Add Copilot repository guidance and tuned Scout knowledge

[sean-parent]

Summary

• Add repository-specific GitHub Copilot guidance for the CMake toolkit, consumer lifecycle, installation, templates, and regression commands.
• Seed eight source-backed Scout memories with inferred provenance and documented relationships.
• Preserve existing Scout rules, add approved artifact exclusions and ranking rules, and retain six parameters selected by a 48-round autotune run.
• Add a reusable 20-query evaluation set (8 identifier, 8 natural-language, 4 filtered). NDCG@5 improved from 0.5296 to 0.5878; natural-language NDCG@5 improved from 0.6553 to 0.8008. Keep bundled:v27 at the default ranker blend for every class.
• No CMake implementation or consumer-template behavior changes.

Validation

The user approved repository-native CMake and Scout checks in place of the invoked PR skill's inapplicable Cargo workspace commands. All seven CMake scripts passed under an x64 MSVC environment, with no detected compiler warnings:

• cmake -P tests/install/CMakeLists.txt
• cmake -P tests/install/test_provider_merge.cmake
• cmake -P tests/install/test_nested_install.cmake
• cmake -P tests/setup/test_target_type.cmake
• cmake -P tests/setup/test_runtime_dlls.cmake
• cmake -P tests/setup/test_setup_version_resolution.cmake
• cmake -P tests/docs/test_docs.cmake
• Scout query-set evaluation; the persisted compact configuration reproduced the tuned score in two evaluations.
• Scout memory lint: zero errors, five known orphan warnings retained without automatic maintenance.
• git diff --cached --check

Known limitations

• Filtered DOCS_OPTIONS retrieval still scores zero.
• CMake function lookups are not fully represented in Scout's symbol graph; some public API lookups resolve README sections.
• Local validation does not claim that hosted CI has already passed.

Generated with GitHub Copilot CLI

[sean-parent]

Talos local review

Verdict: No actionable findings accepted within the reviewed scope.

Reviewed SHA: 847b2912af42de996589b9f157d78b53d8579b47

Scope: Full branch diff against the PR's actual base, main (merge base 2068f2e6522ead4577f147f886e264b66f7db07e). Local HEAD matched the pushed PR head.

Reviewed by: Security, correctness, standards, architecture, operations, and documentation personas, followed by the challenge reviewer.

Counts: 3 provisional observations; 2 excluded because they target the generated synonyms.json artifact; 1 sent to challenge and rejected for lack of a documented rule or concrete consequence. Accepted: 0. Posted inline: 0. Overflow: 0. No previous review comments existed for duplicate/recurrence matching.

Coverage: 14 changed files in the complete snapshot. The generated .scout/memory/synonyms.json artifact is marked skipped for finding eligibility; the other 13 files were eligible. Base-revision CLAUDE.md context was loaded. GitHub custom-instructions matching returned zero instructions at the base revision, so newly added guidance was reviewed as content rather than used to govern its own review. No active OpenSpec context was found.

Verification: The user-approved repository-native replacement for the PR skill's inapplicable Cargo suite ran all seven documented CMake regression scripts successfully under x64 MSVC, without detected compiler warnings. Scout query-set evaluation passed; the persisted compact autotune configuration reproduced NDCG@5 0.587805 twice. Staged whitespace checks passed. Scout memory lint returned zero errors and five known orphan warnings; no automatic maintenance was applied.

Limitations: Architecture and operations personas did not consume the Scout packet/leads. Documentation review checked selected public contracts but did not exhaustively trace install internals; correctness review verified the deferred install ordering against source. Search relevance judgments and Scout's external implementation were not independently certified. Scout reported conservative whole-file roots for 13 changed files. Filtered DOCS_OPTIONS retrieval remains a known zero-scoring evaluation case. Hosted CI results are not claimed by this local review.

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

Scout

Used: risk LOW, evidence HIGH, status COMPLETE (scout 0.9.130, 1s).
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).