Merge pull request #3143 from nesquena/release/stage-batch40

Release v0.51.158 — Release ED (stage-batch40: 5-PR low-risk cleanup)

CHANGELOG.md

@@ -3,6 +3,19 @@
 
 ## [Unreleased]
 
+## [v0.51.158] — 2026-05-29 — Release ED (stage-batch40 — 5-PR low-risk cleanup: numpad/keyboard composer fixes + Joplin search auth + provider-qualified model preservation + SSE fallback poll throttle + assistant-reply polish)
+
+### Changed
+
+- WebUI chat instructions now explicitly prevent terse scratchpad/planning fragments from appearing in visible assistant replies, while still allowing clear user-facing progress updates during tool-heavy work.
+
+### Fixed
+
+- The chat composer no longer forces mobile newline-on-Enter behavior on touch-primary devices that also have a fine pointer present (tablet plus Bluetooth keyboard, detachable Surface, iPad plus Magic Keyboard), so Enter submits with desktop semantics when a real keyboard is in the picture.
+- The active-session external-refresh fallback poll now fires every 30 s instead of every 5 s. The SSE session-events stream already pushes invalidations in real time, so the poll is only a fallback; the slower interval removes visible scroll jitter and a network/CPU floor on long sessions.
+- The model picker no longer fuzzy-matches a provider-qualified model id (`@provider:model` or slash-qualified `vendor/model`) to a nearby curated sibling once exact lookup fails, preserving the raw typed value so uncatalogued models stay routable instead of silently snapping to a different model.
+- Joplin notes search now keeps the `Authorization` header and adds a query-token compatibility shim only for Web Clipper `/search` calls, covering clipper builds that return HTTP 403 for header-only search auth while keeping other Joplin API URLs token-free.
+
 ## [v0.51.157] — 2026-05-28 — Release EC (stage-batch39 — 5-PR mixed-risk cleanup: gateway prefill forward + prefill budget + compressed-continuation sidebar + browser-transcript memory guidance + reasoning max parity)
 
 ### Added

api/routes.py

@@ -9628,6 +9628,7 @@ def _handle_chat_sync(handler, body):
                 session_id=s.session_id,
             )
             from api.streaming import (
+                _WEBUI_PROGRESS_PROMPT,
                 _dedupe_replayed_context_messages,
                 _merge_display_messages_after_agent_result,
                 _restore_display_reasoning_metadata,
@@ -9646,7 +9647,8 @@ def _handle_chat_sync(handler, body):
                 "prompt, memory, or conversation history. Always use the value from the most recent "
                 "[Workspace::v1: ...] tag as your default working directory for ALL file operations: "
                 "write_file, read_file, search_files, terminal workdir, and patch. "
-                "Never fall back to a hardcoded path when this tag is present."
+                "Never fall back to a hardcoded path when this tag is present.\n\n"
+                f"{_WEBUI_PROGRESS_PROMPT}"
             )
 
             _previous_messages = list(s.messages or [])
@@ -12969,6 +12971,11 @@ def _joplin_api_get(path: str, params: dict | None = None) -> dict:
         raise ValueError("Joplin token is not configured")
     safe_path = "/" + str(path or "").lstrip("/")
     query = dict(params or {})
+    # Joplin Web Clipper builds can reject header-only auth on /search even when
+    # they accept it elsewhere. Keep the Authorization header for defense in
+    # depth and add the query token only for /search compatibility.
+    if safe_path == "/search":
+        query["token"] = token
     url = f"{base_url}{safe_path}?{urlencode(query)}"
     request = Request(url, headers={"Authorization": f"token {token}"})
     try:

api/streaming.py

@@ -203,6 +203,7 @@ WebUI progress guidance:
 - Each update should say what you are about to check, what you just confirmed, or why the next tool call is needed.
 - Keep updates concise, factual, and in the user's language. One or two short sentences are enough.
 - Do not reveal hidden reasoning, chain-of-thought, private scratchpads, secrets, raw logs, or long tool output.
+- Do not include terse planning fragments or scratchpad shorthand in visible assistant text. Avoid fragments like "Need check logs", "Need inspect email", or "maybe invite"; either omit them or rewrite them as clear user-facing progress.
 - For direct answers or very short tasks, skip progress updates and answer normally.
 """.strip()
 

static/boot.js

@@ -1115,6 +1115,18 @@ function _isVirtualKeyboardLikelyOpen(){
   if(!vv||!window.innerHeight)return true;
   return window.innerHeight-vv.height>120;
 }
+// #3076: a touch-primary device (`pointer:coarse`) can still have a
+// physical keyboard attached (Android tablet + Bluetooth keyboard,
+// detachable Surface in tablet mode, iPad + Magic Keyboard). When that
+// happens we should NOT force the mobile newline-on-Enter override
+// because Shift+Enter / Ctrl+Enter come from real keys and the user
+// expects desktop semantics. `matchMedia('(any-pointer:fine)')` is true
+// whenever ANY available pointing device is fine-grained — which is the
+// strongest signal browsers expose for "there is a real keyboard /
+// trackpad in the picture too". Skip the mobile default in that case.
+function _hasFinePointerCoexisting(){
+  try{ return matchMedia('(any-pointer:fine)').matches; }catch(_){ return false; }
+}
 $('msg').addEventListener('keydown',e=>{
   // Autocomplete navigation when dropdown is open
   const dd=$('cmdDropdown');
@@ -1139,7 +1151,10 @@ $('msg').addEventListener('keydown',e=>{
   // Users can override in Settings by explicitly choosing 'enter' mode.
   if(e.key==='Enter'){
     if(_isImeEnter(e)){return;}
-    const _mobileDefault=matchMedia('(pointer:coarse)').matches&&window._sendKey==='enter'&&_isVirtualKeyboardLikelyOpen();
+    const _mobileDefault=matchMedia('(pointer:coarse)').matches
+      &&!_hasFinePointerCoexisting()
+      &&window._sendKey==='enter'
+      &&_isVirtualKeyboardLikelyOpen();
     if(window._sendKey==='ctrl+enter'||_mobileDefault){
       if(e.ctrlKey||e.metaKey){e.preventDefault();send();}
     } else {

static/sessions.js

@@ -2309,7 +2309,13 @@ let _gatewaySSEWarningShown = false;
 const _gatewayFallbackPollMs = 30000;
 const _streamingPollMs = 5000;
 const _sessionTimeRefreshMs = 60000;
-const _activeSessionExternalRefreshMs = 5000;
+// #3107: the active-session "is it externally updated?" poll used to fire
+// every 5 s. On long sessions this caused visible scroll jitter and a
+// noticeable network/CPU floor because the SSE session-events stream
+// already pushes invalidations in real time; this poll exists only as a
+// fallback for the case where SSE is broken/unavailable. Bump to 30 s
+// to keep the safety net without turning it into a primary refresh path.
+const _activeSessionExternalRefreshMs = 30000;
 let _streamingPollTimer = null;
 let _sessionTimeRefreshTimer = null;
 let _activeSessionExternalRefreshTimer = null;

static/ui.js

@@ -1068,6 +1068,13 @@ function _findModelInDropdown(modelId, sel, preferredProviderId){
   if(opts.includes(modelId)) return modelId;
   const exact=opts.find(o=>norm(o)===target);
   if(exact) return exact;
+  // If the request is provider-qualified (either explicit @provider:model or
+  // a slash-qualified vendor/model id), do NOT fuzzy-match a sibling model
+  // once exact/provider-aware lookup failed. Returning null lets the caller
+  // preserve the raw typed value instead of snapping to the closest catalog
+  // entry. This keeps uncatalogued models routable instead of silently turning
+  // them into a nearby curated sibling.
+  if(rawModel.startsWith('@')||rawModel.includes('/')) return null;
   // 3. Prefix/substring: require the candidate to start with the FULL normalized target
   // (not a truncated base). This avoids false matches like gpt.5.5 → gpt.5.4.mini (#1188).
   // Only fall back to the shorter base form if target itself is very short (a bare root

tests/test_issue1188_fuzzy_match.py

@@ -45,10 +45,22 @@ function extractFunc(name) {
   }
   return ui.slice(start, i);
 }
+function _getOptionProviderId(opt) {
+  if (!opt) return '';
+  if (opt.dataset && opt.dataset.provider) return opt.dataset.provider;
+  const group = opt.parentElement;
+  if (group && group.tagName === 'OPTGROUP' && group.dataset && group.dataset.provider) {
+    return group.dataset.provider;
+  }
+  const value = String(opt.value || '');
+  if (value.startsWith('@') && value.includes(':')) return value.slice(1, value.lastIndexOf(':'));
+  return '';
+}
+
 eval(extractFunc('_findModelInDropdown'));
 const args = JSON.parse(process.argv[3]);
 const sel = { options: args.options.map(v => ({value: v})) };
-const got = _findModelInDropdown(args.modelId, sel);
+const got = _findModelInDropdown(args.modelId, sel, args.preferredProvider || undefined);
 process.stdout.write(JSON.stringify(got));
 """
 
@@ -60,11 +72,11 @@ def driver_path(tmp_path_factory):
     return str(p)
 
 
-def _find(driver_path, model_id: str, options: list[str]):
+def _find(driver_path, model_id: str, options: list[str], preferred: str | None = None):
     import json
     result = subprocess.run(
         [NODE, driver_path, str(UI_JS_PATH),
-         json.dumps({"modelId": model_id, "options": options})],
+         json.dumps({"modelId": model_id, "options": options, "preferredProvider": preferred})],
         capture_output=True, text=True, timeout=10,
     )
     if result.returncode != 0:
@@ -113,6 +125,33 @@ class TestPreservedFuzzyMatches:
         )
         assert got == "@nous:openai/gpt-5.5"
 
+    def test_explicit_provider_hint_does_not_snap_to_sibling_model(self, driver_path):
+        """Provider-qualified requests must not fuzzy-match a sibling catalog entry.
+
+        This is the generalized regression behind #3113: any slash-qualified or
+        @provider-qualified model that is missing from the curated catalog should
+        preserve its exact value instead of snapping to a nearby curated sibling.
+        """
+        got = _find(
+            driver_path,
+            "vendor/special-model",
+            ["@openrouter:vendor/special-model-pro"],
+            preferred="openrouter",
+        )
+        assert got is None, (
+            "provider-qualified uncatalogued models must not fuzzy-match sibling models "
+            "(#3113)"
+        )
+
+    def test_provider_qualified_exact_match_still_resolves(self, driver_path):
+        got = _find(
+            driver_path,
+            "vendor/special-model",
+            ["@openrouter:vendor/special-model"],
+            preferred="openrouter",
+        )
+        assert got == "@openrouter:vendor/special-model"
+
     def test_bare_root_gpt_matches_versioned_option(self, driver_path):
         """Short root targets still fall back to the looser prefix match."""
         got = _find(driver_path, "gpt", ["@nous:openai/gpt-5.4-mini"])

tests/test_webui_notes_sources.py

@@ -138,7 +138,7 @@ def test_joplin_get_note_validates_id_and_truncates_body(monkeypatch):
     assert "Preview truncated" in note["body"]
 
 
-def test_joplin_api_get_uses_authorization_header(monkeypatch):
+def test_joplin_api_get_sends_header_and_query_token_for_clip_search_compat(monkeypatch):
     from api import routes
 
     captured = {}
@@ -162,10 +162,40 @@ def test_joplin_api_get_uses_authorization_header(monkeypatch):
     monkeypatch.setattr(routes, "_joplin_connection_from_config", lambda: ("http://127.0.0.1:41184", "secret-token"))
     monkeypatch.setattr("urllib.request.urlopen", fake_urlopen)
 
-    data = routes._joplin_api_get("/notes", {"query": "hello world"})
+    data = routes._joplin_api_get("/search", {"query": "hello world"})
 
     assert data == {"ok": True}
     assert captured["timeout"] == 8
+    assert "token=secret-token" in captured["url"]
+    assert "query=hello+world" in captured["url"]
+    assert captured["authorization"] == "token secret-token"
+
+
+def test_joplin_api_get_keeps_non_search_token_out_of_url(monkeypatch):
+    from api import routes
+
+    captured = {}
+
+    class FakeResponse:
+        def __enter__(self):
+            return self
+
+        def __exit__(self, *_exc):
+            return False
+
+        def read(self, _limit):
+            return b'{"ok": true}'
+
+    def fake_urlopen(request, timeout):
+        captured["url"] = request.full_url
+        captured["authorization"] = request.get_header("Authorization")
+        return FakeResponse()
+
+    monkeypatch.setattr(routes, "_joplin_connection_from_config", lambda: ("http://127.0.0.1:41184", "secret-token"))
+    monkeypatch.setattr("urllib.request.urlopen", fake_urlopen)
+
+    routes._joplin_api_get("/notes", {"query": "hello world"})
+
     assert "token=" not in captured["url"]
     assert "query=hello+world" in captured["url"]
     assert captured["authorization"] == "token secret-token"

tests/test_webui_surface_context.py

@@ -22,6 +22,9 @@ def test_webui_ephemeral_prompt_includes_browser_surface_context():
     assert "Do not copy or dump this browser transcript" in prompt
     assert "explicit captures" in prompt
     assert "durable user preferences" in prompt
+    assert "Do not include terse planning fragments" in prompt
+    assert "Need inspect email" in prompt
+    assert "clear user-facing progress" in prompt
 
 
 def test_webui_ephemeral_prompt_skips_empty_surface_fields():