Add evidence-based repository audits and evals

[study8677]

What changed

• add an inspect-first repository architecture audit mode across the native skill, Codex AGENTS.md, and Cursor rule
• classify findings as observed, inferred, or unknown; rank a high-leverage Top 3 with validation and rollback paths
• bundle a read-only repository inventory tool plus architecture-review and ADR output templates
• add native Codex metadata and make native skill installation the recommended Codex path
• add 14 behavior evaluation cases, a scoring rubric, deterministic cross-surface/reference checks, unit tests, and GitHub Actions CI

Why

The project already had strong architecture knowledge, but its highest-leverage gap was turning that knowledge into a repeatable daily workflow for existing codebases while preventing prompt and platform variants from silently drifting. This change adds both the user-visible repository audit workflow and the feedback loop needed to evolve it safely.

Validation

• python3 scripts/check_quality.py
• PYTHONDONTWRITEBYTECODE=1 python3 -m unittest discover -s tests -v (3/3 passed)
• quick_validate.py skills/architecture-copilot
• native-skill copy/install smoke test
• local Markdown-link and YAML parsing checks
• blind forward test: repository audit read the new reference, ran the bundled inventory, cited real paths, separated evidence levels, and produced a read-only Top 3
• blind non-trigger test: a FastAPI unit-test request correctly stayed out of architecture mode