add documentation-rendering and runtime-deployment fixtures, CI coverage, and consumer guidance.

Test plan

User-approved CMake suite replaces the inapplicable Cargo checklist in the PR-opening skill. All seven scripts passed on Windows/MSVC through a temporary short path, with zero emitted warnings.

• cmake -P tests/install/CMakeLists.txt (32 passed; one pre-existing manual-only case skipped)
• cmake -P tests/install/test_provider_merge.cmake
• cmake -P tests/install/test_nested_install.cmake (11 cases)
• cmake -P tests/setup/test_target_type.cmake (4 cases)
• cmake -P tests/setup/test_setup_version_resolution.cmake
• cmake -P tests/setup/test_runtime_dlls.cmake
• cmake -P tests/docs/test_docs.cmake (10 cases, including continuation regressions)
• git diff --check

Generated with GitHub Copilot CLI

[sean-parent]

Talos local review

Verdict: One LOW-priority correctness finding; no blocking findings accepted. The finding remains outstanding.

Reviewed SHA: 6a5890b01217125086215544eece0c71cea749e5

Scope: Full clean PR diff, origin/main...HEAD, merge base b5273fdb8d0c356579883d8419fecb267804d0f2; 16 changed files, no binary/generated-file exclusions.

Reviewers: Security, correctness, standards, architecture, operations, and documentation, followed by the challenge reviewer. Persona routing: static roster (DCAP routing off; rerun with --dcap to enable).

Counts: 1 candidate received; 1 after field/file validation; 0 invalid drops; 0 duplicates; 1 challenged; 1 accepted; 1 inline posted; 0 overflow; 0 unanchored findings; 0 recurrences or unchanged duplicate findings in existing reviews/comments at publication time.

Accepted finding

LOW, HIGH verified confidence: DOCS_OPTIONS accepts a trailing continuation backslash, allowing one assignment to consume the next. The challenge reviewer reproduced the behavior with Doxygen 1.18.0 and the PR's validation/append logic: WARN_IF_UNDOCUMENTED = NO was parsed into PREDEFINED instead of overriding the template's YES value, without a diagnostic.

Coverage and verification limitations

All six baseline personas read the complete 548-line diff. The architecture reviewer did not consume the supplied Scout artifacts. The operations reviewer did not consume the supplied base rules, changed-file list, or Scout artifacts. The security reviewer consumed Scout leads but did not explicitly confirm packet consumption. The challenge reviewer consumed the diff, base rules, and leads, but did not consume the Scout packet. Follow-up delivery to the synchronous persona agents was unavailable. These are disclosed input-coverage limitations, not evidence of absence of defects.

No GitHub custom instructions or active OpenSpec specifications were found at the base snapshot. Local validation ran the user-approved seven-script CMake suite on Windows/MSVC through a temporary short drive path, with zero emitted warnings on the final run. The mapping was removed afterwards. The install mapping script has one pre-existing manual-only skipped case. Linux/macOS execution and the separate Ubuntu scaffold integration job were not reproduced locally. The two new feature suites and existing nested-install/target-type integration suites were executed.

Scout

Used: risk MEDIUM, evidence MEDIUM, status COMPLETE (scout 0.9.130, 11s).
Cross-file leads given to reviewers: 0 changed symbol(s) with uses outside the diff, 0 new value(s) not handled everywhere, 0 possibly unused symbol(s).

Tests to run (Scout: 5 Must Run test(s), in these files):

• tests/docs/fixtures/include/fixture/api.hpp
• tests/setup/fixtures/runtime_dlls/runtime.cpp
• tests/setup/fixtures/runtime_dlls/main.cpp

Scout warnings: index refresh did not finish (exit 124); evidence may lag the working tree

The packet also reported conservative whole-file roots for nine changed files. Its listed fixture files were exercised through the docs and runtime deployment scripts; they are not standalone test commands.

Generated with GitHub Copilot CLI. This is a COMMENT review, not an approval.

[sean-parent]

Talos local review — post-fix pass

Verdict: The continuation finding is fixed and its thread is resolved. One new LOW-priority documentation finding was accepted; no functional or security defects were accepted.

Reviewed SHA: 09dee6069307a258a394a6e3781ae4c342fa8cb9

Scope: Full clean PR diff, origin/main...HEAD, merge base b5273fdb8d0c356579883d8419fecb267804d0f2; all 17 changed files, no binary/generated exclusions.

Reviewers: Security, correctness, standards, architecture, operations, documentation, and challenge. All six personas confirmed reading the entire 580-line diff, changed-file list, and base rules; all four Scout-enabled personas and the challenger confirmed reading both leads and packet. No persona failed. The challenge review read only the start of the prior review body but read the inline history in full. No GitHub custom instructions or active OpenSpec specifications were found.

Counts: 1 candidate received; 1 after validation; 0 drops; 0 duplicates; 1 challenged; 1 accepted; 1 inline posted; 0 overflow; 0 unanchored findings; 0 recurrences. The new finding is distinct from the resolved DOCS_OPTIONS thread.

Accepted finding: LOW, HIGH verified confidence: module descriptions still portray cpp-library-testing.cmake as only a backward-compatibility wrapper, although the PR adds public runtime deployment implementation there. The header and CLAUDE.md architecture bullet should describe both roles. This is inaccurate current documentation; speculative deletion consequences are not used to set severity.

Verification: Public consumer/Doxygen reproduction confirmed the whitespace continuation trigger, and terminal and whitespace continuation regressions each failed before the fix. All 10 documentation cases now pass on Windows/MSVC with no emitted warnings, including quoted Windows-style backslashes and legacy custom Doxyfiles. The original seven-script CMake suite passed before this fix; unchanged suites were not rerun for the localized docs validation change. Linux/macOS and the separate Ubuntu scaffolding integration job were not reproduced locally. Personas performed static reviews rather than additional test execution; the external pinned action implementation and runner tool availability were not independently audited.

Persona routing: static roster (DCAP routing off; rerun with --dcap to enable).

Scout

Used: risk MEDIUM, evidence MEDIUM, status COMPLETE (scout 0.9.130, 2s).
Cross-file leads given to reviewers: 0 changed symbol(s) with uses outside the diff, 0 new value(s) not handled everywhere, 0 possibly unused symbol(s).

Tests to run (Scout: 5 Must Run test(s), in these files):

• tests/docs/fixtures/include/fixture/api.hpp
• tests/setup/fixtures/runtime_dlls/runtime.cpp
• tests/setup/fixtures/runtime_dlls/main.cpp

The packet reported conservative whole-file roots for ten changed files. The listed fixtures are exercised by the docs and runtime deployment scripts, not standalone commands.

Generated with GitHub Copilot CLI. COMMENT review, not an approval.

[sean-parent]

Talos local review — final pass

Verdict: No new issues found. Both previously accepted findings have fixes, thread replies, and resolved threads: continuation validation in 735e22d, and module documentation in e7494f7. The requested Scout configuration changes are committed in 09dee60.

Reviewed SHA: e7494f77e699a694aaa4b95651d2b9a108ca5221

Scope: Full clean PR diff, origin/main...HEAD, merge base b5273fdb8d0c356579883d8419fecb267804d0f2; 17 changed files, 596 diff lines, no binary/generated-file exclusions.

Reviewers: Security, correctness, standards, architecture, operations, documentation, and challenge. All six baseline personas confirmed complete diff, changed-file-list, and base-rule reads. All four Scout-enabled personas confirmed reading the leads and packet. The challenge stage completed with an empty validated observation set, reading leads, packet summary, and file list rather than repeating the complete source review. No persona failed. No custom GitHub instructions or active OpenSpec specifications were found.

Counts: 0 candidates; 0 after validation; 0 dropped; 0 duplicates; 0 accepted; 0 inline posted; 0 overflow; 0 unanchored findings; 0 recurrences. No unchanged duplicate findings were reposted.

Verification and coverage limitations: The continuation issue was reproduced through the real public consumer configuration and Doxygen 1.18.0, including the whitespace suffix that survives CMake argument forwarding. Both new rejection cases failed against the pre-fix implementation and passed after the fix. All 10 documentation cases passed on Windows/MSVC, including quoted Windows-style backslashes and legacy custom Doxyfiles, with no emitted warnings on the final run. Subsequent code changes were comments/documentation only. The original seven-script repository suite passed before the localized validation fix; unaffected suites were not rerun for that fix. Linux/macOS execution and the separate Ubuntu scaffold integration job were not reproduced locally. Persona review is static: it did not independently execute CI, audit the external pinned action's implementation, or verify runner-installed tools. Existing review history was not re-read by personas with zero candidate findings; publication checked current comments/thread state separately.

Persona routing: static roster (DCAP routing off; rerun with --dcap to enable).

Scout

Used: risk MEDIUM, evidence MEDIUM, status COMPLETE (scout 0.9.130, 2s).
Cross-file leads given to reviewers: 0 changed symbol(s) with uses outside the diff, 0 new value(s) not handled everywhere, 0 possibly unused symbol(s).

Tests to run (Scout: 5 Must Run test(s), in these files):

• tests/docs/fixtures/include/fixture/api.hpp
• tests/setup/fixtures/runtime_dlls/runtime.cpp
• tests/setup/fixtures/runtime_dlls/main.cpp

The packet reported conservative whole-file roots for ten changed files and MEDIUM evidence. Its listed fixture files are exercised by the docs and runtime deployment scripts, not standalone test commands. No inferred absence of defects relies on empty Scout impact rows.

Generated with GitHub Copilot CLI. COMMENT review, not an approval.