Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@scripts/build_site.py`:
- Around line 59-60: The call to _count_examples() is duplicated (once in main()
and again inside build_index_md()), causing redundant filesystem scans; modify
the code to compute example/conformance counts once in main() (using
_count_examples() and _count_conformance()) and pass those values as parameters
into build_index_md(count_examples, count_conformance) (or implement a simple
module-level cache in _count_examples() that returns a stored value on
subsequent calls), then update build_index_md() to consume the passed counts (or
read from the cache) so the filesystem is scanned only once.
- Around line 40-60: check_doc_counts.py currently validates counts in
TESTING.md but not the generated site artifacts, so stale counts can be
committed; extend the validation to parse and verify docs/llms.txt and
docs/index.md (or add a CI step that runs build_llms_txt and diffs generated
files) by reusing the same dynamic counting logic (_count_examples,
_count_conformance) used in build_llms_txt and adding checks that the numbers in
docs/llms.txt and docs/index.md match the computed values; update
check_doc_counts.py to locate and read those files, extract the documented
counts, and fail the check on mismatch (alternatively add a pipeline job to run
build_llms_txt and compare outputs to the checked-in docs).
- Around line 45-51: The _count_conformance() function reads the
tests/conformance/manifest.json using (ROOT / "tests" / "conformance" /
"manifest.json").read_text() without specifying encoding; update that
read_text() call in _count_conformance to include encoding="utf-8" to match
other reads in this file (so _count_conformance uses the same UTF-8 explicit
read as the rest of the script).

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@scripts/build_site.py`:
- Around line 41-47: Replace the use of functools.lru_cache(maxsize=None) with
functools.cache for the _count_examples function: update the decorator on the
_count_examples definition to use `@cache` and ensure functools.cache is imported
(replace or add the import of cache alongside or instead of lru_cache) so the
function uses the simpler built-in cache decorator.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@scripts/build_site.py`:
- Around line 136-149: build_llms_full_txt currently compiles only SKILL.md,
AGENTS.md, FAQ.md, error codes and grammar but the site now links additional
docs (README, EXAMPLES, DESIGN, CHANGELOG, ROADMAP, HISTORY, vera/README,
TESTING.md, KNOWN_ISSUES.md), so update scripts/build_site.py's
build_llms_full_txt to include those newly linked documents in its source list
(use the same RAW path expansion used elsewhere), concatenate them into
llms-full.txt in the intended order, and ensure the “complete documentation”
claim matches the actual contents if you prefer not to include a file; also
apply the same inclusion or claim-narrowing to the other related
functions/blocks referenced in the review (the other build paths noted in the
comment).