Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In @.github/workflows/pr-example-validation.yaml:
- Line 43: Update the changed_md grep filter in the workflow to use a literal
dot for the filename/path match, ensuring Markdown files at the repository root
and in subdirectories are detected while retaining the existing .md suffix
requirement.
- Line 43: Update the changed_md command in the PR validation workflow to use
git diff with the existing base and HEAD revision range, or first compute the
merge base before invoking git diff-tree. Preserve the current name and
diff-filter options so only added, modified, copied, or renamed Markdown files
are selected.
- Around line 42-46: Update the output-writing logic in the Markdown
change-detection step so the newline-separated value from changed_md uses GitHub
Actions’ multiline output delimiter syntax instead of echoing it as a plain
list= entry. Preserve the existing output name list and ensure the
delimiter-wrapped value is written safely to GITHUB_OUTPUT for multiple Markdown
files.

In `@pipeline/filters/parser.md`:
- Line 30: Update scripts/test-config.sh to discover each per-file tab title and
pass that title to extract-config.sh, rather than counting only
fluent-bit.yaml/fluent-bit.conf; ensure every renamed example is validated.
Apply this discovery behavior to the tab declarations at
pipeline/filters/parser.md:30-30 and :40-40, and
pipeline/outputs/kafka.md:110-110 and :157-157, using each declaration’s title
as the corresponding configuration name.

In `@pipeline/inputs/kafka.md`:
- Around line 145-156: Update the introductory paragraph to state that each
message is processed by the inline modify_kafka_message function rather than
kafka.lua, and revise the sentence structure so it clearly describes sending the
result back to the fb-sink topic on the same broker.

In `@pipeline/parsers.md`:
- Around line 68-72: Make both standalone YAML examples self-contained: in
pipeline/parsers.md, define custom_parser1 inline or reference a concrete
parser-file fixture before it is used; in pipeline/filters/kubernetes.md,
replace the legacy undefined parsers_file reference with the existing inline
custom-tag definition or another concrete parser-file fixture. Ensure each
extracted YAML tab validates independently.

In `@pipeline/router.md`:
- Line 210: Update the routing.yaml tab title in the routing documentation
around the routing.yaml example so scripts/test-config.sh recognizes and
extracts the routes configuration snippet. Use a supported extractor tab title
while preserving the existing routes syntax example.

In `@scripts/test-config.sh`:
- Around line 56-71: In the suppressed-file loop, replace the three redundant
conditions with one path-anchored match that accepts only an exact path or a
file beneath the suppressed entry as a directory. Preserve the informational
message and exit behavior, while ensuring names embedded in unrelated filenames
or extensions do not suppress validation.
- Around line 96-102: Update the Stage 1 counting flow around extract-config.sh
so genuine command failures remain visible and cause the validation to fail,
rather than being converted into EXAMPLE_COUNT=0. Preserve the legitimate
zero-example path that continues without validation, but capture and report the
extraction error separately using the existing script error-handling
conventions.

---

Nitpick comments:
In `@scripts/extract-config.sh`:
- Around line 146-171: Update the extract-mode validation in the END block so a
missing tab title reports an error and exits nonzero instead of silently
succeeding. Extend the existing found_tab/found_fence handling around
wanted_title, while preserving count mode and the current missing-fence and
missing-code-fence errors.

In `@scripts/test-config.sh`:
- Around line 119-128: Update the validation command in the
configuration-checking flow to capture its combined output during the initial
container run while preserving the exit status. In the failure branch, print the
captured output instead of invoking $CONTAINER_RUNTIME a second time; keep the
existing configuration and failure-context diagnostics unchanged.
- Around line 73-81: Update the image setup around CONTAINER_RUNTIME and
VALIDATION_IMAGE so the script checks whether VALIDATION_IMAGE is already
available locally before pulling it. Invoke the runtime’s image-inspection
command, pull only when the image is absent, and preserve the existing error
message and exit behavior when a required pull fails.

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In @.github/workflows/pr-example-validation.yaml:
- Around line 65-68: Update the changed-file iteration around CHANGED_MD_FILES
to read one line at a time, preserving each newline-delimited path as a single
argument to ./scripts/test-config.sh. Replace the unquoted for loop while
retaining the existing processing message and error_count increment behavior.
- Line 43: Update the changed-file collection around changed_md so git diff runs
separately and its failure causes the validation step to exit nonzero; only
after a successful diff should its output be filtered for .md files, preserving
the existing no-Markdown-files behavior for an empty result.

In `@scripts/README.md`:
- Around line 41-49: Update the all-files validation example to track whether
any invocation of test-config.sh fails, while continuing to process every
Markdown file; after the loop completes, exit with the tracked failure status so
automation receives a non-zero result when any file fails.

---

Outside diff comments:
In `@pipeline/outputs/s3.md`:
- Around line 154-182: Update the Parquet examples in the S3 output
documentation so validation does not run them against the default image lacking
Arrow/Parquet support. Either configure their validation to use an Arrow-enabled
image such as amazon/aws-for-fluent-bit, or add pipeline/outputs/s3.md to the
validator suppression list with an explicit reason, while preserving the
examples themselves.
- Around line 171-182: Add an explicit valid total_file_size setting, such as
50M, to the Parquet PutObject examples in the shown YAML configuration and the
related migratory YAML and classic configuration examples. Preserve the existing
use_put_object settings and formatting while ensuring each relevant example
documents the size explicitly.
- Around line 49-54: The S3 documentation must clarify that Arrow output uses
format: arrow, while compression: arrow is legacy syntax; update the compression
table in pipeline/outputs/s3.md lines 49-54 to document none/unset, and revise
pipeline/outputs/s3.md line 778 to show Parquet files are uncompressed when
format: parquet omits compression, with Arrow compression using zstd or unset
for no compression.
- Around line 194-199: Update the Parquet migration YAML example to include the
documented use_put_object: on setting alongside format: parquet and compression:
snappy, unless the surrounding documentation explicitly establishes that Parquet
enables it automatically.