[finding] AuthManager still carries two more independent basePath normalisers — getAuthIssuer() and getMcpResourceUrl() — and one of them builds a malformed URL #16399

Measured, on the real manager

Driven against the built @objectstack/plugin-auth dist at 38463491893 (branch claude/issue-16025-hono-auth-mount-basepath), constructing a real AuthManager with baseURL: 'http://localhost:3000':

Two defects, both pre-existing (they reproduce on the merge base too):

1. A basePath written without a leading slash produces a malformed MCP resource URL. getMcpResourceUrl() strips a trailing /auth and concatenates, adding no leading slash: http://localhost:3000 + api/v1 + /mcp. That value is the RFC 8707 resource identifier and the token aud (auth-manager.ts — validAudiences, the AS metadata resource, and the jwt audience), and auth-plugin.ts:3176 parses a path back out of it with new URL(...).pathname.
2. A basePath written with a trailing slash keeps it in the OAuth issuer. getAuthIssuer() adds a leading slash when absent but never strips a trailing one, so the advertised issuer carries a slash that the string better-auth is now configured with does not.

Where they come from

auth-manager.ts had four readers of this.config.basePath. getBasePath() (added on #16025, PR #16380) collapsed two of them — the string handed to better-auth and betterAuthEndpointPath's copy. These two remain, each re-deriving this.config.basePath || '/api/v1/auth' with its own normaliser:

getAuthIssuer()       adds a leading slash, KEEPS a trailing one
getMcpResourceUrl()   adds nothing, strips a trailing `/auth`

Why PR #16380 did not collapse them, and why this is its own card

The at-tier contract review of that PR raised this and gave two acceptable routes; the implementing seat took the documentation route deliberately and is reporting the rest here.

Collapsing these two onto getBasePath() is not a mechanical de-duplication. getAuthIssuer() is the OAuth iss this AS advertises (and one of validAudiences); getMcpResourceUrl() is the resource identifier a token's aud is matched against. Both are compared by exact string by relying parties, so normalising either re-selects tokens issued before the change — a published-identifier decision with a security blast radius, on a card whose scope was where an HTTP adapter mounts /auth/*.

⇒ Filed unassigned, for triage rather than for a route. The two sub-questions are genuinely separate:

• The malformed URL (defect 1) looks like a straightforward repair, but "which spelling of the resource identifier is canonical" still moves an aud for any deployment configuring basePath without a leading slash.
• The trailing slash in iss (defect 2) is a live divergence created by PR fix(hono): mount /auth where the auth service serves, and refuse a prefix it cannot serve under #16380 in one direction: better-auth is now configured with the stripped form while getAuthIssuer() still reports the unstripped one. Before that PR both carried the slash.

Not measured here

• Whether any deployment configures basePath without a leading slash or with a trailing one. @objectstack/hono has no in-repo consumer (adapters/hono 的 ${prefix}/auth/* 与 ${prefix}/storage/* 是终结式通配 —— 与 #4088 同一缺陷，只是这个包在仓内没有消费者 #4117), and nothing here can answer for out-of-repo hosts.
• Whether an iss change would break a live relying party. That is the decision, not a fact this card can settle.

Adjacent

• [finding] /api/v1/auth is declared as two independent literals — AuthPlugin's constructor default and AuthManager's fallback — and a divergence between them is silent #16384 — the other basePath duplication on this manager (AuthPlugin's constructor default vs AuthManager's fallback, two literals of /api/v1/auth). Different pair, same family; neither subsumes the other.
• PR fix(hono): mount /auth where the auth service serves, and refuse a prefix it cannot serve under #16380 — where this was measured. Its docblock, changeset and commit messages now state explicitly that getBasePath() is not the single definition, and point here.