`share-link-routes.ts` is the fifth drifting envelope module — no body carries `success`, and one answers the `{ ok: true }` #3689 retired

[os-zhuang]

Found by the repo-wide guard added in #3843 (PR #3972), not by hand. #3843's survey listed four modules; this is a fifth, and it was missed for a specific and repeatable reason: every guard before this one was per-package, so a module nobody thought to convert was never audited. Filing it rather than letting it ride along, the same call #3675's changeset made about storage's success bodies.

Mildly ironic: #3675's own changeset cites this file as an example of the good shape — "the nested-error shape the sibling services already use (settings-routes.ts, share-link-routes.ts)". That was true of its error object and not of anything else.

The drift

packages/plugins/plugin-sharing/src/share-link-routes.ts, 6 response write sites:

res.status(status).json({ error: { code, message } });          // :52  — sendError: nested, but NO `success: false`
await res.status(201).json({ link });                          // :104 — bare success
await res.json({ links: link });                               // :127 — bare success
await res.status(200).json({ ok: true });                       // :139 — the private second word for `success`
await res.json({ record: …, link: { … } });                    // :209 — bare success
res.status(200).json({ data: rows ?? [] });                     // :265 — `data` present, flag absent

So this module carries three of the four dialects the #3636 → #3843 sequence has been retiring:

dialect	retired by	still here
no success flag on any body	#3689 (storage), #3843 (four more)	✅ all 6
ok: true as a second success word	#3689 (storage's { ok: true, key })	✅ :139
payload at the top level instead of under data	#3689, #3843	✅ 4 sites
error as a bare string	#3675	❌ already nested — the one it got right

:265 is the interesting one: it already answers { data }, so it is one field short of conformant.

Why it is not fixed in #3843

Converting it is breaking for share-link consumers and needs a consumer sweep of its own — public share links are an end-user-facing feature, and GET /share/:token is read by the anonymous share view, not just by the SDK. That is exactly the shape of change #3687 did for storage: producer + consumers, landed deliberately, with the console taught both shapes first. It is not a quiet ride along with someone else's PR.

Meanwhile the guard pins it: scripts/check-route-envelope.mjs declares this module at its current responses: 6, ok: 0, err: 0, privateOk: 1 with a ratchet pointing here, so nothing gets worse while this is open, and the numbers drop to the conformant 2 / 1 / 1 / 0 when it lands.

What needs deciding

1. Convert it like the other five — a sendOk / sendError pair, payloads under data, { ok: true } at :139 becomes the envelope's own flag. Needs the anonymous share view and any SDK/console reader taught both shapes first (the service-storage success bodies are three shapes, none carrying success: true — the other half of #3675 #3689 pattern), then the ratchet deleted.
2. Convert the flag only, leave payload depth alone — add success to all six, keep top-level payloads. Cheaper and BaseResponseSchema.safeParse would pass, but it leaves data meaning different things on different routes in the same module, which is the "two shapes from one registrar" state Envelope drift is not just service-storage: four more route modules emit bare bodies, two of them the pre-#3675 { error: '<string>' } #3843 called worse than untouched.
3. Declare the share surface exempt — argue that an anonymous, tokenised share endpoint is not part of the /api/v1 SDK contract (the way hmr-routes.ts is exempted as a dev-only SSE endpoint). Honest only if nothing on it is SDK-addressable; worth checking the route ledger before assuming.

Recommendation: 1. :265 already emits data, and :139's { ok: true } is the precise thing #3689 removed one module over — leaving it is how the next survey misses it again. Worth sequencing after #3841 settles the code vocabulary, since sendError here will need to pick codes.