.github/workflows/npm-publish.yaml (1)

33-33: Empty-string guard is dead code under workflow_dispatch.

inputs.version is required: true (line 8), so [ -n "$INPUT_VERSION" ] is always true at this trigger. The || fallback at line 12 (github.event.release.tag_name || inputs.version) suggests this workflow once had a release: trigger that has since been removed. Either re-add the trigger or drop the dead RELEASE_TAG_NAME plumbing in a follow-up — not blocking.

🤖 Prompt for AI Agents

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

In @.github/workflows/npm-publish.yaml at line 33, The branch contains dead-code
guarding INPUT_VERSION with [ -n "$INPUT_VERSION" ] because workflow_dispatch
declares inputs.version as required and INPUT_VERSION is always set; update the
action to either remove the redundant empty-string check (delete the [ -n
"$INPUT_VERSION" ] && prefix) or restore the prior release trigger and
RELEASE_TAG_NAME plumbing so the conditional makes sense; locate the conditional
using INPUT_VERSION and the expression github.event.release.tag_name ||
inputs.version and apply one of those two fixes consistently across the
workflow.

.github/workflows/docker-manual.yaml (2)

58-73: LGTM — case matching against $INPUT_REF is the right pattern.

The same env→shell-var rewrite repeated identically in docker-default, docker-dev, and docker-alpine (lines 58–73, 112–125, 165–178). Since the three jobs are otherwise close to identical aside from Dockerfile/tag suffix, consider extracting to a composite action or reusable workflow in a follow-up — purely a maintenance win, not blocking.

🤖 Prompt for AI Agents

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

In @.github/workflows/docker-manual.yaml around lines 58 - 73, The "Resolve
version" step (id: get_version) that sets INPUT_REF -> VERSION via the case
block is duplicated across jobs docker-default, docker-dev, and docker-alpine;
extract that logic into a single reusable unit (either a composite action or a
reusable workflow) and replace the duplicated steps in each job with a call to
that unit, ensuring the composite accepts INPUT_REF (or inputs.ref) and exposes
VERSION via outputs or writes to GITHUB_ENV exactly as the current step does to
preserve behavior.

---

15-31: Prefer an allow-list for git refs over a deny-list of metacharacters.

The current deny-list [[:space:];|&$\()<>!] misses several shell-significant characters (`, ', ", *, ?, {, }, ~, #). It currently doesn't matter because INPUT_REF is passed via env: and double-quoted in case "$INPUT_REF", but a future refactor that drops the env-passing would silently re-open injection.

Git refs are constrained by git check-ref-format, so an allow-list is both stricter and easier to reason about. Also: empty check should come first to short-circuit.

🛡️ Suggested allow-list

       - name: Validate ref input
         env:
           # Pass via env so the value is treated as data, not interpolated code.
           INPUT_REF: ${{ inputs.ref }}
         run: |
-          # Reject refs containing shell metacharacters (space, ;, &, |, $, (, ), `, <, >, !).
-          if echo "$INPUT_REF" | grep -qE '[[:space:];|&$`()<>!]'; then
-            echo "::error::inputs.ref contains invalid characters: ${INPUT_REF}"
-            exit 1
-          fi
           if [ -z "$INPUT_REF" ]; then
             echo "::error::inputs.ref must not be empty"
             exit 1
           fi
+          # Allow only characters valid in git refs / version tags.
+          if ! printf '%s' "$INPUT_REF" | grep -qE '^(refs/(heads|tags)/)?[A-Za-z0-9._/-]+$'; then
+            echo "::error::inputs.ref must be a valid git ref (got: ${INPUT_REF})"
+            exit 1
+          fi

🤖 Prompt for AI Agents

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

In @.github/workflows/docker-manual.yaml around lines 15 - 31, Replace the
deny-list metacharacter check in the validate-inputs step with an allow-list
that enforces Git ref syntax and move the emptiness check first: first test if
INPUT_REF is empty and fail, then validate INPUT_REF against a strict
allow-regex equivalent to git check-ref-format (e.g. allow only characters and
patterns permitted in refs: no ASCII control chars, no consecutive dots, no
component starting/ending with dot, no components with @{-}, no space, and only
the standard safe characters), failing with a clear "::error::" message if the
ref does not match; update the script that references INPUT_REF in the
validate-inputs job so the condition uses this allow-list regex and preserve
quoting of "$INPUT_REF" when logging errors.

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes