security(smime): fork upstream plugin and fix two audit findings

S-01 audited bulwarkmail/plugins/smime @ 91085a3 (2,935 lines). Nine
findings, two HIGH. No backdoor and no exfiltration path anywhere in the
bundle — the problems are trust-model and input-validation gaps. Full
report in vnc/audits/SMIME-PLUGIN-AUDIT-2026-08-04.md.

Fork is source-only. The upstream smime.zip is a 1.77 MB prebuilt bundle
whose manifest reads 1.0.1 while the source reads 1.0.2, so auditing
src/ would not audit what that zip installs. We build from source.

Finding 1 (HIGH) — certificate substitution. maybeAutoImportSigner gated
on signatureValid alone, but smimeVerify runs checkChain:false, so that
only proves "signed by whoever holds this key", not that the claimed
identity is real. Self-sign a cert asserting victim@example.com, send one
signed message, and it was stored as the encryption target for that
address — the user's next Encrypt to the victim went to the attacker.
Now requires signerEmailMatch === true and !selfSigned. Both values were
already computed and displayed as untrusted in the banner; only the
import path ignored them. Tests for `true` explicitly so an undefined
match (missing From header) fails closed.

Finding 3 (MED-HIGH) — CRLF header injection. Escaping reached only
Subject and attachment filename; display names, raw addresses,
Message-ID, In-Reply-To, References and attachment Content-Type were
emitted verbatim, and formatAddress escapes only backslash and quote.
In-Reply-To/References/display names are copied from inbound mail when
replying or forwarding, so the value is attacker-supplied. Sanitising
inside formatHeader covers all 17 call sites by construction; the three
headers assembled directly get stripCrlf explicitly.

Also adds auth:observe to the manifest. The plugin registers
onAfterLogout/onAccountSwitch — real hooks (lib/plugin-hooks.ts:362-363)
— without declaring the permission, so under B-09 the session-key wipe
would silently stop running.

verify-fixes.mjs carries 19 assertions including source checks that fail
if either guard is removed or a new unsanitised interpolated header
appears. That last one immediately caught the interpolated smime-type
Content-Type header, which manual review had dismissed as static.

Finding 2 (unauthenticated CBC accepted on decrypt) is NOT fixed. This
is not safe for real mail yet — sandbox accounts only.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

vnc/VNC-CHANGES.md

@@ -50,6 +50,12 @@ microfrontends integration was also added and reverted the same day._
 | 2026-08-04 | `lib/plugin-sandbox/loader.ts` | **B-04 security fix** — gate hook registration on granted permissions via new `HOOK_PERMISSIONS` map; refused hooks are skipped, logged and counted | `info.hooks` is self-reported by the sandbox, so an untrusted plugin could claim `onRenderEmailBody` and replace any rendered email body without holding `email:render-takeover`. Consent copy gated what the user was *asked*, not what the host *allowed*. |
 | 2026-08-04 | `lib/plugin-sandbox/host-api.ts` | export `hasPermission()` (was module-private) | one source of truth for the permission rule — the loader gate and the RPC gate must not drift apart |
 | 2026-08-04 | `app/api/admin/plugins/route.ts` | **B-01** — scan all `.js`/`.mjs` in the bundle (was entrypoint only); return structured `findings` + `canOverride`; allow admin `overrideWarnings=true` with a `plugin.install.scan_override` audit entry; echo accepted `findings` on success | hard-reject on `eval(`/`new Function(`/`innerHTML =` made every crypto plugin uninstallable (minified openpgp.js/pkijs trip it), while only scanning the entrypoint left a trivial bypass. Route is already admin-authenticated, so the scan is defence-in-depth, not a trust boundary. |
+| 2026-08-04 | `vnc/audits/SMIME-PLUGIN-AUDIT-2026-08-04.md` (new) | **S-01** security audit of `bulwarkmail/plugins/smime` @ `91085a3` — 9 findings (2 HIGH, 1 MED-HIGH), verdict: fork and fix before shipping | privileged same-origin plugin that handles users' private keys; verdict must precede any deploy |
+| 2026-08-04 | `vnc/plugins/smime/` (new) | fork of the upstream S/MIME plugin, **source only — upstream `smime.zip` deliberately NOT vendored** | shipped zip is a 1.77 MB bundle at manifest 1.0.1 while source is 1.0.2, so auditing `src/` would not audit what the zip installs. We build from source via `npm run package`. |
+| 2026-08-04 | `vnc/plugins/smime/src/index.js` | **audit fix 1 (HIGH)** — `maybeAutoImportSigner` now requires `signerEmailMatch === true` and `!selfSigned` before trusting a signer cert | upstream gated on `signatureValid` alone, but `smimeVerify` runs `checkChain:false`, so a self-signed cert asserting any address was silently stored as the ENCRYPTION TARGET for it. Both values were already computed and ignored. |
+| 2026-08-04 | `vnc/plugins/smime/src/mime-builder.js` | **audit fix 3 (MED-HIGH)** — `stripCrlf()` applied inside `formatHeader` + the 3 directly-assembled headers (`att.contentType`, `att.cid`, `smimeType`) | CRLF escaping reached only Subject and filename; display names, Message-ID, In-Reply-To and References were raw — and those are copied from inbound mail on reply/forward, making it remotely reachable header injection |
+| 2026-08-04 | `vnc/plugins/smime/manifest.json` | add `auth:observe` | plugin registers `onAfterLogout`/`onAccountSwitch` (real hooks, `lib/plugin-hooks.ts:362-363`) without declaring the permission; under `B-09` the session-key wipe would silently stop running |
+| 2026-08-04 | `vnc/plugins/smime/verify-fixes.mjs` (new) | 19 regression assertions for both fixes, incl. source checks that fail if a guard is removed | the source assertion caught an interpolated header manual review had wrongly dismissed as static |
 | 2026-08-04 | `app/(main)/admin/_tabs/plugins.tsx` | **B-01 (UI)** — scanner-findings review panel: holds the rejected file, lists pattern-per-file, offers "Install anyway" / "Cancel"; success message reports how many findings were accepted | without this the override was API-only — an admin uploading a crypto bundle through the web form hit a 400 they could not act on. Also replaces a dead `data.warnings` read (never returned by the route) with the live `findings` field. |
 
 _(append new rows as you diverge)_