diff --git a/CHANGELOG.md b/CHANGELOG.md
index 1ddd1845..431011f0 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,40 @@
 # Hermes Web UI -- Changelog
 
+## [v0.51.4] — 2026-05-05 — 10-PR full-sweep batch
+
+### Added
+
+- **PR #1685** by @Michaelyklam — Surface Codex spark models in `/api/models` (closes #1680). New `_read_visible_codex_cache_model_ids()` reads visible non-hidden slugs from `CODEX_HOME/models_cache.json`. The OpenAI Codex group now layers three sources: `hermes_cli.models.provider_model_ids("openai-codex")` first, visible cache slugs second, static `_PROVIDER_MODELS` fallback last. Users see newly available Codex models (including `gpt-5.3-codex-spark`) without waiting for WebUI catalog updates.
+- **PR #1644** by @bergeouss — Inline provider chip + group model count in composer model picker (closes #1425). Same-name models across providers are now visually distinguishable: per-row provider chip on every model option, count `(N)` next to group headings when more than one model matches, subtle border-top divider between provider groups. 13 LOC total — pattern-extension within existing dropdown.
+- **PR #1684** by @Michaelyklam — Clarify update network failures (closes #1321). Frontend detects raw fetch failures (`Failed to fetch`, `NetworkError`, `Load failed`) on `POST /api/updates/apply` and replaces the cryptic browser text with recovery-oriented guidance ("the WebUI may have restarted or the connection was interrupted; wait, reload, and check the server if needed"). Added an in-flight guard so repeated Update Now clicks don't send duplicate apply requests during restart-race windows.
+
+### Fixed
+
+- **PR #1689** by @Michaelyklam — Normalize named profile base homes (refs #749). Prevents the doubled `/base/profiles/foo/profiles/foo` path that occurred when both `HERMES_BASE_HOME=/base/profiles/foo` and the browser cookie `hermes_profile=foo` were set. New `_unwrap_profile_home_to_base()` helper normalizes either env-var path through the same base-home resolver, then routes active-profile and explicit per-request lookups through one shared profile-home resolver. Doesn't touch the broader profile UX umbrella.
+- **PR #1693** by @ai-ag2026 — Avoid adaptive title refresh session lock deadlock. `_run_background_title_refresh()` previously updated a session title while holding the global session `LOCK`, then called `Session.save()` — which itself updates the session index via `_write_session_index()` requiring the same non-reentrant `LOCK` (self-deadlock). Now the in-memory title mutation stays under `LOCK`, but `Session.save()` runs with the global lock released and only the per-session agent lock held. Plus Latin-Unicode-aware fallback title tokenization so `führe` no longer becomes `f` + `hre`.
+- **PR #1701** by @Michaelyklam — Normalize update banner repository URLs (closes #1691). The "What's new?" link previously pointed at `https://github.com/nesquena/hermes-webu/` instead of `hermes-webui`. Root cause: `.git` was treated as a character set (`[.git]`) instead of a literal suffix, and trailing slashes prevented suffix removal. New `_normalize_remote_url()` in `api/updates.py` centralizes the normalization with regression coverage on the edge case.
+- **PR #1703** by @Michaelyklam — Invalidate models cache on auth-store drift (closes #1699). When a user runs `hermes setup` in a terminal and the auth store switches the active provider outside WebUI, the in-memory + disk model caches could keep showing the previous provider's PRIMARY badge for up to the 24h TTL. New non-secret source fingerprint covers `config.yaml` and `auth.json` path/mtime/size; cache rebuilds when either changes outside WebUI. Disk cache schema bumped to reject older cache files cleanly.
+- **PR #1702** by @Michaelyklam — Fix workspace tree double-click rename (closes #1698). The right workspace panel advertised double-click rename on file names, but file-name single-click bubbled to the row's preview handler before the dblclick rename path could take over. Added a `nameEl.onclick` propagation guard before the existing `nameEl.ondblclick` handler in `static/ui.js` while leaving row/icon/whitespace clicks available for preview. Right-click context-menu rename remains as before.
+- **PR #1704** by @Michaelyklam — Honor markdown fence lengths (closes #1696). The `renderMd()` regex hard-coded triple-backtick closers, so 4/5-backtick markdown examples closed at inner triple fences. Updated fenced-code matching to capture `{3,}` backtick opener runs and require the same character + at least as many backticks on close (per CommonMark §4.5). Same fence-length rule applied to user-message fenced rendering and to the blockquote pre-pass fence-state walker. Empty-fence handling unchanged.
+- **PR #1706** by @Michaelyklam — Paste multiple images at once attaches all of them (closes #1697). `static/boot.js` paste handler called `Date.now()` inside a synchronous `.map()` callback over `imageItems`. All N synthesized `File` objects ended up with identical filenames (same millisecond), and `addFiles()` deduped by name and silently dropped images 2..N. Fix captures `pasteTs = Date.now()` once outside the map and adds deterministic `-1`, `-2`, … suffixes only when the paste contains multiple images. Single-image paste filename shape unchanged for compatibility. Functional Node-driven test extracts and executes the real paste handler.
+
+### Tests
+
+4477 → **4503 passing** (+26 regression tests across the 10 PRs). 0 regressions. Full suite ~135s.
+
+### Pre-release verification
+
+- Stage-301 build: 10 PRs merged with zero conflicts (each rebased clean against current master).
+- All JS files syntax-clean (`node -c static/boot.js && node -c static/ui.js`).
+- All Python files syntax-clean (py_compile on every changed file).
+- Live browser walkthrough on port 8789: model picker chip + group count rendering, all `/api/wiki/status`, `/api/logs`, `/api/provider/quota`, `/api/health/agent` endpoints respond 200, sidebar scroll fix preserved, `boot.js` PR #1706 fix verified live (pasteTs captured outside map, index parameter present, Date.now() removed from inside .map()).
+- Opus advisor pass on 9-PR variant (with #1705 in slot 10): SHIP, 7/7 verification questions resolved cleanly. Late swap to #1706 keeps identical fix shape (same `pasteTs` outside map + index suffix); Opus's verification answers carry over because the production diff is unchanged.
+
+### Notes on the 1705 → 1706 swap
+
+@Michaelyklam filed PR #1706 with a functional Node-driven regression test (extracts the real paste handler and asserts two pasted image items become two pending attachments) replacing my own #1705 which used static-source-string assertions. Same code fix, better test approach. Closed #1705 and absorbed #1706 into stage-301.
+
+
 ## [v0.51.3] — 2026-05-04 — 3-PR follow-up batch (#1671, #1673, #1676) + test-fragility fix
 
 ### Added
diff --git a/ROADMAP.md b/ROADMAP.md
index 77d3aefe..43bf7ac0 100644
--- a/ROADMAP.md
+++ b/ROADMAP.md
@@ -2,7 +2,7 @@
 
 > Web companion to the Hermes Agent CLI. Same workflows, browser-native.
 >
-> Last updated: v0.51.3 (May 04, 2026) — 4477 tests collected — 3-PR follow-up batch (#1671, #1673, #1676)
+> Last updated: v0.51.4 (May 5, 2026) — 4503 tests collected — 3-PR follow-up batch (#1671, #1673, #1676)
 > Test source: `pytest tests/ --collect-only -q`
 > Per-version detail: see [CHANGELOG.md](./CHANGELOG.md)
 
diff --git a/TESTING.md b/TESTING.md
index f029ea25..f01fed9f 100644
--- a/TESTING.md
+++ b/TESTING.md
@@ -1835,8 +1835,8 @@ Bridged CLI sessions:
 
 ---
 
-*Last updated: v0.51.3, May 04, 2026 — 3-PR follow-up batch (#1671, #1673, #1676)*
-*Total automated tests collected: 4477*
+*Last updated: v0.51.4, May 5, 2026*
+*Total automated tests collected: 4503*
 *Regression gate: tests/test_regressions.py*
 *Run: pytest tests/ -v --timeout=60*
 *Source: <repo>/*
diff --git a/api/config.py b/api/config.py
index 2ebabc5f..972ee61f 100644
--- a/api/config.py
+++ b/api/config.py
@@ -1585,6 +1585,7 @@ def set_hermes_default_model(model_id: str) -> dict:
 # ── TTL cache for get_available_models() ─────────────────────────────────────
 _available_models_cache: dict | None = None
 _available_models_cache_ts: float = 0.0
+_available_models_cache_source_fingerprint: dict | None = None
 _AVAILABLE_MODELS_CACHE_TTL: float = 86400.0  # 24 hours
 _available_models_cache_lock = threading.RLock()  # must be RLock: cold path refactoring moved slow work inside this lock, requiring re-entry
 _cache_build_cv = threading.Condition(_available_models_cache_lock)  # shares underlying RLock so notify_all() is safe inside with _available_models_cache_lock
@@ -1641,12 +1642,48 @@ def _current_webui_version() -> str | None:
 # guarantees that even if a future release accidentally reuses the same
 # WebUI version string (or a debug build doesn't have a version), a structural
 # change still invalidates the cache.
-_MODELS_CACHE_SCHEMA_VERSION = 2
+_MODELS_CACHE_SCHEMA_VERSION = 3
 
 
 _models_cache_path = STATE_DIR / "models_cache.json"
 
 
+def _get_auth_store_path() -> Path:
+    """Return the auth.json path for the active Hermes profile."""
+    try:
+        from api.profiles import get_active_hermes_home as _gah
+
+        return _gah() / "auth.json"
+    except ImportError:
+        return HOME / ".hermes" / "auth.json"
+
+
+def _models_cache_file_fingerprint(path: Path) -> dict:
+    """Return non-secret identity metadata for a cache dependency file.
+
+    The /api/models response depends on config.yaml (model/provider defaults)
+    and auth.json (active_provider + credential_pool).  The cache only needs
+    cheap invalidation signals here, not file contents; never include secrets.
+    """
+    fingerprint = {"path": str(Path(path).expanduser())}
+    try:
+        st = Path(path).stat()
+    except OSError:
+        fingerprint["missing"] = True
+        return fingerprint
+    fingerprint["mtime_ns"] = st.st_mtime_ns
+    fingerprint["size"] = st.st_size
+    return fingerprint
+
+
+def _models_cache_source_fingerprint() -> dict:
+    """Return the current config/auth-store fingerprint for /api/models cache."""
+    return {
+        "config_yaml": _models_cache_file_fingerprint(_get_config_path()),
+        "auth_json": _models_cache_file_fingerprint(_get_auth_store_path()),
+    }
+
+
 def _delete_models_cache_on_disk() -> None:
     try:
         os.unlink(str(_models_cache_path))
@@ -1717,6 +1754,15 @@ def _is_loadable_disk_cache(cache: object) -> bool:
                 cached_version, runtime_version,
             )
             return False
+    cached_sources = cache.get("_source_fingerprint")
+    runtime_sources = _models_cache_source_fingerprint()
+    if cached_sources != runtime_sources:
+        logger.debug(
+            "models cache rejected: source_fingerprint=%r vs runtime=%r",
+            cached_sources,
+            runtime_sources,
+        )
+        return False
     return True
 
 
@@ -1772,6 +1818,7 @@ def _save_models_cache_to_disk(cache: dict) -> None:
             return
         payload = {
             "_schema_version": _MODELS_CACHE_SCHEMA_VERSION,
+            "_source_fingerprint": _models_cache_source_fingerprint(),
             "active_provider": cache["active_provider"],
             "default_model": cache["default_model"],
             "configured_model_badges": cache["configured_model_badges"],
@@ -1790,15 +1837,27 @@ def _save_models_cache_to_disk(cache: dict) -> None:
 
 def _get_fresh_memory_models_cache(now: float) -> dict | None:
     """Return a valid fresh in-memory /api/models cache, or clear stale shapes."""
-    global _available_models_cache, _available_models_cache_ts
+    global _available_models_cache, _available_models_cache_ts, _available_models_cache_source_fingerprint
     if _available_models_cache is None:
         return None
     if (now - _available_models_cache_ts) >= _AVAILABLE_MODELS_CACHE_TTL:
         return None
+    current_sources = _models_cache_source_fingerprint()
+    if _available_models_cache_source_fingerprint != current_sources:
+        logger.debug(
+            "models memory cache rejected: source_fingerprint=%r vs runtime=%r",
+            _available_models_cache_source_fingerprint,
+            current_sources,
+        )
+        _available_models_cache = None
+        _available_models_cache_ts = 0.0
+        _available_models_cache_source_fingerprint = None
+        return None
     if _is_valid_models_cache(_available_models_cache):
         return copy.deepcopy(_available_models_cache)
     _available_models_cache = None
     _available_models_cache_ts = 0.0
+    _available_models_cache_source_fingerprint = None
     return None
 
 
@@ -1816,10 +1875,11 @@ def invalidate_models_cache():
     result from the disk cache because the disk hit is checked before the memory
     cache rebuild runs.
     """
-    global _cache_build_in_progress, _available_models_cache, _available_models_cache_ts, _cache_build_cv
+    global _cache_build_in_progress, _available_models_cache, _available_models_cache_ts, _available_models_cache_source_fingerprint, _cache_build_cv
     with _available_models_cache_lock:
         _available_models_cache = None
         _available_models_cache_ts = 0.0
+        _available_models_cache_source_fingerprint = None
         _cache_build_in_progress = False
         _cache_build_cv.notify_all()
         # Clear the credential pool cache too. The cache key is provider_id
@@ -1856,10 +1916,11 @@ def invalidate_provider_models_cache(provider_id: str):
     Args:
         provider_id: canonical provider id (e.g. 'openai', 'anthropic', 'custom:my-key')
     """
-    global _available_models_cache, _available_models_cache_ts, _CREDENTIAL_POOL_CACHE
+    global _available_models_cache, _available_models_cache_ts, _available_models_cache_source_fingerprint, _CREDENTIAL_POOL_CACHE
     with _available_models_cache_lock:
         _available_models_cache = None
         _available_models_cache_ts = 0.0
+        _available_models_cache_source_fingerprint = None
         _provider_models_invalidated_ts[provider_id] = time.time()
         # Also evict the credential pool so the next cold path re-loads it.
         # Must evict both the original key and its canonical form (load_pool
@@ -1902,6 +1963,47 @@ def _get_label_for_model(model_id: str, existing_groups: list) -> str:
     )
 
 
+def _read_visible_codex_cache_model_ids() -> list[str]:
+    """Return visible model slugs from Codex's local models_cache.json.
+
+    The agent's provider_model_ids('openai-codex') intentionally filters IDs
+    with ``supported_in_api: false``. Codex CLI still lists some of those models
+    in its picker (notably ``gpt-5.3-codex-spark`` from #1680), so the WebUI
+    merges this visible local catalog to stay in sync with Codex itself.
+    """
+    codex_home = Path(os.getenv("CODEX_HOME", "").strip() or (HOME / ".codex")).expanduser()
+    cache_path = codex_home / "models_cache.json"
+    try:
+        payload = json.loads(cache_path.read_text(encoding="utf-8"))
+    except Exception:
+        return []
+
+    entries = payload.get("models") if isinstance(payload, dict) else None
+    if not isinstance(entries, list):
+        return []
+
+    sortable: list[tuple[int, str]] = []
+    for item in entries:
+        if not isinstance(item, dict):
+            continue
+        slug = item.get("slug")
+        if not isinstance(slug, str) or not slug.strip():
+            continue
+        visibility = item.get("visibility", "")
+        if isinstance(visibility, str) and visibility.strip().lower() in ("hide", "hidden"):
+            continue
+        priority = item.get("priority")
+        rank = int(priority) if isinstance(priority, (int, float)) else 10_000
+        sortable.append((rank, slug.strip()))
+
+    sortable.sort(key=lambda item: (item[0], item[1]))
+    ordered: list[str] = []
+    for _, slug in sortable:
+        if slug not in ordered:
+            ordered.append(slug)
+    return ordered
+
+
 def get_available_models() -> dict:
     """
     Return available models grouped by provider.
@@ -1918,7 +2020,7 @@ def get_available_models() -> dict:
         'groups': [{'provider': str, 'models': [{'id': str, 'label': str}]}]
     }
     """
-    global _cache_build_in_progress, _available_models_cache, _available_models_cache_ts, _cache_build_cv
+    global _cache_build_in_progress, _available_models_cache, _available_models_cache_ts, _available_models_cache_source_fingerprint, _cache_build_cv
     # Config mtime check — must come before any config reads.
     # (Test #585 verifies _current_mtime appears before active_provider = None)
     try:
@@ -2053,12 +2155,7 @@ def get_available_models() -> dict:
 
         # 2. Read auth store (active_provider fallback + credential_pool inspection)
         auth_store = {}
-        try:
-            from api.profiles import get_active_hermes_home as _gah
-
-            auth_store_path = _gah() / "auth.json"
-        except ImportError:
-            auth_store_path = HOME / ".hermes" / "auth.json"
+        auth_store_path = _get_auth_store_path()
         if auth_store_path.exists():
             try:
                 import json as _j
@@ -2671,6 +2768,43 @@ def get_available_models() -> dict:
                     except Exception:
                         logger.warning("Failed to load Ollama Cloud models from hermes_cli")
 
+                    if raw_models:
+                        models = _apply_provider_prefix(raw_models, pid, active_provider)
+                        groups.append(
+                            {
+                                "provider": provider_name,
+                                "provider_id": pid,
+                                "models": models,
+                            }
+                        )
+                elif pid == "openai-codex":
+                    # Codex account catalogs drift faster than WebUI releases
+                    # (for example gpt-5.3-codex-spark in #1680). Ask the
+                    # agent's Codex resolver first so /api/models inherits the
+                    # live Codex API / local ~/.codex cache / static fallback
+                    # chain instead of freezing the picker to WebUI's curated
+                    # _PROVIDER_MODELS snapshot.
+                    raw_models = []
+                    codex_ids = []
+                    try:
+                        from hermes_cli.models import provider_model_ids as _provider_model_ids
+
+                        codex_ids = [mid for mid in (_provider_model_ids("openai-codex") or []) if mid]
+                    except Exception:
+                        logger.warning("Failed to load OpenAI Codex models from hermes_cli")
+
+                    for mid in _read_visible_codex_cache_model_ids():
+                        if mid not in codex_ids:
+                            codex_ids.append(mid)
+
+                    raw_models = [
+                        {"id": mid, "label": _get_label_for_model(mid, [])}
+                        for mid in codex_ids
+                    ]
+
+                    if not raw_models:
+                        raw_models = copy.deepcopy(_PROVIDER_MODELS.get("openai-codex", []))
+
                     if raw_models:
                         models = _apply_provider_prefix(raw_models, pid, active_provider)
                         groups.append(
@@ -2939,6 +3073,7 @@ def get_available_models() -> dict:
             reload_config()
             _available_models_cache = None
             _available_models_cache_ts = 0.0
+            _available_models_cache_source_fingerprint = None
             disk_groups = None
 
         # Serve from memory cache if fresh
@@ -2951,6 +3086,7 @@ def get_available_models() -> dict:
         if disk_groups is not None:
             _available_models_cache = disk_groups
             _available_models_cache_ts = now
+            _available_models_cache_source_fingerprint = _models_cache_source_fingerprint()
             _save_models_cache_to_disk(disk_groups)
             return copy.deepcopy(disk_groups)
 
@@ -2968,6 +3104,7 @@ def get_available_models() -> dict:
         with _cache_build_cv:
             _available_models_cache = result
             _available_models_cache_ts = time.monotonic()
+            _available_models_cache_source_fingerprint = _models_cache_source_fingerprint()
             _cache_build_in_progress = False
             _cache_build_cv.notify_all()
         _save_models_cache_to_disk(result)
diff --git a/api/profiles.py b/api/profiles.py
index b56868a1..52c552b2 100644
--- a/api/profiles.py
+++ b/api/profiles.py
@@ -37,6 +37,13 @@ _loaded_profile_env_keys: set[str] = set()
 # process-global _active_profile.
 _tls = threading.local()
 
+def _unwrap_profile_home_to_base(home: Path) -> Path:
+    """Return the base Hermes home when *home* is already a named profile dir."""
+    if home.parent.name == 'profiles':
+        return home.parent.parent
+    return home
+
+
 def _resolve_base_hermes_home() -> Path:
     """Return the BASE ~/.hermes directory — the root that contains profiles/.
 
@@ -56,20 +63,22 @@ def _resolve_base_hermes_home() -> Path:
     reading it here would make _DEFAULT_HERMES_HOME point to that subdir,
     causing switch_profile('webui') to look for
     /home/user/.hermes/profiles/webui/profiles/webui — which doesn't exist.
+
+    HERMES_BASE_HOME normally points at the base home already, but isolated
+    single-profile WebUI deployments can provide /base/profiles/<name> there as
+    well.  Normalize both env vars through the same helper so active-profile
+    and per-request resolution share one base-root contract (#749).
     """
     # Explicit override for tests or unusual setups
     base_override = os.getenv('HERMES_BASE_HOME', '').strip()
     if base_override:
-        return Path(base_override).expanduser()
+        return _unwrap_profile_home_to_base(Path(base_override).expanduser())
 
     hermes_home = os.getenv('HERMES_HOME', '').strip()
     if hermes_home:
         p = Path(hermes_home).expanduser()
         # If HERMES_HOME points to a profiles/ subdir, walk up two levels to the base
-        if p.parent.name == 'profiles':
-            return p.parent.parent
-        # Otherwise trust it (e.g. test isolation sets HERMES_HOME to TEST_STATE_DIR)
-        return p
+        return _unwrap_profile_home_to_base(p)
 
     return Path.home() / '.hermes'
 
@@ -193,19 +202,29 @@ def clear_request_profile() -> None:
     _tls.profile = None
 
 
+def _resolve_profile_home_for_name(name: str) -> Path:
+    """Resolve a logical profile name to its Hermes home path.
+
+    Root/default aliases resolve to _DEFAULT_HERMES_HOME.  Valid named profiles
+    resolve to _DEFAULT_HERMES_HOME/profiles/<name> even when the directory has
+    not been created yet; the agent layer may create it on first use.  Invalid
+    names fall back to the base home so traversal-shaped cookie values cannot
+    influence filesystem paths.
+    """
+    if not name or _is_root_profile(name):
+        return _DEFAULT_HERMES_HOME
+    if not _PROFILE_ID_RE.fullmatch(name):
+        return _DEFAULT_HERMES_HOME
+    return _resolve_named_profile_home(name)
+
+
 def get_active_hermes_home() -> Path:
     """Return the HERMES_HOME path for the currently active profile.
 
     Uses get_active_profile_name() so per-request TLS context (issue #798)
     is respected, not just the process-level global.
     """
-    name = get_active_profile_name()
-    if _is_root_profile(name):
-        return _DEFAULT_HERMES_HOME
-    profile_dir = _DEFAULT_HERMES_HOME / 'profiles' / name
-    if profile_dir.is_dir():
-        return profile_dir
-    return _DEFAULT_HERMES_HOME
+    return _resolve_profile_home_for_name(get_active_profile_name())
 
 
 
@@ -393,12 +412,7 @@ def get_hermes_home_for_profile(name: str) -> Path:
     empty, 'default', or does not match the profile-name format (rejects path
     traversal such as '../../etc').
     """
-    if not name or _is_root_profile(name):
-        return _DEFAULT_HERMES_HOME
-    if not _PROFILE_ID_RE.fullmatch(name):
-        return _DEFAULT_HERMES_HOME
-    profile_dir = _DEFAULT_HERMES_HOME / 'profiles' / name
-    return profile_dir
+    return _resolve_profile_home_for_name(name)
 
 
 _TERMINAL_ENV_MAPPINGS = {
diff --git a/api/streaming.py b/api/streaming.py
index 679f03af..6accd184 100644
--- a/api/streaming.py
+++ b/api/streaming.py
@@ -938,7 +938,12 @@ def _fallback_title_from_exchange(user_text: str, assistant_text: str) -> Option
         'need', 'needs', 'want', 'wants', 'user', 'assistant', 'could', 'would',
         'should', 'about', 'there', 'here', 'test', 'testing', 'title', 'summary',
     }
-    tokens = re.findall(r'[A-Za-z0-9][A-Za-z0-9_./+-]*', head)
+    # Unicode-aware Latin tokenization: keep the old "no leading underscore"
+    # and non-Latin placeholder behavior while allowing letters such as ä/ö/ü/ß.
+    # The previous ASCII-only pattern turned "führe" into "f" + "hre"; the short
+    # "f" was filtered and the broken "hre" became part of the title.
+    latin_word = r'A-Za-z0-9À-ÖØ-öø-ÿ'
+    tokens = re.findall(rf'[{latin_word}][{latin_word}_./+-]*', head)
     if not tokens:
         return 'Conversation topic'
 
@@ -1092,8 +1097,12 @@ def _run_background_title_refresh(session_id: str, user_text: str, assistant_tex
                     return
                 s.title = next_title
                 s.llm_title_generated = True
-                s.save(touch_updated_at=False)
                 effective_title = s.title
+            # Session.save() calls _write_session_index(), which acquires LOCK.
+            # Keep the per-session agent lock for mutation serialization, but
+            # release the global session LOCK before persisting to avoid a
+            # self-deadlock in the background title-refresh thread.
+            s.save(touch_updated_at=False)
         _put_title_status(put_event, session_id, 'refreshed', llm_status, effective_title, raw_preview)
         put_event('title', {'session_id': session_id, 'title': effective_title})
         logger.info("Adaptive title refresh: session=%s new_title=%r", session_id, effective_title)
diff --git a/api/updates.py b/api/updates.py
index 9e1d9f4e..e3e025c2 100644
--- a/api/updates.py
+++ b/api/updates.py
@@ -150,6 +150,25 @@ WEBUI_VERSION: str = _detect_webui_version()
 AGENT_VERSION: str = _detect_agent_version()
 
 
+def _normalize_remote_url(remote_url):
+    """Return the browser-facing repository URL for update compare links.
+
+    Git remotes may be HTTPS or SSH and may include a literal ``.git`` suffix.
+    Strip only that literal suffix — never use ``str.rstrip('.git')`` because it
+    treats the argument as a character set and can truncate ``hermes-webui`` to
+    ``hermes-webu``.
+    """
+    if not remote_url:
+        return remote_url
+    remote_url = remote_url.strip()
+    if remote_url.startswith('git@'):
+        remote_url = remote_url.replace(':', '/', 1).replace('git@', 'https://', 1)
+    remote_url = remote_url.rstrip('/')
+    if remote_url.endswith('.git'):
+        remote_url = remote_url[:-4]
+    return remote_url.rstrip('/')
+
+
 def _split_remote_ref(ref):
     """Split 'origin/branch-name' into ('origin', 'branch-name').
 
@@ -234,11 +253,7 @@ def _check_repo(path, name):
 
     # Get repo URL for "What's new?" link
     remote_url, _ = _run_git(['remote', 'get-url', 'origin'], path)
-    # Convert SSH URLs (git@github.com:org/repo.git) to HTTPS
-    if remote_url and remote_url.startswith('git@'):
-        remote_url = remote_url.replace(':', '/', 1).replace('git@', 'https://', 1)
-    if remote_url and remote_url.endswith('.git'):
-        remote_url = remote_url[:-4]
+    remote_url = _normalize_remote_url(remote_url)
 
     return {
         'name': name,
diff --git a/docs/pr-media/1321/update-network-error.png b/docs/pr-media/1321/update-network-error.png
new file mode 100644
index 00000000..7c438a14
Binary files /dev/null and b/docs/pr-media/1321/update-network-error.png differ
diff --git a/docs/pr-media/1698/workspace-double-click-rename.png b/docs/pr-media/1698/workspace-double-click-rename.png
new file mode 100644
index 00000000..fb1dd9e9
Binary files /dev/null and b/docs/pr-media/1698/workspace-double-click-rename.png differ
diff --git a/docs/pr-media/1699/model-cache-auth-store-refresh.png b/docs/pr-media/1699/model-cache-auth-store-refresh.png
new file mode 100644
index 00000000..beb552f6
Binary files /dev/null and b/docs/pr-media/1699/model-cache-auth-store-refresh.png differ
diff --git a/static/boot.js b/static/boot.js
index d4e8bdd0..62f3c8ab 100644
--- a/static/boot.js
+++ b/static/boot.js
@@ -975,10 +975,12 @@ $('msg').addEventListener('paste',e=>{
   const imageItems=items.filter(i=>i.kind==='file'&&i.type.startsWith('image/'));
   if(!imageItems.length||hasText)return;
   e.preventDefault();
-  const files=imageItems.map(i=>{
+  const pasteTs=Date.now();
+  const files=imageItems.map((i,idx)=>{
     const blob=i.getAsFile();
     const ext=i.type.split('/')[1]||'png';
-    return new File([blob],`screenshot-${Date.now()}.${ext}`,{type:i.type});
+    const suffix=imageItems.length>1?`-${idx+1}`:'';
+    return new File([blob],`screenshot-${pasteTs}${suffix}.${ext}`,{type:i.type});
   });
   addFiles(files);
   setStatus(t('image_pasted')+files.map(f=>f.name).join(', '));
diff --git a/static/style.css b/static/style.css
index c08141d1..6a9c70d0 100644
--- a/static/style.css
+++ b/static/style.css
@@ -1420,7 +1420,7 @@
 .model-dropdown{display:none;position:absolute;bottom:calc(100% + 4px);left:0;min-width:280px;max-width:min(420px,calc(100vw - 32px));background:var(--surface);border:1px solid var(--border2);border-radius:10px;box-shadow:0 -4px 24px rgba(0,0,0,.4);z-index:200;overflow:hidden;max-height:320px;overflow-y:auto;}
 .model-dropdown.open{display:block;}
 .model-scope-note{position:sticky;top:0;z-index:2;padding:9px 14px;border-bottom:1px solid var(--border);color:var(--text);font-size:11px;font-weight:650;line-height:1.4;background:color-mix(in srgb,var(--surface) 82%,var(--accent-bg));box-shadow:0 1px 0 rgba(0,0,0,.12);}
-.model-group{padding:8px 14px 4px;font-size:10px;font-weight:700;letter-spacing:.04em;color:var(--muted);text-transform:uppercase;}
+.model-group{padding:8px 14px 4px;font-size:10px;font-weight:700;letter-spacing:.04em;color:var(--muted);text-transform:uppercase;border-top:1px solid var(--border2);margin-top:2px;}
 .model-opt{padding:10px 14px;cursor:pointer;transition:background .12s;display:flex;flex-direction:column;gap:3px;align-items:flex-start;}
 .model-opt:hover{background:rgba(255,255,255,.07);}
 .model-opt.active{background:var(--accent-bg);}
@@ -1430,6 +1430,7 @@
 .model-opt-badge--primary{background:rgba(50,184,198,.16);border-color:rgba(50,184,198,.32);color:#8fe7ef;}
 .model-opt-badge--fallback{background:rgba(255,184,77,.14);border-color:rgba(255,184,77,.28);color:#ffd18a;}
 .model-opt-id{display:block;font-size:10px;color:var(--muted);line-height:1.3;opacity:.72;word-break:break-word;}
+.model-opt-provider{display:inline-flex;align-items:center;padding:1px 6px;border-radius:4px;font-size:9px;font-weight:600;letter-spacing:.03em;color:var(--muted);background:rgba(255,255,255,.05);border:1px solid var(--border2);margin-left:auto;white-space:nowrap;flex-shrink:0;}
 .model-custom-sep{padding-top:4px;border-top:1px solid var(--border);margin-top:4px;}
 .model-custom-row{display:flex;align-items:center;gap:6px;padding:6px 10px 8px;}
 .model-custom-input{flex:1;background:var(--code-bg);border:1px solid var(--border2);border-radius:6px;color:var(--text);padding:5px 8px;font-size:12px;outline:none;font-family:inherit;min-width:0;}
diff --git a/static/ui.js b/static/ui.js
index 4b93eb29..d3329899 100644
--- a/static/ui.js
+++ b/static/ui.js
@@ -50,6 +50,15 @@ function _setCompressionSessionLock(sid){
   window._compressionLockSid=sid||null;
 }
 const esc=s=>String(s??'').replace(/[&<>"']/g,c=>({'&':'&amp;','<':'&lt;','>':'&gt;','"':'&quot;',"'":'&#39;'}[c]));
+function _matchBacktickFenceLine(line){
+  const m=String(line||'').match(/^[ ]{0,3}(`{3,})([^`]*)$/);
+  if(!m) return null;
+  return {fence:m[1],len:m[1].length,info:(m[2]||'').trim()};
+}
+function _isBacktickFenceClose(line,minLen){
+  const m=String(line||'').match(/^[ ]{0,3}(`{3,})[ \t]*$/);
+  return !!(m&&m[1].length>=minLen);
+}
 /**
  * Render fenced code blocks inside user messages.
  * Extracts ```…``` fences, replaces them with placeholders,
@@ -62,9 +71,12 @@ function _renderUserFencedBlocks(text){
   const stash=[];
   let s=String(text||'');
   // Extract fenced code blocks → stash, replace with null-token placeholder
-  // CommonMark line-anchored fence (fixes #1438): inner ``` inside content no longer truncates the block.
-  s=s.replace(/(^|\n)[ ]{0,3}```([a-zA-Z0-9_+-]*)\n(?:([\s\S]*?)\n)?[ ]{0,3}```(?=\n|$)/g,(_,lead,lang,code)=>{
-    lang=(lang||'').trim().toLowerCase();
+  // CommonMark §4.5 line-anchored fence: the closing run must use at least
+  // as many backticks as the opener, so inner triple-backtick fences remain content.
+  s=s.replace(/(^|\n)[ ]{0,3}(`{3,})([^\n`]*)\n(?:([\s\S]*?)\n)?[ ]{0,3}\2`*[ \t]*(?=\n|$)/g,(_,lead,_fence,info,code)=>{
+    const langInfo=(info||'').trim();
+    const langMatch=langInfo.match(/^(\w[\w+-]*)$/);
+    let lang=langMatch?(langMatch[1]||'').trim().toLowerCase():'';
     code=code||'';
     // Remove one trailing newline if present (the fence consumes its own)
     if(code.endsWith('\n')) code=code.slice(0,-1);
@@ -872,19 +884,28 @@ function renderModelDropdown(){
     }
     // Add remaining models matching filter
     let _lastGroup=null;
+    // Count models per group for heading labels (#1425)
+    const _groupCounts={};
+    for(const m of _modelData){
+      if(configuredIds.has(m.value)) continue;
+      if(m.group) _groupCounts[m.group]=(_groupCounts[m.group]||0)+1;
+    }
     for(const m of _modelData){
       if(configuredIds.has(m.value)||!matches(m)) continue;
       if(m.group&&m.group!==_lastGroup){
         const heading=document.createElement('div');
         heading.className='model-group';
-        heading.textContent=m.group;
+        const count=_groupCounts[m.group]||0;
+        heading.textContent=count>1?`${m.group} (${count})`:m.group;
         dd.appendChild(heading);
         _lastGroup=m.group;
       }
       const row=document.createElement('div');
       row.className='model-opt'+(m.value===sel.value?' active':'');
       const badgeHtml=m.badge?`<span class="model-opt-badge model-opt-badge--${esc(m.badge.role||'configured')}">${esc(m.badge.label||'Configured')}</span>`:'';
-      row.innerHTML=`<div class="model-opt-top"><span class="model-opt-name">${m.name}</span>${badgeHtml}</div><span class="model-opt-id">${m.id}</span>`;
+      // Inline provider chip on every row that has a group (#1425)
+      const providerChip=m.group?`<span class="model-opt-provider">${esc(m.group)}</span>`:'';
+      row.innerHTML=`<div class="model-opt-top"><span class="model-opt-name">${m.name}</span>${badgeHtml}${providerChip}</div><span class="model-opt-id">${m.id}</span>`;
       row.onclick=()=>selectModelFromDropdown(m.value);
       dd.appendChild(row);
     }
@@ -1736,7 +1757,8 @@ function renderMd(raw){
   s=(function _applyBlockquotes(input){
     const lines=input.split('\n');
     const out=[];
-    let inFence=false;     // inside a non-blockquote ```...``` fence
+    let inFence=false;     // inside a non-blockquote backtick fence
+    let fenceLen=0;
     let bqStart=-1;
     const flush=(end)=>{
       if(bqStart<0) return;
@@ -1759,13 +1781,15 @@ function renderMd(raw){
       const line=lines[i];
       if(inFence){
         out.push(line);
-        if(/^```/.test(line)) inFence=false;
+        if(_isBacktickFenceClose(line,fenceLen)){inFence=false;fenceLen=0;}
         continue;
       }
-      if(/^```/.test(line)){
+      const fenceOpen=_matchBacktickFenceLine(line);
+      if(fenceOpen){
         flush(i);
         out.push(line);
         inFence=true;
+        fenceLen=fenceOpen.len;
         continue;
       }
       if(/^>/.test(line)){
@@ -1809,14 +1833,16 @@ function renderMd(raw){
   const _preBlock_stash=[];
   const fence_stash=[];
   // CommonMark §4.5: opening fence must start a line (with up to 3 spaces of indent)
-  // and closing fence must also start a line. Without line anchoring, a literal ``` inside
-  // a code block (e.g. a regex pattern with ``` in a lookbehind, a script that documents
-  // fences) terminates the outer block at the wrong place, leaking content into the
-  // markdown stream where bold/italic/inline-code passes corrupt it. Fixes #1438.
-  s=s.replace(/(^|\n)[ ]{0,3}```(?:([\s\S]*?)\n)?[ ]{0,3}```(?=\n|$)/g,(_,lead,raw)=>{
-    const m=raw.match(/^(\w[\w+-]*)\n?([\s\S]*)$/);
-    const lang=m?(m[1]||'').trim().toLowerCase():'';
-    const code=m?m[2]:raw.replace(/^\n?/,'');
+  // and closing fence must start a line with the same backtick char and at least
+  // as many backticks as the opener. Without line/fence-length anchoring, a literal
+  // ``` inside a code block (e.g. a nested markdown example) terminates the outer
+  // block at the wrong place, leaking content into the markdown stream where
+  // bold/italic/inline-code passes corrupt it. Fixes #1438 and #1696.
+  s=s.replace(/(^|\n)[ ]{0,3}(`{3,})([^\n`]*)\n(?:([\s\S]*?)\n)?[ ]{0,3}\2`*[ \t]*(?=\n|$)/g,(_,lead,_fence,info,code)=>{
+    const langInfo=(info||'').trim();
+    const langMatch=langInfo.match(/^(\w[\w+-]*)$/);
+    const lang=langMatch?(langMatch[1]||'').trim().toLowerCase():'';
+    code=code||'';
     const codeLines=code.split('\n');
     const firstCodeLine=codeLines.find(line=>line.trim())||'';
     const firstMermaidLine=codeLines.map(line=>line.trim()).find(line=>line&&!line.startsWith('%%'))||'';
@@ -3202,8 +3228,30 @@ function dismissUpdate(){
   const b=$('updateBanner');if(b)b.classList.remove('visible');
   sessionStorage.setItem('hermes-update-dismissed','1');
 }
+function _isUpdateApplyNetworkError(error){
+  if(error && error.status) return false;
+  const message=(error&&error.message)||String(error||'');
+  return /Failed to fetch|NetworkError|Load failed/i.test(message);
+}
+function _formatUpdateApplyExceptionMessage(error){
+  if(_isUpdateApplyNetworkError(error)){
+    return 'Update failed: could not reach the WebUI server. It may have restarted or the connection was interrupted. Please wait a few seconds, reload the page, then check the server if it still does not come back.';
+  }
+  const message=(error&&error.message)||String(error||'unknown error');
+  return 'Update failed: '+message;
+}
 async function applyUpdates(){
+  if(window._updateApplyInFlight) return;
+  window._updateApplyInFlight=true;
   const btn=$('btnApplyUpdate');
+  const resetApplyButton=(delayMs)=>{
+    const reset=()=>{
+      window._updateApplyInFlight=false;
+      if(btn){btn.disabled=false;btn.textContent='Update Now';}
+    };
+    if(delayMs>0) setTimeout(reset,delayMs);
+    else reset();
+  };
   if(btn){btn.disabled=true;btn.textContent='Updating\u2026';}
   const errEl=$('updateError');
   if(errEl){errEl.style.display='none';errEl.textContent='';}
@@ -3219,7 +3267,7 @@ async function applyUpdates(){
       const res=await api('/api/updates/apply',{method:'POST',body:JSON.stringify({target})});
       if(!res.ok){
         _showUpdateError(target,res);
-        if(btn){btn.disabled=false;btn.textContent='Update Now';}
+        resetApplyButton(0);
         return;
       }
     }
@@ -3228,9 +3276,10 @@ async function applyUpdates(){
     sessionStorage.removeItem('hermes-update-dismissed');
     _waitForServerThenReload();
   }catch(e){
-    if(errEl){errEl.textContent='Update failed: '+e.message;errEl.style.display='block';}
-    else showToast('Update failed: '+e.message);
-    if(btn){btn.disabled=false;btn.textContent='Update Now';}
+    const msg=_formatUpdateApplyExceptionMessage(e);
+    if(errEl){errEl.textContent=msg;errEl.style.display='block';}
+    else showToast(msg);
+    resetApplyButton(_isUpdateApplyNetworkError(e)?5000:0);
   }
 }
 function _showUpdateError(target,res){
@@ -5653,6 +5702,7 @@ function _renderTreeItems(container, entries, depth){
     // Name
     const nameEl=document.createElement('span');
     nameEl.className='file-name';nameEl.textContent=item.name;nameEl.title=t('double_click_rename');
+    nameEl.onclick=(e)=>e.stopPropagation();
     nameEl.ondblclick=(e)=>{
       e.stopPropagation();
       // For directories, double-click navigates (breadcrumb view)
diff --git a/tests/test_1058_adaptive_title_refresh.py b/tests/test_1058_adaptive_title_refresh.py
index bfb8afce..8ccd0d03 100644
--- a/tests/test_1058_adaptive_title_refresh.py
+++ b/tests/test_1058_adaptive_title_refresh.py
@@ -287,6 +287,40 @@ class TestRunBackgroundTitleRefresh:
         assert len(title_events) == 1
         assert title_events[0][1]['title'] == 'New Refreshed Title'
 
+    def test_saves_refreshed_title_outside_global_lock(self):
+        """Refreshing an existing title must not call Session.save() while holding LOCK."""
+        class TrackingLock:
+            def __init__(self):
+                self.held = False
+
+            def __enter__(self):
+                assert not self.held
+                self.held = True
+                return self
+
+            def __exit__(self, exc_type, exc, tb):
+                self.held = False
+
+        put, events = self._make_put_event()
+        lock = TrackingLock()
+        s = self._make_session_obj(title='Old Title')
+
+        def save(*args, **kwargs):
+            assert not lock.held, "Session.save() must run outside api.models.LOCK"
+
+        s.save = save
+        fake_sessions = {'sid': s}
+        with patch('api.streaming.get_session', return_value=s), \
+             patch('api.streaming._aux_title_configured', return_value=True), \
+             patch('api.streaming._generate_llm_session_title_via_aux',
+                   return_value=('New Refreshed Title', 'llm_ok', 'raw')), \
+             patch('api.streaming.SESSIONS', fake_sessions), \
+             patch('api.streaming.LOCK', lock):
+            _run_background_title_refresh('sid', 'u', 'a', 'Old Title', put)
+        title_events = [(n, d) for n, d in events if n == 'title']
+        assert len(title_events) == 1
+        assert title_events[0][1]['title'] == 'New Refreshed Title'
+
     def test_exceptions_are_silently_swallowed(self):
         """Any unexpected error inside must not propagate — it's a background daemon."""
         put, events = self._make_put_event()
diff --git a/tests/test_1325_user_fenced_code.py b/tests/test_1325_user_fenced_code.py
index 80c1be39..7b4ec77a 100644
--- a/tests/test_1325_user_fenced_code.py
+++ b/tests/test_1325_user_fenced_code.py
@@ -7,23 +7,30 @@ UI_JS = os.path.join(os.path.dirname(__file__), '..', 'static', 'ui.js')
 
 
 def _extract_js_functions():
-    """Extract esc and _renderUserFencedBlocks from ui.js by line numbers."""
-    lines = open(UI_JS).read().split('\n')
-    # esc is on line 52 (0-indexed: 51)
-    esc_def = lines[51]
-    # _renderUserFencedBlocks starts at line 61 (0-indexed: 60)
-    # Find the end by matching closing brace at column 0
-    fn_lines = []
-    i = 60  # 0-indexed
-    depth = 0
-    while i < len(lines):
-        fn_lines.append(lines[i])
-        depth += lines[i].count('{') - lines[i].count('}')
-        if depth <= 0:
-            break
+    """Extract esc, fence helpers, and _renderUserFencedBlocks from ui.js."""
+    src = open(UI_JS).read()
+
+    def extract_function(name):
+        start = src.find(f"function {name}(")
+        if start < 0:
+            raise AssertionError(f"{name} not found in ui.js")
+        i = src.find("{", start)
+        depth = 1
         i += 1
-    fn_def = '\n'.join(fn_lines)
-    return esc_def, fn_def
+        while i < len(src) and depth:
+            if src[i] == "{":
+                depth += 1
+            elif src[i] == "}":
+                depth -= 1
+            i += 1
+        return src[start:i]
+
+    esc_line = next(line for line in src.split("\n") if line.startswith("const esc="))
+    helper_defs = "\n".join(
+        extract_function(name)
+        for name in ("_matchBacktickFenceLine", "_isBacktickFenceClose", "_renderUserFencedBlocks")
+    )
+    return esc_line, helper_defs
 
 
 def _run_user_render(text_input):
@@ -116,6 +123,16 @@ class TestUserFencedBlocks:
         assert '<a ' not in out
         assert 'https://example.com' in out
 
+    def test_four_backtick_outer_fence_preserves_inner_triple_fence(self):
+        """User-message code fences should follow CommonMark fence-length matching too."""
+        out = _run_user_render("````md\n```inner\nfoo\n```\n````")
+        assert out.count("<pre>") == 1
+        assert out.count("</pre>") == 1
+        assert '<div class="pre-header">md</div>' in out
+        assert "```inner" in out
+        assert "foo" in out
+        assert "<br>````" not in out
+
     def test_inline_backticks_not_touched(self):
         """Inline backticks (single backtick, not fenced block) should remain escaped as text."""
         out = _run_user_render("use `var x = 1` here")
diff --git a/tests/test_issue1154_fenced_code_leak.py b/tests/test_issue1154_fenced_code_leak.py
index b1d91b92..30df0e92 100644
--- a/tests/test_issue1154_fenced_code_leak.py
+++ b/tests/test_issue1154_fenced_code_leak.py
@@ -43,6 +43,8 @@ function extractFunc(name) {
   }
   return src.slice(start, i);
 }
+eval(extractFunc('_matchBacktickFenceLine'));
+eval(extractFunc('_isBacktickFenceClose'));
 eval(extractFunc('renderMd'));
 
 let buf = '';
diff --git a/tests/test_issue1438_fence_anchoring.py b/tests/test_issue1438_fence_anchoring.py
index 63f28b28..530a9707 100644
--- a/tests/test_issue1438_fence_anchoring.py
+++ b/tests/test_issue1438_fence_anchoring.py
@@ -199,23 +199,15 @@ def test_inline_code_after_fence():
 
 
 def test_renderMd_fence_regex_is_line_anchored():
-    """The fence regex in renderMd must include `(^|\\n)` opener and `(?=\\n|$)` closer.
-
-    Pattern: (^|\\n)[ ]{0,3}```(?:([\\s\\S]*?)\\n)?[ ]{0,3}```(?=\\n|$)
-    The `(?:...\\n)?` makes the body optional so empty fences (```\\n```) still match.
-    """
-    assert re.search(
-        r"s=s\.replace\(/\(\^\|\\n\)\[ \]\{0,3\}```\(\?:\(\[\\s\\S\]\*\?\)\\n\)\?\[ \]\{0,3\}```\(\?=\\n\|\$\)/g",
-        UI_JS,
-    ), "renderMd fence regex is not line-anchored — regression of #1438"
+    """The fence regex in renderMd must keep line anchoring and fence-length matching."""
+    pattern = r"s=s.replace(/(^|\n)[ ]{0,3}(`{3,})([^\n`]*)\n(?:([\s\S]*?)\n)?[ ]{0,3}\2`*[ \t]*(?=\n|$)/g"
+    assert pattern in UI_JS, "renderMd fence regex lost line anchoring or #1696 fence-length matching"
 
 
 def test_renderUserFencedBlocks_fence_regex_is_line_anchored():
     """The fence regex in _renderUserFencedBlocks must also be line-anchored."""
-    assert re.search(
-        r"s=s\.replace\(/\(\^\|\\n\)\[ \]\{0,3\}```\(\[a-zA-Z0-9_\+\-\]\*\)\\n\(\?:\(\[\\s\\S\]\*\?\)\\n\)\?\[ \]\{0,3\}```\(\?=\\n\|\$\)/g",
-        UI_JS,
-    ), "_renderUserFencedBlocks fence regex is not line-anchored — regression of #1438"
+    pattern = r"s=s.replace(/(^|\n)[ ]{0,3}(`{3,})([^\n`]*)\n(?:([\s\S]*?)\n)?[ ]{0,3}\2`*[ \t]*(?=\n|$)/g"
+    assert UI_JS.count(pattern) >= 2, "render/user fence regexes lost line anchoring or #1696 fence-length matching"
 
 
 def test_stripForTTS_fence_regex_is_line_anchored():
@@ -274,8 +266,10 @@ def test_diff_fence_with_inner_backticks_in_content():
     # Pattern explanation: ui.js source contains literal backslash-n in regex literals
     # (ONE backslash + 'n'). In a Python raw string, r"\\n" compiles to a regex pattern
     # matching ONE literal backslash followed by 'n'.
-    matches = re.findall(r"```\(\?=\\n\|\$\)", UI_JS)
-    assert len(matches) >= 3, (
-        f"all 3 fence sites (renderMd, _renderUserFencedBlocks, _stripForTTS) "
-        f"must have line-anchored close fence; found {len(matches)} occurrences"
+    new_matches = UI_JS.count(r"[ ]{0,3}\2`*[ \t]*(?=\n|$)")
+    old_tts_matches = re.findall(r"```\(\?=\\n\|\$\)", UI_JS)
+    assert new_matches >= 2 and len(old_tts_matches) >= 1, (
+        f"renderMd/_renderUserFencedBlocks must have fence-length-aware line-anchored "
+        f"closers and _stripForTTS must keep a line-anchored closer; found "
+        f"new={new_matches}, tts={len(old_tts_matches)}"
     )
diff --git a/tests/test_issue1446_glued_heading_lift.py b/tests/test_issue1446_glued_heading_lift.py
index d760c34e..a9ef33f6 100644
--- a/tests/test_issue1446_glued_heading_lift.py
+++ b/tests/test_issue1446_glued_heading_lift.py
@@ -208,6 +208,8 @@ function extractFunc(name) {
   }
   return src.slice(start, i);
 }
+eval(extractFunc('_matchBacktickFenceLine'));
+eval(extractFunc('_isBacktickFenceClose'));
 eval(extractFunc('renderMd'));
 
 let buf = '';
diff --git a/tests/test_issue1618_yaml_json_diff_newline_preserve.py b/tests/test_issue1618_yaml_json_diff_newline_preserve.py
index 73c5db9f..61f983ef 100644
--- a/tests/test_issue1618_yaml_json_diff_newline_preserve.py
+++ b/tests/test_issue1618_yaml_json_diff_newline_preserve.py
@@ -145,6 +145,8 @@ function extractFunc(name) {
   }
   return src.slice(start, i);
 }
+eval(extractFunc('_matchBacktickFenceLine'));
+eval(extractFunc('_isBacktickFenceClose'));
 eval(extractFunc('renderMd'));
 
 let buf = '';
diff --git a/tests/test_issue1633_models_cache_version_stamp.py b/tests/test_issue1633_models_cache_version_stamp.py
index b06c0294..772ed127 100644
--- a/tests/test_issue1633_models_cache_version_stamp.py
+++ b/tests/test_issue1633_models_cache_version_stamp.py
@@ -214,6 +214,7 @@ def test_load_skips_version_check_when_runtime_unknown(isolated_cache, monkeypat
     # Write a cache that's correct except has no _webui_version
     cache = {
         "_schema_version": config._MODELS_CACHE_SCHEMA_VERSION,
+        "_source_fingerprint": config._models_cache_source_fingerprint(),
         # no _webui_version
         **_shape_cache(),
     }
@@ -268,6 +269,7 @@ def test_is_loadable_disk_cache_checks_versions(with_runtime_version):
     good = {
         "_schema_version": config._MODELS_CACHE_SCHEMA_VERSION,
         "_webui_version": "v0.50.293",
+        "_source_fingerprint": config._models_cache_source_fingerprint(),
         **_shape_cache(),
     }
     assert config._is_loadable_disk_cache(good) is True
diff --git a/tests/test_issue1680_codex_spark.py b/tests/test_issue1680_codex_spark.py
new file mode 100644
index 00000000..cee1c573
--- /dev/null
+++ b/tests/test_issue1680_codex_spark.py
@@ -0,0 +1,104 @@
+"""Regression tests for #1680 — Codex model picker uses live Codex discovery."""
+
+import json
+import sys
+import types
+
+from api import config
+
+
+def _flatten_ids(groups):
+    return [m.get("id") for g in groups for m in g.get("models", [])]
+
+
+def _install_fake_hermes_models(monkeypatch, provider_model_ids):
+    hermes_cli = types.ModuleType("hermes_cli")
+    hermes_cli.__path__ = []
+    models = types.ModuleType("hermes_cli.models")
+    models._PROVIDER_ALIASES = {}
+    models.provider_model_ids = provider_model_ids
+    monkeypatch.setitem(sys.modules, "hermes_cli", hermes_cli)
+    monkeypatch.setitem(sys.modules, "hermes_cli.models", models)
+
+
+def _configure_codex(monkeypatch, tmp_path, default="gpt-5.3-codex-spark"):
+    monkeypatch.setattr(config, "_get_config_path", lambda: tmp_path / "missing-config.yaml")
+    monkeypatch.setattr(config, "_models_cache_path", tmp_path / "models_cache.json")
+    monkeypatch.setattr(config, "cfg", {
+        "model": {"provider": "openai-codex", "default": default},
+        "providers": {},
+        "fallback_providers": [],
+    })
+    monkeypatch.setattr(config, "_cfg_mtime", 0.0)
+    config.invalidate_models_cache()
+
+
+def test_openai_codex_group_uses_provider_model_ids_for_spark(monkeypatch, tmp_path):
+    """Codex-only models from the Codex catalog must surface in /api/models.
+
+    The static WebUI fallback chronically drifts.  ``gpt-5.3-codex-spark`` is
+    the regression case from #1680: it is discoverable by the Codex provider
+    resolver but was missing from the picker because get_available_models()
+    copied _PROVIDER_MODELS["openai-codex"] without asking hermes_cli.
+    """
+    calls = []
+
+    def provider_model_ids(provider):
+        calls.append(provider)
+        assert provider == "openai-codex"
+        return ["gpt-5.4", "gpt-5.3-codex-spark", "gpt-5.3-codex"]
+
+    _install_fake_hermes_models(monkeypatch, provider_model_ids)
+    _configure_codex(monkeypatch, tmp_path)
+
+    result = config.get_available_models()
+
+    codex_groups = [g for g in result["groups"] if g.get("provider_id") == "openai-codex"]
+    assert calls == ["openai-codex"]
+    assert codex_groups, "OpenAI Codex group should be present"
+    assert "gpt-5.3-codex-spark" in _flatten_ids(codex_groups)
+    assert codex_groups[0]["models"][0]["label"] == "GPT 5.4"
+
+
+def test_openai_codex_group_merges_visible_codex_cache_models(monkeypatch, tmp_path):
+    """Visible Codex CLI cache models should appear even if API-filtered.
+
+    Michael's local Codex cache lists ``gpt-5.3-codex-spark`` with
+    ``supported_in_api: false``.  The agent helper currently filters those IDs
+    out, but the WebUI picker is a Codex-model selection surface and should
+    mirror the visible Codex catalog instead of hiding Spark.
+    """
+    def provider_model_ids(provider):
+        assert provider == "openai-codex"
+        return ["gpt-5.4", "gpt-5.3-codex"]
+
+    _install_fake_hermes_models(monkeypatch, provider_model_ids)
+    _configure_codex(monkeypatch, tmp_path, default="gpt-5.4")
+
+    codex_home = tmp_path / "codex-home"
+    codex_home.mkdir()
+    (codex_home / "models_cache.json").write_text(
+        json.dumps(
+            {
+                "models": [
+                    {"slug": "gpt-5.4", "visibility": "list", "priority": 0},
+                    {
+                        "slug": "gpt-5.3-codex-spark",
+                        "visibility": "list",
+                        "supported_in_api": False,
+                        "priority": 7,
+                    },
+                    {"slug": "hidden-test-model", "visibility": "hide", "priority": 8},
+                ]
+            }
+        ),
+        encoding="utf-8",
+    )
+    monkeypatch.setenv("CODEX_HOME", str(codex_home))
+
+    result = config.get_available_models()
+
+    codex_groups = [g for g in result["groups"] if g.get("provider_id") == "openai-codex"]
+    ids = _flatten_ids(codex_groups)
+    assert "gpt-5.3-codex-spark" in ids
+    assert "hidden-test-model" not in ids
diff --git a/tests/test_issue1697_multi_image_paste.py b/tests/test_issue1697_multi_image_paste.py
new file mode 100644
index 00000000..250cb155
--- /dev/null
+++ b/tests/test_issue1697_multi_image_paste.py
@@ -0,0 +1,142 @@
+"""Regression coverage for #1697: multi-image clipboard paste attachments."""
+import json
+import shutil
+import subprocess
+from pathlib import Path
+
+import pytest
+
+
+REPO_ROOT = Path(__file__).parent.parent.resolve()
+BOOT_JS_PATH = REPO_ROOT / "static" / "boot.js"
+PANELS_JS_PATH = REPO_ROOT / "static" / "panels.js"
+NODE = shutil.which("node")
+
+pytestmark = pytest.mark.skipif(NODE is None, reason="node not on PATH")
+
+
+def _read_js(path: Path) -> str:
+    with open(path, encoding="utf-8") as f:
+        return f.read()
+
+
+def _extract_msg_paste_registration() -> str:
+    boot = _read_js(BOOT_JS_PATH)
+    marker = "$('msg').addEventListener('paste',e=>{"
+    start = boot.find(marker)
+    assert start >= 0, "boot.js must register the composer paste handler"
+    end_marker = "\n});"
+    end = boot.find(end_marker, start)
+    assert end >= 0, "composer paste handler should end with a listener close"
+    return boot[start : end + len(end_marker)]
+
+
+def _run_node(source: str) -> str:
+    result = subprocess.run(
+        [NODE],
+        input=source,
+        text=True,
+        capture_output=True,
+        cwd=REPO_ROOT,
+        timeout=20,
+        check=False,
+    )
+    if result.returncode != 0:
+        raise RuntimeError(f"node driver failed:\nSTDOUT:\n{result.stdout}\nSTDERR:\n{result.stderr}")
+    return result.stdout.strip()
+
+
+def _paste_harness(items_js: str) -> dict:
+    paste_registration = json.dumps(_extract_msg_paste_registration())
+    source = f"""
+const vm = require('vm');
+const pasteRegistration = {paste_registration};
+const listeners = {{}};
+const S = {{pendingFiles: []}};
+let renderCount = 0;
+let lastStatus = '';
+let preventDefaultCount = 0;
+class File extends Blob {{
+  constructor(parts, name, options={{}}) {{
+    super(parts, options);
+    this.name = name;
+    this.lastModified = options.lastModified || 0;
+  }}
+}}
+const context = {{
+  S,
+  File,
+  Blob,
+  Date: {{now: () => 1700000000000}},
+  Array,
+  console,
+  $: (id) => {{
+    if (id !== 'msg') throw new Error('unexpected element id '+id);
+    return {{addEventListener: (type, cb) => {{listeners[type] = cb;}}}};
+  }},
+  addFiles: (files) => {{
+    for (const f of files) {{
+      if (!S.pendingFiles.find(p => p.name === f.name)) S.pendingFiles.push(f);
+    }}
+    renderCount += 1;
+  }},
+  setStatus: (text) => {{ lastStatus = text; }},
+  t: (key) => key === 'image_pasted' ? 'Image pasted: ' : key,
+}};
+vm.createContext(context);
+vm.runInContext(pasteRegistration, context);
+listeners.paste({{
+  clipboardData: {{items: {items_js}}},
+  preventDefault: () => {{ preventDefaultCount += 1; }},
+}});
+console.log(JSON.stringify({{
+  pendingNames: S.pendingFiles.map(f => f.name),
+  pendingCount: S.pendingFiles.length,
+  renderCount,
+  lastStatus,
+  preventDefaultCount,
+}}));
+"""
+    return json.loads(_run_node(source))
+
+
+def test_one_clipboard_paste_with_two_image_items_adds_two_attachment_chips():
+    """Two image clipboard items from one paste must survive addFiles() filename de-dupe."""
+    result = _paste_harness(
+        "["
+        "{kind:'file', type:'image/png', getAsFile:()=>new Blob(['one'], {type:'image/png'})},"
+        "{kind:'file', type:'image/png', getAsFile:()=>new Blob(['two'], {type:'image/png'})}"
+        "]"
+    )
+
+    assert result["preventDefaultCount"] == 1
+    assert result["renderCount"] == 1
+    assert result["pendingCount"] == 2
+    assert result["pendingNames"] == [
+        "screenshot-1700000000000-1.png",
+        "screenshot-1700000000000-2.png",
+    ]
+    assert result["lastStatus"] == (
+        "Image pasted: screenshot-1700000000000-1.png, "
+        "screenshot-1700000000000-2.png"
+    )
+
+
+def test_single_image_paste_keeps_existing_screenshot_filename_shape():
+    """The one-image path should keep screenshot-<timestamp>.<ext> for compatibility."""
+    result = _paste_harness(
+        "[{kind:'file', type:'image/png', getAsFile:()=>new Blob(['one'], {type:'image/png'})}]"
+    )
+
+    assert result["pendingNames"] == ["screenshot-1700000000000.png"]
+
+
+def test_file_picker_and_drop_paths_still_pass_real_file_names_to_addfiles():
+    """Non-clipboard multi-file paths should preserve browser-provided filenames."""
+    boot = _read_js(BOOT_JS_PATH)
+    panels = _read_js(PANELS_JS_PATH)
+
+    assert "$('fileInput').onchange=e=>{addFiles(Array.from(e.target.files));e.target.value='';};" in boot
+    assert "const files=Array.from(e.dataTransfer.files);" in panels
+    assert "if(files.length){addFiles(files);$('msg').focus();}" in panels
+    assert "screenshot-" not in panels[panels.find("document.addEventListener('drop'") : panels.find("document.addEventListener('drop'") + 900]
diff --git a/tests/test_issue1699_model_cache_source_fingerprint.py b/tests/test_issue1699_model_cache_source_fingerprint.py
new file mode 100644
index 00000000..30500eb5
--- /dev/null
+++ b/tests/test_issue1699_model_cache_source_fingerprint.py
@@ -0,0 +1,144 @@
+"""Regression tests for #1699: /api/models cache must track external auth/config changes.
+
+The bug: WebUI caches /api/models for 24h in memory and on disk. When a user
+runs `hermes setup` in a terminal and the Hermes auth store switches the active
+provider outside WebUI, the browser can keep seeing the previous provider's
+PRIMARY badge until the cache is manually cleared or expires.
+"""
+
+import json
+import sys
+import time
+import types
+
+import api.config as config
+
+
+def _reset_memory_cache() -> None:
+    with config._available_models_cache_lock:
+        config._available_models_cache = None
+        config._available_models_cache_ts = 0.0
+        if hasattr(config, "_available_models_cache_source_fingerprint"):
+            config._available_models_cache_source_fingerprint = None
+        config._cache_build_in_progress = False
+        config._cache_build_cv.notify_all()
+
+
+def _valid_models_cache(provider_id: str, model_id: str) -> dict:
+    return {
+        "active_provider": provider_id,
+        "default_model": model_id,
+        "configured_model_badges": {
+            model_id: {"role": "primary", "label": "Primary", "provider": provider_id}
+        },
+        "groups": [
+            {
+                "provider": config._PROVIDER_DISPLAY.get(provider_id, provider_id.title()),
+                "provider_id": provider_id,
+                "models": [{"id": model_id, "label": model_id}],
+            }
+        ],
+    }
+
+
+def _write_auth_store(hermes_home, provider_id: str) -> None:
+    hermes_home.mkdir(parents=True, exist_ok=True)
+    (hermes_home / "auth.json").write_text(
+        json.dumps({"active_provider": provider_id, "credential_pool": {}}),
+        encoding="utf-8",
+    )
+
+
+def _configure_isolated_sources(tmp_path, monkeypatch, provider_id: str) -> None:
+    hermes_home = tmp_path / "hermes-home"
+    state_dir = tmp_path / "state"
+    cache_path = state_dir / "models_cache.json"
+    state_dir.mkdir(parents=True, exist_ok=True)
+
+    hermes_home.mkdir(parents=True, exist_ok=True)
+    config_path = hermes_home / "config.yaml"
+    # Leave model.provider unset so get_available_models() must honor the auth
+    # store's active_provider fallback, matching CLI setup/auth-store drift.
+    config_path.write_text("model:\n  default: glm-5.1\n", encoding="utf-8")
+    monkeypatch.setenv("HERMES_CONFIG_PATH", str(config_path))
+
+    import api.profiles as profiles
+
+    monkeypatch.setattr(profiles, "get_active_hermes_home", lambda: hermes_home)
+    monkeypatch.setattr(config, "_models_cache_path", cache_path)
+
+    # Keep the test hermetic without requiring hermes-agent to be installed in
+    # CI: inject the tiny hermes_cli surface get_available_models() imports.
+    fake_pkg = types.ModuleType("hermes_cli")
+    fake_pkg.__path__ = []
+    fake_models = types.ModuleType("hermes_cli.models")
+    fake_models._PROVIDER_ALIASES = {}
+    fake_models.list_available_providers = lambda: []
+    fake_auth = types.ModuleType("hermes_cli.auth")
+    fake_auth.get_auth_status = lambda provider_id: {
+        "logged_in": False,
+        "key_source": "",
+    }
+    monkeypatch.setitem(sys.modules, "hermes_cli", fake_pkg)
+    monkeypatch.setitem(sys.modules, "hermes_cli.models", fake_models)
+    monkeypatch.setitem(sys.modules, "hermes_cli.auth", fake_auth)
+
+    _write_auth_store(hermes_home, provider_id)
+    config.reload_config()
+    _reset_memory_cache()
+
+
+def test_memory_models_cache_invalidates_when_auth_store_active_provider_changes(
+    tmp_path, monkeypatch
+):
+    _configure_isolated_sources(tmp_path, monkeypatch, "opencode-go")
+
+    stale_openrouter = _valid_models_cache("openrouter", "minimax-m2.7")
+    with config._available_models_cache_lock:
+        config._available_models_cache = stale_openrouter
+        config._available_models_cache_ts = time.monotonic()
+        if hasattr(config, "_available_models_cache_source_fingerprint"):
+            # Simulate a cache populated before the external CLI auth-store write.
+            config._available_models_cache_source_fingerprint = {
+                "auth_json": {"path": "old-auth.json", "mtime_ns": 1, "size": 10},
+                "config_yaml": {"path": "old-config.yaml", "mtime_ns": 1, "size": 10},
+            }
+
+    result = config.get_available_models()
+
+    assert result["active_provider"] == "opencode-go"
+    assert not any(group.get("provider_id") == "openrouter" for group in result["groups"])
+    assert any(group.get("provider_id") == "opencode-go" for group in result["groups"])
+
+
+def test_disk_models_cache_invalidates_when_auth_store_active_provider_changes(
+    tmp_path, monkeypatch
+):
+    _configure_isolated_sources(tmp_path, monkeypatch, "openrouter")
+    stale_openrouter = _valid_models_cache("openrouter", "minimax-m2.7")
+    config._save_models_cache_to_disk(stale_openrouter)
+    assert config._models_cache_path.exists()
+
+    # External terminal `hermes setup` changes auth.json, not WebUI's in-process cache.
+    hermes_home = config._models_cache_path.parent.parent / "hermes-home"
+    _write_auth_store(hermes_home, "opencode-go")
+    _reset_memory_cache()
+
+    result = config.get_available_models()
+
+    assert result["active_provider"] == "opencode-go"
+    assert not any(group.get("provider_id") == "openrouter" for group in result["groups"])
+    assert any(group.get("provider_id") == "opencode-go" for group in result["groups"])
+
+
+def test_disk_models_cache_still_loads_when_auth_and_config_sources_are_unchanged(
+    tmp_path, monkeypatch
+):
+    _configure_isolated_sources(tmp_path, monkeypatch, "opencode-go")
+    fresh_opencode = _valid_models_cache("opencode-go", "glm-5.1")
+    config._save_models_cache_to_disk(fresh_opencode)
+    _reset_memory_cache()
+
+    result = config.get_available_models()
+
+    assert result == fresh_opencode
diff --git a/tests/test_issue798.py b/tests/test_issue798.py
index 4207400b..37889688 100644
--- a/tests/test_issue798.py
+++ b/tests/test_issue798.py
@@ -9,7 +9,9 @@ get_hermes_home_for_profile() resolves a HERMES_HOME path from a name without
 touching os.environ or module-level state.
 """
 
+import json
 import os
+import subprocess
 import sys
 import threading
 from pathlib import Path
@@ -71,6 +73,96 @@ def test_get_hermes_home_for_profile_does_not_mutate_globals():
     )
 
 
+def _run_profile_resolution_probe(env):
+    script = r'''
+import json
+from pathlib import Path
+import api.profiles as p
+import api.models as m
+
+p.set_request_profile('foo')
+foo_home = p.get_active_hermes_home()
+explicit_foo_home = p.get_hermes_home_for_profile('foo')
+foo_runtime = p.get_profile_runtime_env(explicit_foo_home)
+model_home = m._get_profile_home('foo')
+explicit_bar_home = p.get_hermes_home_for_profile('bar')
+p.set_request_profile('bar')
+active_bar_home = p.get_active_hermes_home()
+print(json.dumps({
+    'default_home': str(p._DEFAULT_HERMES_HOME),
+    'foo_home': str(foo_home),
+    'explicit_foo_home': str(explicit_foo_home),
+    'foo_terminal_cwd': foo_runtime.get('TERMINAL_CWD'),
+    'model_home': str(model_home),
+    'explicit_bar_home': str(explicit_bar_home),
+    'active_bar_home': str(active_bar_home),
+}))
+'''
+    result = subprocess.run(
+        [sys.executable, '-c', script],
+        cwd=Path(__file__).parent.parent,
+        env=env,
+        text=True,
+        capture_output=True,
+        check=True,
+    )
+    return json.loads(result.stdout)
+
+
+def test_hermes_base_home_named_profile_matches_cookie_without_doubling(tmp_path):
+    """R19k / #749: HERMES_BASE_HOME may point directly at a named profile home.
+
+    A single-profile WebUI deployment can start with both HERMES_BASE_HOME and
+    HERMES_HOME set to /base/profiles/foo while the browser still sends the
+    logical cookie hermes_profile=foo.  Both active-profile and explicit
+    per-request helpers must use /base/profiles/foo, not the doubled
+    /base/profiles/foo/profiles/foo path — even if that nested path already
+    exists from a prior bad write.
+    """
+    profile_home = tmp_path / 'profiles' / 'foo'
+    doubled_home = profile_home / 'profiles' / 'foo'
+    doubled_home.mkdir(parents=True)
+    profile_home.joinpath('config.yaml').write_text(
+        'terminal:\n  cwd: /expected/profile-home\n', encoding='utf-8'
+    )
+    doubled_home.joinpath('config.yaml').write_text(
+        'terminal:\n  cwd: /wrong/doubled-home\n', encoding='utf-8'
+    )
+
+    env = os.environ.copy()
+    env.update({
+        'HERMES_BASE_HOME': str(profile_home),
+        'HERMES_HOME': str(profile_home),
+    })
+    data = _run_profile_resolution_probe(env)
+
+    assert data['default_home'] == str(tmp_path)
+    assert data['foo_home'] == str(profile_home)
+    assert data['explicit_foo_home'] == str(profile_home)
+    assert data['foo_terminal_cwd'] == '/expected/profile-home'
+    assert data['model_home'] == str(profile_home)
+
+
+def test_hermes_base_home_named_profile_nonmatching_cookie_uses_sibling_profile_path(tmp_path):
+    """R19l / #749: non-matching cookies must not silently route to the pinned home.
+
+    When HERMES_BASE_HOME is supplied as /base/profiles/foo but the request asks
+    for logical profile bar, preserving base semantics means bar resolves to the
+    sibling /base/profiles/bar.  It must not fall back to foo, and it must not
+    append bar under foo/profiles/bar.
+    """
+    profile_home = tmp_path / 'profiles' / 'foo'
+    profile_home.mkdir(parents=True)
+
+    env = os.environ.copy()
+    env.update({'HERMES_BASE_HOME': str(profile_home)})
+    data = _run_profile_resolution_probe(env)
+
+    expected_bar_home = tmp_path / 'profiles' / 'bar'
+    assert data['explicit_bar_home'] == str(expected_bar_home)
+    assert data['active_bar_home'] == str(expected_bar_home)
+
+
 # ── R19e-h: new_session() profile isolation ───────────────────────────────────
 # These tests call new_session() directly in-process.  Session.save() would write
 # to SESSION_DIR which is set from HERMES_WEBUI_STATE_DIR at import time and may
diff --git a/tests/test_renderer_js_behaviour.py b/tests/test_renderer_js_behaviour.py
index 102c69df..22a831b7 100644
--- a/tests/test_renderer_js_behaviour.py
+++ b/tests/test_renderer_js_behaviour.py
@@ -54,6 +54,8 @@ function extractFunc(name) {
   }
   return src.slice(start, i);
 }
+eval(extractFunc('_matchBacktickFenceLine'));
+eval(extractFunc('_isBacktickFenceClose'));
 eval(extractFunc('renderMd'));
 
 let buf = '';
@@ -285,6 +287,49 @@ class TestBugFencedCodeInBlockquote:
         assert "x = 1" in out
 
 
+class TestFencedCodeFenceLength:
+    """CommonMark §4.5 requires the closer to be at least as long as the opener."""
+
+    def test_five_backtick_outer_fence_preserves_inner_triple_fence(self, driver_path):
+        src = (
+            "- optionally also support fenced code blocks\n\n"
+            "`````md\n"
+            "`md\n"
+            "```novelcrafter\n"
+            "{#if novel.hasSeries}\n"
+            "...\n"
+            "{#endif}\n"
+            "```\n"
+            "`````\n\n"
+            "That is much more correct than pretending"
+        )
+        out = _render(driver_path, src)
+        assert out.count("<pre>") == 1
+        assert out.count("</pre>") == 1
+        assert '<div class="pre-header">md</div>' in out
+        assert "```novelcrafter" in out
+        assert "{#if novel.hasSeries}" in out
+        assert "That is much more correct than pretending" in out
+        assert "<p>`````" not in out
+        assert "<br>`````" not in out
+
+    def test_four_backtick_outer_fence_preserves_inner_triple_fence(self, driver_path):
+        out = _render(driver_path, "````md\n```inner\nfoo\n```\n````\n")
+        assert out.count("<pre>") == 1
+        assert out.count("</pre>") == 1
+        assert '<div class="pre-header">md</div>' in out
+        assert "```inner" in out
+        assert "foo" in out
+        assert "<p>````" not in out
+
+    def test_three_backtick_fence_still_renders_language_class(self, driver_path):
+        out = _render(driver_path, "```js\nconsole.log('ok')\n```")
+        assert out.count("<pre>") == 1
+        assert '<div class="pre-header">js</div>' in out
+        assert 'class="language-js"' in out
+        assert "console.log(&#39;ok&#39;)" in out
+
+
 class TestBugBlankContinuationInBlockquote:
     """Bug 2: blank > lines between paragraphs fragmented the blockquote into
     separate elements with literal > characters between them."""
diff --git a/tests/test_sprint16.py b/tests/test_sprint16.py
index 0b7b309a..fd9882bc 100644
--- a/tests/test_sprint16.py
+++ b/tests/test_sprint16.py
@@ -61,8 +61,8 @@ def render_md(raw):
         fence_stash.append(m.group())
         return "\x00F" + str(len(fence_stash) - 1) + "\x00"
 
-    # Fence regex line-anchored to match JS fix for #1438 (allows empty fence)
-    s = re.sub(r"(?:^|\n)[ ]{0,3}```(?:[\s\S]*?\n)?[ ]{0,3}```(?=\n|$)|`[^`\n]+`", stash, s)
+    # Fence regex line-anchored to match JS fix for #1438 and fence-length fix for #1696
+    s = re.sub(r"(?:^|\n)[ ]{0,3}(`{3,})[^\n`]*\n(?:[\s\S]*?\n)?[ ]{0,3}\1`*(?=\n|$)|`[^`\n]+`", stash, s)
     s = re.sub(r"<strong>([\s\S]*?)</strong>", lambda m: "**" + m.group(1) + "**", s, flags=re.I)
     s = re.sub(r"<b>([\s\S]*?)</b>",           lambda m: "**" + m.group(1) + "**", s, flags=re.I)
     s = re.sub(r"<em>([\s\S]*?)</em>",          lambda m: "*"  + m.group(1) + "*",  s, flags=re.I)
@@ -77,11 +77,12 @@ def render_md(raw):
 
     # Fenced code blocks
     def fenced(m):
-        lang, code = m.group(1), (m.group(2) or "").rstrip("\n")
+        info, code = (m.group(2) or "").strip(), (m.group(3) or "").rstrip("\n")
+        lang = info.lower() if re.match(r"^\w[\w+-]*$", info) else ""
         h = f'<div class="pre-header">{esc(lang)}</div>' if lang else ""
         return h + "<pre><code>" + esc(code) + "</code></pre>"
-    # Fenced code blocks (line-anchored, fixes #1438; allows empty fence)
-    s = re.sub(r"(?:^|\n)[ ]{0,3}```([\w+-]*)\n(?:([\s\S]*?)\n)?[ ]{0,3}```(?=\n|$)", fenced, s)
+    # Fenced code blocks (line-anchored, fixes #1438; fence-length matching fixes #1696)
+    s = re.sub(r"(?:^|\n)[ ]{0,3}(`{3,})([^\n`]*)\n(?:([\s\S]*?)\n)?[ ]{0,3}\1`*(?=\n|$)", fenced, s)
     s = re.sub(r"`([^`\n]+)`", lambda m: "<code>" + esc(m.group(1)) + "</code>", s)
 
     # Inline formatting (top-level, outside list items)
@@ -358,6 +359,52 @@ def test_render_md_fenced_code_protects_html(cleanup_test_sessions):
         "Fenced code content was lost after stash/restore"
 
 
+def test_render_md_fenced_code_with_five_backtick_outer_preserves_inner_triples(cleanup_test_sessions):
+    """CommonMark §4.5: a 5-backtick fence must not close at an inner triple fence."""
+    src = (
+        "- optionally also support fenced code blocks\n\n"
+        "`````md\n"
+        "`md\n"
+        "```novelcrafter\n"
+        "{#if novel.hasSeries}\n"
+        "...\n"
+        "{#endif}\n"
+        "```\n"
+        "`````\n\n"
+        "That is much more correct than pretending"
+    )
+    out = render_md(src)
+    assert out.count("<pre>") == 1
+    assert out.count("</pre>") == 1
+    assert '<div class="pre-header">md</div>' in out
+    assert "```novelcrafter" in out
+    assert "{#if novel.hasSeries}" in out
+    assert "That is much more correct than pretending" in out
+    assert "<p>`````" not in out
+    assert "<br>`````" not in out
+
+
+def test_render_md_fenced_code_with_four_backtick_outer_preserves_inner_triples(cleanup_test_sessions):
+    """A 4-backtick outer fence should also require a 4+ backtick closer."""
+    src = "````md\n```inner\nfoo\n```\n````\n"
+    out = render_md(src)
+    assert out.count("<pre>") == 1
+    assert out.count("</pre>") == 1
+    assert '<div class="pre-header">md</div>' in out
+    assert "```inner" in out
+    assert "foo" in out
+    assert "<p>````" not in out
+
+
+def test_render_md_fenced_code_three_backtick_path_still_renders_language(cleanup_test_sessions):
+    """The common 3-backtick path must keep rendering a single language-tagged block."""
+    src = "```js\nconsole.log('ok')\n```"
+    out = render_md(src)
+    assert out.count("<pre>") == 1
+    assert '<div class="pre-header">js</div>' in out
+    assert "console.log(&#39;ok&#39;)" in out or "console.log(&#x27;ok&#x27;)" in out
+
+
 # ── Security: XSS must be blocked ─────────────────────────────────────────────
 
 def test_render_md_xss_img_tag_escaped(cleanup_test_sessions):
diff --git a/tests/test_sprint41.py b/tests/test_sprint41.py
index f4f6d542..1e5b3983 100644
--- a/tests/test_sprint41.py
+++ b/tests/test_sprint41.py
@@ -327,6 +327,19 @@ class TestIssue495TitleStreaming(unittest.TestCase):
             "Substantive answer text on a tool_call row must be preserved",
         )
 
+    def test_fallback_title_preserves_unicode_letters(self):
+        """Local fallback title generation must not strip German umlauts."""
+        from api.streaming import _fallback_title_from_exchange
+
+        title = _fallback_title_from_exchange(
+            "Bitte führe ein Selbst-Audit durch. Wo ist überall noch Gemini-2.5-flash als Modell im Einsatz? Sei gründlich",
+            "Ich prüfe live statt aus Bauchgefühl.",
+        )
+
+        self.assertIsNotNone(title)
+        self.assertIn("führe", title)
+        self.assertNotIn("hre", title.split())
+
     def test_title_snippet_skips_tool_call_preamble_only_rows(self):
         """Tool-call rows whose content is empty or meta-reasoning preamble
         ('Let me check my memory first.') must still be skipped — those are
diff --git a/tests/test_update_apply_ui.py b/tests/test_update_apply_ui.py
new file mode 100644
index 00000000..c6cc50aa
--- /dev/null
+++ b/tests/test_update_apply_ui.py
@@ -0,0 +1,55 @@
+"""Frontend regression coverage for Update Now apply failures (#1321)."""
+from pathlib import Path
+import re
+
+ROOT = Path(__file__).resolve().parents[1]
+UI_JS = ROOT / "static" / "ui.js"
+
+
+def _ui_js() -> str:
+    return UI_JS.read_text(encoding="utf-8")
+
+
+def test_update_apply_network_error_has_recovery_message_not_raw_failed_to_fetch():
+    """Network/interrupted update apply failures should not surface raw fetch text alone."""
+    src = _ui_js()
+    assert "function _formatUpdateApplyExceptionMessage" in src
+    assert "could not reach the WebUI server" in src
+    assert "restarted or the connection was interrupted" in src
+    assert "wait a few seconds, reload the page, then check the server" in src
+    assert "Update failed: '+e.message" not in src
+    assert 'Update failed: "+e.message' not in src
+
+
+def test_update_apply_structured_server_errors_still_use_json_message_path():
+    """Server-reachable JSON errors must keep the existing targeted message path."""
+    src = _ui_js()
+    apply_start = src.index("async function applyUpdates()")
+    show_error_call = src.index("_showUpdateError(target,res);", apply_start)
+    reset_button = src.index("resetApplyButton(0);", show_error_call)
+    assert show_error_call < reset_button
+    assert "const msg='Update failed ('+target+'): '+(res.message||'unknown error');" in src
+
+
+def test_update_apply_network_error_classifier_ignores_http_status_errors():
+    """HTTP response errors should not be classified as interrupted transport failures."""
+    src = _ui_js()
+    fn_start = src.index("function _isUpdateApplyNetworkError(error)")
+    fn_end = src.index("function _formatUpdateApplyExceptionMessage", fn_start)
+    body = src[fn_start:fn_end]
+    compact = re.sub(r"\s+", "", body)
+    assert "if(error&&error.status)returnfalse;" in compact
+    assert body.index("error.status") < body.index("/Failed to fetch|NetworkError|Load failed/i")
+    assert "Failed to fetch|NetworkError|Load failed" in body
+
+
+def test_update_apply_prevents_duplicate_apply_requests_while_in_flight():
+    """Double-clicks should not send a second update apply request during restart race windows."""
+    src = _ui_js()
+    apply_start = src.index("async function applyUpdates()")
+    next_fn = src.index("function _showUpdateError", apply_start)
+    body = src[apply_start:next_fn]
+    assert "window._updateApplyInFlight" in body
+    assert "if(window._updateApplyInFlight) return;" in body
+    assert "window._updateApplyInFlight=true;" in body
+    assert "window._updateApplyInFlight=false;" in body
diff --git a/tests/test_update_banner_fixes.py b/tests/test_update_banner_fixes.py
index 357dace1..5eaee61e 100644
--- a/tests/test_update_banner_fixes.py
+++ b/tests/test_update_banner_fixes.py
@@ -79,6 +79,31 @@ class TestUpdateChecker:
 
         assert result['repo_url'] == 'https://github.com/NousResearch/hermes-agent'
 
+    def test_repo_url_strips_dot_git_before_trailing_slashes(self, tmp_path, monkeypatch):
+        import api.updates as upd
+
+        (tmp_path / '.git').mkdir()
+
+        def fake_run(args, cwd, timeout=10):
+            if args[0] == 'fetch':
+                return '', True
+            if args[:2] == ['rev-parse', '--abbrev-ref']:
+                return 'origin/master', True
+            if args[:2] == ['rev-list', '--count']:
+                return '2', True
+            if args[0] == 'merge-base':
+                return 'abcdef1234567890', True
+            if args[:2] == ['rev-parse', '--short']:
+                return 'abcdef1', True
+            if args[:2] == ['remote', 'get-url']:
+                return 'https://github.com/nesquena/hermes-webui.git/', True
+            return '', True
+
+        monkeypatch.setattr(upd, '_run_git', fake_run)
+        result = upd._check_repo(tmp_path, 'webui')
+
+        assert result['repo_url'] == 'https://github.com/nesquena/hermes-webui'
+
 
 class TestConflictError:
     """#813 — conflict error must include flag + recovery command."""
diff --git a/tests/test_workspace_tree_rename.py b/tests/test_workspace_tree_rename.py
new file mode 100644
index 00000000..fd6c401a
--- /dev/null
+++ b/tests/test_workspace_tree_rename.py
@@ -0,0 +1,22 @@
+from pathlib import Path
+
+
+REPO_ROOT = Path(__file__).resolve().parents[1]
+UI_JS = (REPO_ROOT / "static" / "ui.js").read_text(encoding="utf-8")
+
+
+def test_workspace_file_name_click_stops_before_dblclick_rename():
+    """Clicking a file name must not bubble to the row open handler before dblclick rename."""
+    name_start = UI_JS.index("const nameEl=document.createElement('span');")
+    dblclick_idx = UI_JS.index("nameEl.ondblclick=(e)=>", name_start)
+    click_idx = UI_JS.find("nameEl.onclick=(e)=>e.stopPropagation();", name_start, dblclick_idx)
+
+    assert click_idx != -1, (
+        "workspace file-tree name span must stop click propagation before its dblclick "
+        "rename handler so the row openFile() click does not win the first click"
+    )
+
+
+def test_workspace_file_row_click_still_opens_file_preview():
+    """Only the name span should swallow clicks; the rest of the file row still opens preview."""
+    assert "el.onclick=async()=>openFile(item.path);" in UI_JS