check-*.mjs under bash: the script exits 2, but a piped grep reports it as a pass

[explosivebit]

What is wrong

website/scripts/check-*.mjs are Node ESM scripts, but they are runnable-looking as
bash scripts/check-content-completeness.mjs - and that form fails in a way that reads as success
in the most common usage.

$ bash website/scripts/check-content-completeness.mjs
website/scripts/check-content-completeness.mjs: line 2: //: is a directory
website/scripts/check-content-completeness.mjs: line 3: //: is a directory
$ echo $?
2

The script itself fails loudly. The trap is the pipeline: bash script.mjs | grep … reports
grep's status (1 for "no matches"), so a caller checking the exit code of the pipeline sees "no
issues found" - a green that never ran the check. Three agents hit exactly this in one session and
two of them reported the completeness gate as green before realising.

Evidence

• bash website/scripts/check-content-completeness.mjs → exit 2, //: is a directory
• node website/scripts/check-content-completeness.mjs → exit 0, [content] total 176 files scanned, 0 issues
• the same shape applies to check-translation-drift.mjs and check-dead-links.mjs

Fix options

1. Add a #!/usr/bin/env node-aware guard is not possible in bash - the practical fixes are:
2. a one-line note next to each documented invocation (docs/ and AGENTS.md) that the runner is
   node, and
3. make the scripts refuse to be interpreted by anything but Node is not feasible; instead, prefer
   wiring the common ones into npm scripts (npm run check:content), so the invocation people copy
   is already correct.

Option 3 removes the failure mode instead of warning about it.

Revisit trigger

Act if a review or CI summary reports one of these checks green while the underlying command was
bash. The two are indistinguishable in every summary format we use today.