test(symlink): say what a local skip means — CI is not skipping these

[askalf]

A local run on Windows without Developer Mode reports # skipped 4 for test/symlink-skills.test.mjs, and that count reads as a Windows coverage hole.

It is not one. The skip has always been a runtime capability probe rather than a platform check, and all three CI runners create symlinks fine:

job	skipped
windows-latest, node 20	0
windows-latest, node 22	0
ubuntu / macos (20 & 22)	1 — the DPAPI test, Windows-only by design

Every test in that file already runs on every platform. The only machines that skip are developer boxes without Developer Mode.

I misread that count myself and proposed rebuilding these tests around Windows junctions to "un-skip them in CI" — which would have added a second set of fixtures for coverage that already exists. Checking the runner logs first is what killed that idea.

So the fix is to stop the count from misleading the next person: the comment now says where the skip bites and where it does not, and the skip message tells you how to run them locally rather than only stating that a privilege is missing.

# SKIP needs symlink privilege — enable Windows Developer Mode to run these
#      locally (CI runs them on all three OSes)

No behaviour change, no new coverage, tests only. Independent of #110-#115.

[sprayberry-reviewer]

Automated review from the Sprayberry Labs fleet code reviewer.

Verdict: Clean — no blocking issues. Comment/message-only change, approving.

What I checked:

• Full diff (test/symlink-skills.test.mjs, +17/-6): every hunk either rewrites the top-of-file comment explaining the symlink-privilege skip, or swaps four call sites' inline 'symlinks unprivileged here' skip string for a shared NO_LINKS constant with a more actionable message. No test logic, assertions, or control flow changed — confirmed by re-reading each hunk context (link(...) calls, t.skip(...) calls) around the changed lines.
• CI check-rollup: all 8 checks green (test on ubuntu/macos/windows × node 20/22, CodeQL analyze, CodeQL run, "verify pinned skills"), completed ~21:15Z, none pending/failing.
• PR body's claim (Windows CI runs report # skipped 0, i.e. the skip only bites on unprivileged local dev boxes) is consistent with the CI rollup showing the windows-latest jobs passing same as others.

What's good: Extracting the repeated skip-message string into NO_LINKS (test/symlink-skills.test.mjs) removes duplication across the four call sites and the new message tells a developer how to unblock locally instead of just stating a fact. Comment rewrite is accurate and scoped — no new coverage claimed, none needed.

No correctness, security, reuse, or test-coverage findings on this diff.