fix: fixed bds2: protocol handling

specs/editor_settings.allium

@@ -157,8 +157,13 @@ surface SettingsViewSurface {
         --   Blogmark Category (select), Blogmark Bookmarklet (copy button), Save button.
 
     @guarantee BookmarkletCopy
-        -- Copy button copies bookmarklet JavaScript to clipboard.
-        -- Bookmarklet uses project's publicUrl to construct POST endpoint.
+        -- Copy button copies the generated bookmarklet JavaScript to the system
+        -- clipboard for pasting into a browser bookmark.
+        -- The bookmarklet captures the active tab's document.title and
+        -- location.href (each URI-encoded) and navigates to a
+        -- bds2://new-post?title=...&url=... deep link. When a project is active
+        -- its id is appended as &project_id=... so the link targets that
+        -- project regardless of which one is currently open.
 
     @guarantee EditorSection
         -- Section 2: Default Editor Mode (select: Markdown/Preview),

specs/script.allium

@@ -12,6 +12,8 @@ config {
     transform_max_toasts_per_script: Integer = 5
     transform_max_toasts_total: Integer = 20
     transform_max_toast_length: Integer = 300
+    blogmark_max_title_length: Integer = 200
+    blogmark_max_url_length: Integer = 2048
 }
 
 enum ScriptStatus {
@@ -243,6 +245,36 @@ rule ExecuteTransform {
         requires: t.entrypoint != ""
         ensures: TransformApplied(t, data)
 
+    @guarantee BlogmarkInputHardening
+        -- The deep link is sanitized before transforms run. Title and url have
+        -- control characters stripped and are trimmed. The url is kept only
+        -- when it is http(s) with a host, after removing any credentials and
+        -- fragment, and at most config.blogmark_max_url_length characters;
+        -- otherwise it is dropped so it can never reach the post body. The
+        -- title is capped at config.blogmark_max_title_length and falls back to
+        -- the url host when blank.
+
+    @guarantee BlogmarkDefaultBody
+        -- When the deep link supplies no content, the post body defaults to a
+        -- markdown link [title](url) to the bookmarked page, with the title
+        -- escaped, so transforms operate on the same candidate whether or not
+        -- an explicit body was provided. Explicit content always wins.
+
+    @guarantee BlogmarkProjectTargeting
+        -- A link's own project_id determines the target project. When it names a
+        -- project that exists and differs from the active one, the shell
+        -- switches to it before importing so the new post is shown in context.
+        -- When it names a project that does not exist in this install (e.g. a
+        -- bookmarklet copied from another instance) the import is refused with
+        -- an error — the post is never redirected into a different project.
+        -- With no link project_id, the active project is used; with no link
+        -- project_id and no active project the import is refused with a warning.
+
+    @guarantee BlogmarkColdStartDelivery
+        -- A deep link that arrives before the shell is connected (e.g. the
+        -- bookmarklet launching the app) is queued and replayed once a shell
+        -- attaches, so launching via a bookmarklet still creates the post.
+
     @guarantee TransformTrigger
         -- Transform scripts are triggered automatically by blogmark import.
         -- Each script receives the current post candidate plus a context with