cc-viewer 1.6.316 → 1.6.318

package/concepts/ar/Tool-ToolSearch.md

@@ -0,0 +1,48 @@
+# ToolSearch
+
+يجلب عند الطلب التعريفات الكاملة لمخططات «الأدوات المؤجَّلة» حتى تصبح قابلة للاستدعاء. عندما يتوفر عدد كبير من الأدوات، لا يُحمَّل بعضها مسبقاً — إذ تظهر بالاسم فقط داخل رسائل `<system-reminder>`. وإلى أن يُجلب مخططها، يكون الاسم وحده معروفاً ولا يوجد تعريف للمعاملات، لذا لا يمكن استدعاء الأداة. تأخذ `ToolSearch` استعلاماً، وتطابقه مع قائمة الأدوات المؤجَّلة، وتُعيد تعريفات JSONSchema الكاملة للأدوات المطابقة داخل كتلة `<functions>`. وبمجرد ظهور مخطط الأداة في النتيجة، تصبح قابلة للاستدعاء تماماً مثل أي أداة معرَّفة في بداية الموجِّه.
+
+## متى يُستخدم
+
+- تحتاج إلى أداة مؤجَّلة — يظهر اسمها في `<system-reminder>`، لكن لا يوجد لها تعريف معاملات في قائمة الأدوات ذات المستوى الأعلى.
+- تريد استخدام أدوات خادم MCP (مثل Slack أو Gmail أو computer-use) التي تُحمَّل عند الطلب.
+- لست متأكداً من الاسم الدقيق لأداة تؤدي قدرة معينة وتريد إظهار المرشحين بكلمة مفتاحية دفعة واحدة.
+
+إذا كان مخطط الأداة موجوداً بالفعل في السياق، فلا تبحث مجدداً — اكتفِ باستدعائها.
+
+## المعاملات
+
+- `query` (سلسلة، مطلوب): الاستعلام المستخدَم لتحديد موقع الأدوات المؤجَّلة. تُدعَم ثلاث صيغ:
+  - `select:Read,Edit,Grep` — جلب هذه الأدوات بالتحديد بالاسم.
+  - `notebook jupyter` — بحث بالكلمات المفتاحية، يُعيد حتى `max_results` من أفضل المطابقات.
+  - `+slack send` — اشتراط ظهور `slack` في اسم الأداة، ثم الترتيب وفق بقية المصطلحات.
+- `max_results` (عدد، اختياري): الحد الأقصى لعدد النتائج المُعادة. القيمة الافتراضية 5.
+
+## أمثلة
+
+### مثال 1: الجلب بالاسم الدقيق
+
+```
+ToolSearch(query="select:WebFetch,WebSearch", max_results=5)
+```
+
+### مثال 2: البحث بالكلمات المفتاحية
+
+```
+ToolSearch(query="notebook jupyter", max_results=5)
+```
+
+### مثال 3: تحميل مجموعة أدوات MCP بأكملها دفعة واحدة
+
+عند التحميل الجماعي لكل أدوات خادم MCP (مثل computer-use)، استخدم بحثاً واحداً بكلمة مفتاحية بدلاً من اختيار كل أداة على حدة — فاسم الخادم كسلسلة فرعية يطابق كل أداة تابعة لذلك الخادم:
+
+```
+ToolSearch(query="computer-use", max_results=30)
+```
+
+## ملاحظات
+
+- قبل استدعاء أداة مؤجَّلة يجب أن تجلب أولاً مخططها باستخدام `ToolSearch` — فاستدعاؤها مباشرةً يفشل لأن تعريف المعاملات مفقود.
+- عند التحميل الجماعي لمجموعة أدوات كاملة (مثل كل أدوات خادم MCP)، فضِّل بحثاً واحداً بكلمة مفتاحية على استدعاءات `select:` المتعددة لتقليل عدد جولات الطلب.
+- بمجرد جلب المخطط، تتصرف الأداة تماماً مثل أي أداة عادية؛ لا تُعِد البحث عن الأداة نفسها.
+- تُعاد النتائج بوصفها كتلة `<functions>`، كل أداة في سطر `<function>{...}</function>` واحد — بالترميز نفسه المستخدَم في قائمة الأدوات ذات المستوى الأعلى.

package/concepts/ar/UltraPlan.md

@@ -99,6 +99,7 @@ Your final plan must include the following elements:
 First run `git diff --quiet && git diff --cached --quiet` (or equivalent) to detect whether the working tree actually has non-trivial changes; if there are no real changes (or only whitespace/comment-only edits), skip the UltraReview step.
 Otherwise, if the project is managed with Git:
 Initiate a team (`TeamCreate`), dynamically allocating the number of teammates based on task complexity (5 is recommended);
+If the `TeamCreate` tool is unavailable, spawn teammates directly with the `Agent` tool (note: as teammates, not plain subagents);
 Task: Conduct a Code Review of the current git changes from multiple perspectives;
 Pre-requisites:
 - The git repository may be located in a subdirectory of the current directory; prefer `git rev-parse --show-toplevel` (fall back to recursive lookup) before proceeding;
@@ -115,6 +116,7 @@ Workflow:
 - Distill specific, actionable modification suggestions and annotate them with priority levels (P0/P1/P2/P3);
 - Upon completion, adopt P0 items, and selectively adopt P1 items when they are concrete and low-risk; defer P2/P3 to backlog;
 - After execution is complete, close the team (`TeamDelete`);
+- If the `TeamDelete` tool is unavailable, notify each teammate to exit individually;
 </system-reminder></textarea>
 
 ### Research Expert (researchExpert)

package/concepts/da/Tool-ToolSearch.md

@@ -0,0 +1,48 @@
+# ToolSearch
+
+Henter on demand de fuldstændige schema-definitioner for "udskudte værktøjer", så de bliver kaldbare. Når mange værktøjer er tilgængelige, indlæses nogle ikke på forhånd — de optræder kun ved navn inde i `<system-reminder>`-beskeder. Indtil dets schema er hentet, kendes kun navnet, og der er ingen parameterdefinition, så værktøjet kan ikke invokeres. `ToolSearch` modtager en forespørgsel, matcher den mod listen over udskudte værktøjer og returnerer de matchede værktøjers fuldstændige JSONSchema-definitioner inde i en `<functions>`-blok. Så snart et værktøjs schema optræder i resultatet, kan det kaldes præcis som ethvert værktøj defineret i toppen af prompten.
+
+## Hvornår skal den bruges
+
+- Du har brug for et udskudt værktøj — dets navn optræder i en `<system-reminder>`, men der er ingen parameterdefinition for det i værktøjslisten på øverste niveau.
+- Du vil bruge en MCP-servers værktøjer (f.eks. Slack, Gmail, computer-use), der indlæses on demand.
+- Du er ikke sikker på det eksakte værktøjsnavn for en kapacitet og vil få kandidater frem efter nøgleord i ét hug.
+
+Hvis et værktøjs schema allerede er i konteksten, så søg ikke igen — kald det blot.
+
+## Parametre
+
+- `query` (string, påkrævet): Forespørgslen, der bruges til at lokalisere udskudte værktøjer. Tre former understøttes:
+  - `select:Read,Edit,Grep` — henter disse eksakte værktøjer ved navn.
+  - `notebook jupyter` — nøgleordssøgning, der returnerer op til `max_results` bedste match.
+  - `+slack send` — kræver, at `slack` optræder i værktøjsnavnet, og rangerer derefter efter de resterende termer.
+- `max_results` (number, valgfri): Maksimalt antal resultater, der returneres. Standard er 5.
+
+## Eksempler
+
+### Eksempel 1: Hent ved eksakt navn
+
+```
+ToolSearch(query="select:WebFetch,WebSearch", max_results=5)
+```
+
+### Eksempel 2: Nøgleordssøgning
+
+```
+ToolSearch(query="notebook jupyter", max_results=5)
+```
+
+### Eksempel 3: Indlæs et helt MCP-værktøjssæt på én gang
+
+Når du masseindlæser alle værktøjer fra en MCP-server (f.eks. computer-use), så brug en enkelt nøgleordssøgning i stedet for at vælge hvert enkelt — servernavnet som delstreng matcher alle værktøjer under den server:
+
+```
+ToolSearch(query="computer-use", max_results=30)
+```
+
+## Noter
+
+- Før du invokerer et udskudt værktøj, skal du først hente dets schema med `ToolSearch` — at kalde det direkte fejler, fordi parameterdefinitionen mangler.
+- Når du masseindlæser et helt værktøjssæt (f.eks. alle en MCP-servers værktøjer), så foretræk én nøgleordssøgning frem for mange `select:`-kald for at skære ned på frem og tilbage.
+- Så snart et schema er hentet, opfører værktøjet sig præcis som ethvert normalt værktøj; søg ikke efter det samme værktøj igen.
+- Resultaterne kommer tilbage som en `<functions>`-blok, hvert værktøj en enkelt `<function>{...}</function>`-linje — den samme kodning som værktøjslisten på øverste niveau.

package/concepts/da/UltraPlan.md

@@ -99,6 +99,7 @@ Your final plan must include the following elements:
 First run `git diff --quiet && git diff --cached --quiet` (or equivalent) to detect whether the working tree actually has non-trivial changes; if there are no real changes (or only whitespace/comment-only edits), skip the UltraReview step.
 Otherwise, if the project is managed with Git:
 Initiate a team (`TeamCreate`), dynamically allocating the number of teammates based on task complexity (5 is recommended);
+If the `TeamCreate` tool is unavailable, spawn teammates directly with the `Agent` tool (note: as teammates, not plain subagents);
 Task: Conduct a Code Review of the current git changes from multiple perspectives;
 Pre-requisites:
 - The git repository may be located in a subdirectory of the current directory; prefer `git rev-parse --show-toplevel` (fall back to recursive lookup) before proceeding;
@@ -115,6 +116,7 @@ Workflow:
 - Distill specific, actionable modification suggestions and annotate them with priority levels (P0/P1/P2/P3);
 - Upon completion, adopt P0 items, and selectively adopt P1 items when they are concrete and low-risk; defer P2/P3 to backlog;
 - After execution is complete, close the team (`TeamDelete`);
+- If the `TeamDelete` tool is unavailable, notify each teammate to exit individually;
 </system-reminder></textarea>
 
 ### Research Expert (researchExpert)

package/concepts/de/Tool-ToolSearch.md

@@ -0,0 +1,48 @@
+# ToolSearch
+
+Ruft bei Bedarf die vollständigen Schema-Definitionen „aufgeschobener Tools" ab, damit diese aufrufbar werden. Wenn viele Tools verfügbar sind, werden einige nicht von vornherein geladen – sie erscheinen nur namentlich in `<system-reminder>`-Nachrichten. Bis sein Schema abgerufen ist, ist nur der Name bekannt und es gibt keine Parameterdefinition, sodass das Tool nicht aufgerufen werden kann. `ToolSearch` nimmt eine Anfrage entgegen, gleicht sie mit der Liste der aufgeschobenen Tools ab und liefert die vollständigen JSONSchema-Definitionen der übereinstimmenden Tools innerhalb eines `<functions>`-Blocks zurück. Sobald das Schema eines Tools im Ergebnis erscheint, ist es genauso aufrufbar wie jedes oben im Prompt definierte Tool.
+
+## Wann verwenden
+
+- Sie benötigen ein aufgeschobenes Tool – sein Name erscheint in einem `<system-reminder>`, aber in der Tool-Liste auf oberster Ebene gibt es keine Parameterdefinition dafür.
+- Sie möchten die Tools eines MCP-Servers verwenden (z. B. Slack, Gmail, computer-use), die bei Bedarf geladen werden.
+- Sie sind sich des exakten Tool-Namens für eine Fähigkeit nicht sicher und möchten Kandidaten per Stichwort in einem Zug zutage fördern.
+
+Wenn das Schema eines Tools bereits im Kontext vorliegt, nicht erneut suchen – einfach aufrufen.
+
+## Parameter
+
+- `query` (string, erforderlich): Die Anfrage zum Auffinden aufgeschobener Tools. Drei Formen werden unterstützt:
+  - `select:Read,Edit,Grep` – diese exakten Tools namentlich abrufen.
+  - `notebook jupyter` – Stichwortsuche, liefert bis zu `max_results` beste Treffer zurück.
+  - `+slack send` – verlangt, dass `slack` im Tool-Namen vorkommt, und ordnet dann nach den übrigen Begriffen.
+- `max_results` (number, optional): Maximale Anzahl zurückzugebender Ergebnisse. Standard ist 5.
+
+## Beispiele
+
+### Beispiel 1: Abruf per exaktem Namen
+
+```
+ToolSearch(query="select:WebFetch,WebSearch", max_results=5)
+```
+
+### Beispiel 2: Stichwortsuche
+
+```
+ToolSearch(query="notebook jupyter", max_results=5)
+```
+
+### Beispiel 3: Ein ganzes MCP-Toolkit auf einmal laden
+
+Beim Massenladen aller Tools eines MCP-Servers (z. B. computer-use) eine einzelne Stichwortsuche verwenden, statt jedes einzeln auszuwählen – der Servername als Teilzeichenkette passt auf jedes Tool unter diesem Server:
+
+```
+ToolSearch(query="computer-use", max_results=30)
+```
+
+## Hinweise
+
+- Bevor Sie ein aufgeschobenes Tool aufrufen, müssen Sie zuerst sein Schema mit `ToolSearch` abrufen – ein direkter Aufruf schlägt fehl, weil die Parameterdefinition fehlt.
+- Beim Massenladen eines ganzen Toolkits (z. B. aller Tools eines MCP-Servers) eine einzige Stichwortsuche gegenüber vielen `select:`-Aufrufen bevorzugen, um Hin-und-Her-Runden zu sparen.
+- Sobald ein Schema abgerufen ist, verhält sich das Tool genau wie jedes normale Tool; dasselbe Tool nicht erneut suchen.
+- Ergebnisse kommen als `<functions>`-Block zurück, jedes Tool als eine einzelne `<function>{...}</function>`-Zeile – dieselbe Kodierung wie die Tool-Liste auf oberster Ebene.

package/concepts/de/UltraPlan.md

@@ -99,6 +99,7 @@ Your final plan must include the following elements:
 First run `git diff --quiet && git diff --cached --quiet` (or equivalent) to detect whether the working tree actually has non-trivial changes; if there are no real changes (or only whitespace/comment-only edits), skip the UltraReview step.
 Otherwise, if the project is managed with Git:
 Initiate a team (`TeamCreate`), dynamically allocating the number of teammates based on task complexity (5 is recommended);
+If the `TeamCreate` tool is unavailable, spawn teammates directly with the `Agent` tool (note: as teammates, not plain subagents);
 Task: Conduct a Code Review of the current git changes from multiple perspectives;
 Pre-requisites:
 - The git repository may be located in a subdirectory of the current directory; prefer `git rev-parse --show-toplevel` (fall back to recursive lookup) before proceeding;
@@ -115,6 +116,7 @@ Workflow:
 - Distill specific, actionable modification suggestions and annotate them with priority levels (P0/P1/P2/P3);
 - Upon completion, adopt P0 items, and selectively adopt P1 items when they are concrete and low-risk; defer P2/P3 to backlog;
 - After execution is complete, close the team (`TeamDelete`);
+- If the `TeamDelete` tool is unavailable, notify each teammate to exit individually;
 </system-reminder></textarea>
 
 ### Research Expert (researchExpert)

package/concepts/en/Tool-ToolSearch.md

@@ -0,0 +1,48 @@
+# ToolSearch
+
+Fetches the full schema definitions for "deferred tools" on demand so they become callable. When many tools are available, some are not loaded up front — they appear only by name inside `<system-reminder>` messages. Until its schema is fetched, only the name is known and there is no parameter definition, so the tool cannot be invoked. `ToolSearch` takes a query, matches it against the deferred-tool list, and returns the matched tools' complete JSONSchema definitions inside a `<functions>` block. Once a tool's schema appears in the result, it is callable exactly like any tool defined at the top of the prompt.
+
+## When to Use
+
+- You need a deferred tool — its name appears in a `<system-reminder>`, but there is no parameter definition for it in the top-level tool list.
+- You want to use an MCP server's tools (e.g. Slack, Gmail, computer-use) that are loaded on demand.
+- You are not sure of the exact tool name for a capability and want to surface candidates by keyword in one shot.
+
+If a tool's schema is already in context, do not search again — just call it.
+
+## Parameters
+
+- `query` (string, required): The query used to locate deferred tools. Three forms are supported:
+  - `select:Read,Edit,Grep` — fetch these exact tools by name.
+  - `notebook jupyter` — keyword search, returning up to `max_results` best matches.
+  - `+slack send` — require `slack` to appear in the tool name, then rank by the remaining terms.
+- `max_results` (number, optional): Maximum number of results to return. Defaults to 5.
+
+## Examples
+
+### Example 1: Fetch by exact name
+
+```
+ToolSearch(query="select:WebFetch,WebSearch", max_results=5)
+```
+
+### Example 2: Keyword search
+
+```
+ToolSearch(query="notebook jupyter", max_results=5)
+```
+
+### Example 3: Load an entire MCP toolkit at once
+
+When bulk-loading every tool of an MCP server (e.g. computer-use), use a single keyword search instead of selecting each one — the server name as a substring matches every tool under that server:
+
+```
+ToolSearch(query="computer-use", max_results=30)
+```
+
+## Notes
+
+- Before invoking a deferred tool you must first fetch its schema with `ToolSearch` — calling it directly fails because the parameter definition is missing.
+- When bulk-loading a whole toolkit (e.g. all of an MCP server's tools), prefer one keyword search over many `select:` calls to cut round-trips.
+- Once a schema is fetched, the tool behaves exactly like any normal tool; do not re-search the same tool.
+- Results come back as a `<functions>` block, each tool a single `<function>{...}</function>` line — the same encoding as the top-level tool list.

package/concepts/en/UltraPlan.md

@@ -99,6 +99,7 @@ Your final plan must include the following elements:
 First run `git diff --quiet && git diff --cached --quiet` (or equivalent) to detect whether the working tree actually has non-trivial changes; if there are no real changes (or only whitespace/comment-only edits), skip the UltraReview step.
 Otherwise, if the project is managed with Git:
 Initiate a team (`TeamCreate`), dynamically allocating the number of teammates based on task complexity (5 is recommended);
+If the `TeamCreate` tool is unavailable, spawn teammates directly with the `Agent` tool (note: as teammates, not plain subagents);
 Task: Conduct a Code Review of the current git changes from multiple perspectives;
 Pre-requisites:
 - The git repository may be located in a subdirectory of the current directory; prefer `git rev-parse --show-toplevel` (fall back to recursive lookup) before proceeding;
@@ -115,6 +116,7 @@ Workflow:
 - Distill specific, actionable modification suggestions and annotate them with priority levels (P0/P1/P2/P3);
 - Upon completion, adopt P0 items, and selectively adopt P1 items when they are concrete and low-risk; defer P2/P3 to backlog;
 - After execution is complete, close the team (`TeamDelete`);
+- If the `TeamDelete` tool is unavailable, notify each teammate to exit individually;
 </system-reminder></textarea>
 
 ### Research Expert (researchExpert)

package/concepts/es/Tool-ToolSearch.md

@@ -0,0 +1,48 @@
+# ToolSearch
+
+Recupera bajo demanda las definiciones de esquema completas de las «herramientas diferidas» para que pasen a ser invocables. Cuando hay muchas herramientas disponibles, algunas no se cargan de entrada — aparecen únicamente por su nombre dentro de los mensajes `<system-reminder>`. Hasta que se recupera su esquema, solo se conoce el nombre y no existe ninguna definición de parámetros, por lo que la herramienta no puede invocarse. `ToolSearch` toma una consulta, la coteja con la lista de herramientas diferidas y devuelve las definiciones JSONSchema completas de las herramientas coincidentes dentro de un bloque `<functions>`. Una vez que el esquema de una herramienta aparece en el resultado, es invocable exactamente igual que cualquier herramienta definida al principio del prompt.
+
+## Cuándo usar
+
+- Necesitas una herramienta diferida — su nombre aparece en un `<system-reminder>`, pero no hay ninguna definición de parámetros para ella en la lista de herramientas de nivel superior.
+- Quieres usar las herramientas de un servidor MCP (p. ej. Slack, Gmail, computer-use) que se cargan bajo demanda.
+- No estás seguro del nombre exacto de la herramienta para una capacidad y quieres hacer aflorar candidatas por palabra clave de una sola vez.
+
+Si el esquema de una herramienta ya está en el contexto, no vuelvas a buscar — simplemente invócala.
+
+## Parámetros
+
+- `query` (string, obligatorio): La consulta usada para localizar herramientas diferidas. Se admiten tres formas:
+  - `select:Read,Edit,Grep` — recuperar estas herramientas exactas por su nombre.
+  - `notebook jupyter` — búsqueda por palabra clave, devolviendo hasta `max_results` mejores coincidencias.
+  - `+slack send` — exigir que `slack` aparezca en el nombre de la herramienta y luego clasificar por los términos restantes.
+- `max_results` (number, opcional): Número máximo de resultados a devolver. Por defecto es 5.
+
+## Ejemplos
+
+### Ejemplo 1: Recuperar por nombre exacto
+
+```
+ToolSearch(query="select:WebFetch,WebSearch", max_results=5)
+```
+
+### Ejemplo 2: Búsqueda por palabra clave
+
+```
+ToolSearch(query="notebook jupyter", max_results=5)
+```
+
+### Ejemplo 3: Cargar un kit de herramientas MCP completo de una vez
+
+Al cargar en bloque todas las herramientas de un servidor MCP (p. ej. computer-use), usa una sola búsqueda por palabra clave en lugar de seleccionar cada una — el nombre del servidor como subcadena coincide con todas las herramientas de ese servidor:
+
+```
+ToolSearch(query="computer-use", max_results=30)
+```
+
+## Notas
+
+- Antes de invocar una herramienta diferida debes recuperar primero su esquema con `ToolSearch` — invocarla directamente falla porque falta la definición de parámetros.
+- Al cargar en bloque un kit de herramientas completo (p. ej. todas las herramientas de un servidor MCP), prefiere una sola búsqueda por palabra clave en lugar de muchas llamadas `select:` para reducir los viajes de ida y vuelta.
+- Una vez recuperado el esquema, la herramienta se comporta exactamente como cualquier herramienta normal; no vuelvas a buscar la misma herramienta.
+- Los resultados regresan como un bloque `<functions>`, cada herramienta en una sola línea `<function>{...}</function>` — la misma codificación que la lista de herramientas de nivel superior.

package/concepts/es/UltraPlan.md

@@ -99,6 +99,7 @@ Your final plan must include the following elements:
 First run `git diff --quiet && git diff --cached --quiet` (or equivalent) to detect whether the working tree actually has non-trivial changes; if there are no real changes (or only whitespace/comment-only edits), skip the UltraReview step.
 Otherwise, if the project is managed with Git:
 Initiate a team (`TeamCreate`), dynamically allocating the number of teammates based on task complexity (5 is recommended);
+If the `TeamCreate` tool is unavailable, spawn teammates directly with the `Agent` tool (note: as teammates, not plain subagents);
 Task: Conduct a Code Review of the current git changes from multiple perspectives;
 Pre-requisites:
 - The git repository may be located in a subdirectory of the current directory; prefer `git rev-parse --show-toplevel` (fall back to recursive lookup) before proceeding;
@@ -115,6 +116,7 @@ Workflow:
 - Distill specific, actionable modification suggestions and annotate them with priority levels (P0/P1/P2/P3);
 - Upon completion, adopt P0 items, and selectively adopt P1 items when they are concrete and low-risk; defer P2/P3 to backlog;
 - After execution is complete, close the team (`TeamDelete`);
+- If the `TeamDelete` tool is unavailable, notify each teammate to exit individually;
 </system-reminder></textarea>
 
 ### Research Expert (researchExpert)

package/concepts/fr/Tool-ToolSearch.md

@@ -0,0 +1,48 @@
+# ToolSearch
+
+Récupère à la demande les définitions de schéma complètes des « outils différés » afin de les rendre appelables. Lorsque de nombreux outils sont disponibles, certains ne sont pas chargés d'emblée — ils n'apparaissent que par leur nom dans les messages `<system-reminder>`. Tant que son schéma n'est pas récupéré, seul le nom est connu et il n'existe aucune définition de paramètres, de sorte que l'outil ne peut pas être invoqué. `ToolSearch` prend une requête, la compare à la liste des outils différés et renvoie les définitions JSONSchema complètes des outils correspondants dans un bloc `<functions>`. Une fois que le schéma d'un outil apparaît dans le résultat, il est appelable exactement comme tout outil défini en haut du prompt.
+
+## Quand l'utiliser
+
+- Vous avez besoin d'un outil différé — son nom apparaît dans un `<system-reminder>`, mais il n'existe aucune définition de paramètres pour lui dans la liste d'outils de premier niveau.
+- Vous voulez utiliser les outils d'un serveur MCP (p. ex. Slack, Gmail, computer-use) qui sont chargés à la demande.
+- Vous n'êtes pas sûr du nom exact de l'outil correspondant à une capacité et vous voulez faire apparaître des candidats par mot-clé en une seule fois.
+
+Si le schéma d'un outil est déjà dans le contexte, ne relancez pas de recherche — appelez-le simplement.
+
+## Paramètres
+
+- `query` (string, requis) : la requête utilisée pour localiser les outils différés. Trois formes sont prises en charge :
+  - `select:Read,Edit,Grep` — récupérer ces outils précis par leur nom.
+  - `notebook jupyter` — recherche par mot-clé, renvoyant jusqu'à `max_results` meilleures correspondances.
+  - `+slack send` — exiger que `slack` apparaisse dans le nom de l'outil, puis classer selon les termes restants.
+- `max_results` (number, optionnel) : nombre maximal de résultats à renvoyer. Par défaut 5.
+
+## Exemples
+
+### Exemple 1 : récupération par nom exact
+
+```
+ToolSearch(query="select:WebFetch,WebSearch", max_results=5)
+```
+
+### Exemple 2 : recherche par mot-clé
+
+```
+ToolSearch(query="notebook jupyter", max_results=5)
+```
+
+### Exemple 3 : charger un kit d'outils MCP entier en une fois
+
+Lors du chargement en masse de tous les outils d'un serveur MCP (p. ex. computer-use), utilisez une seule recherche par mot-clé plutôt que de sélectionner chacun d'eux — le nom du serveur en tant que sous-chaîne correspond à chaque outil de ce serveur :
+
+```
+ToolSearch(query="computer-use", max_results=30)
+```
+
+## Notes
+
+- Avant d'invoquer un outil différé, vous devez d'abord récupérer son schéma avec `ToolSearch` — l'appeler directement échoue car la définition de paramètres est manquante.
+- Lors du chargement en masse d'un kit d'outils complet (p. ex. tous les outils d'un serveur MCP), préférez une seule recherche par mot-clé à de nombreux appels `select:` pour réduire les allers-retours.
+- Une fois le schéma récupéré, l'outil se comporte exactement comme n'importe quel outil normal ; ne relancez pas de recherche sur le même outil.
+- Les résultats reviennent sous forme de bloc `<functions>`, chaque outil étant une seule ligne `<function>{...}</function>` — le même encodage que la liste d'outils de premier niveau.

package/concepts/fr/UltraPlan.md

@@ -99,6 +99,7 @@ Your final plan must include the following elements:
 First run `git diff --quiet && git diff --cached --quiet` (or equivalent) to detect whether the working tree actually has non-trivial changes; if there are no real changes (or only whitespace/comment-only edits), skip the UltraReview step.
 Otherwise, if the project is managed with Git:
 Initiate a team (`TeamCreate`), dynamically allocating the number of teammates based on task complexity (5 is recommended);
+If the `TeamCreate` tool is unavailable, spawn teammates directly with the `Agent` tool (note: as teammates, not plain subagents);
 Task: Conduct a Code Review of the current git changes from multiple perspectives;
 Pre-requisites:
 - The git repository may be located in a subdirectory of the current directory; prefer `git rev-parse --show-toplevel` (fall back to recursive lookup) before proceeding;
@@ -115,6 +116,7 @@ Workflow:
 - Distill specific, actionable modification suggestions and annotate them with priority levels (P0/P1/P2/P3);
 - Upon completion, adopt P0 items, and selectively adopt P1 items when they are concrete and low-risk; defer P2/P3 to backlog;
 - After execution is complete, close the team (`TeamDelete`);
+- If the `TeamDelete` tool is unavailable, notify each teammate to exit individually;
 </system-reminder></textarea>
 
 ### Research Expert (researchExpert)

package/concepts/it/Tool-ToolSearch.md

@@ -0,0 +1,48 @@
+# ToolSearch
+
+Recupera su richiesta le definizioni di schema complete degli «strumenti differiti» affinché diventino richiamabili. Quando sono disponibili molti strumenti, alcuni non vengono caricati fin dall'inizio — compaiono solo per nome all'interno dei messaggi `<system-reminder>`. Finché il suo schema non viene recuperato, è noto solo il nome e non esiste alcuna definizione di parametri, quindi lo strumento non può essere invocato. `ToolSearch` prende una query, la confronta con l'elenco degli strumenti differiti e restituisce le definizioni JSONSchema complete degli strumenti corrispondenti all'interno di un blocco `<functions>`. Una volta che lo schema di uno strumento compare nel risultato, è richiamabile esattamente come qualsiasi strumento definito in cima al prompt.
+
+## Quando usare
+
+- Hai bisogno di uno strumento differito — il suo nome compare in un `<system-reminder>`, ma non c'è alcuna definizione di parametri per esso nell'elenco degli strumenti di primo livello.
+- Vuoi usare gli strumenti di un server MCP (es. Slack, Gmail, computer-use) che vengono caricati su richiesta.
+- Non sei sicuro del nome esatto dello strumento per una capacità e vuoi far emergere candidati per parola chiave in un colpo solo.
+
+Se lo schema di uno strumento è già nel contesto, non cercare di nuovo — limitati a richiamarlo.
+
+## Parametri
+
+- `query` (string, obbligatorio): La query usata per localizzare gli strumenti differiti. Sono supportate tre forme:
+  - `select:Read,Edit,Grep` — recuperare questi strumenti esatti per nome.
+  - `notebook jupyter` — ricerca per parola chiave, che restituisce fino a `max_results` migliori corrispondenze.
+  - `+slack send` — richiede che `slack` compaia nel nome dello strumento, poi ordina in base ai termini rimanenti.
+- `max_results` (number, opzionale): Numero massimo di risultati da restituire. Default: 5.
+
+## Esempi
+
+### Esempio 1: Recupero per nome esatto
+
+```
+ToolSearch(query="select:WebFetch,WebSearch", max_results=5)
+```
+
+### Esempio 2: Ricerca per parola chiave
+
+```
+ToolSearch(query="notebook jupyter", max_results=5)
+```
+
+### Esempio 3: Caricare un intero toolkit MCP in una volta
+
+Quando si caricano in blocco tutti gli strumenti di un server MCP (es. computer-use), usa una singola ricerca per parola chiave invece di selezionarli uno per uno — il nome del server come sottostringa corrisponde a ogni strumento sotto quel server:
+
+```
+ToolSearch(query="computer-use", max_results=30)
+```
+
+## Note
+
+- Prima di invocare uno strumento differito devi recuperarne lo schema con `ToolSearch` — richiamarlo direttamente fallisce perché manca la definizione dei parametri.
+- Quando si carica in blocco un intero toolkit (es. tutti gli strumenti di un server MCP), preferisci una singola ricerca per parola chiave a molte chiamate `select:` per ridurre i viaggi di andata e ritorno.
+- Una volta recuperato lo schema, lo strumento si comporta esattamente come qualsiasi strumento normale; non cercare di nuovo lo stesso strumento.
+- I risultati tornano come blocco `<functions>`, ogni strumento su una singola riga `<function>{...}</function>` — la stessa codifica dell'elenco di strumenti di primo livello.

package/concepts/it/UltraPlan.md

@@ -99,6 +99,7 @@ Your final plan must include the following elements:
 First run `git diff --quiet && git diff --cached --quiet` (or equivalent) to detect whether the working tree actually has non-trivial changes; if there are no real changes (or only whitespace/comment-only edits), skip the UltraReview step.
 Otherwise, if the project is managed with Git:
 Initiate a team (`TeamCreate`), dynamically allocating the number of teammates based on task complexity (5 is recommended);
+If the `TeamCreate` tool is unavailable, spawn teammates directly with the `Agent` tool (note: as teammates, not plain subagents);
 Task: Conduct a Code Review of the current git changes from multiple perspectives;
 Pre-requisites:
 - The git repository may be located in a subdirectory of the current directory; prefer `git rev-parse --show-toplevel` (fall back to recursive lookup) before proceeding;
@@ -115,6 +116,7 @@ Workflow:
 - Distill specific, actionable modification suggestions and annotate them with priority levels (P0/P1/P2/P3);
 - Upon completion, adopt P0 items, and selectively adopt P1 items when they are concrete and low-risk; defer P2/P3 to backlog;
 - After execution is complete, close the team (`TeamDelete`);
+- If the `TeamDelete` tool is unavailable, notify each teammate to exit individually;
 </system-reminder></textarea>
 
 ### Research Expert (researchExpert)

package/concepts/ja/Tool-ToolSearch.md

@@ -0,0 +1,48 @@
+# ToolSearch
+
+「遅延ツール」の完全なスキーマ定義をオンデマンドで取得し、呼び出せるようにします。多くのツールが利用可能な場合、一部は事前にロードされず、`<system-reminder>` メッセージの中に名前だけが現れます。スキーマが取得されるまでは名前しか分からず、パラメータ定義が存在しないため、そのツールは呼び出せません。`ToolSearch` はクエリを受け取り、遅延ツールのリストと照合し、一致したツールの完全な JSONSchema 定義を `<functions>` ブロックの中で返します。ツールのスキーマが結果に現れると、プロンプトの先頭で定義されたツールと全く同じように呼び出せるようになります。
+
+## 使用タイミング
+
+- 遅延ツールが必要な場合 — その名前は `<system-reminder>` に現れますが、トップレベルのツールリストにはパラメータ定義がありません。
+- オンデマンドでロードされる MCP サーバーのツール（例: Slack、Gmail、computer-use）を使用したい場合。
+- ある機能に対応する正確なツール名が分からず、キーワードで一度に候補を浮かび上がらせたい場合。
+
+ツールのスキーマがすでにコンテキストにある場合は、再度検索しないでください — そのまま呼び出してください。
+
+## パラメータ
+
+- `query` (string, required): 遅延ツールを特定するために使用するクエリ。3 つの形式がサポートされています:
+  - `select:Read,Edit,Grep` — これらの正確な名前でツールを取得します。
+  - `notebook jupyter` — キーワード検索で、最大 `max_results` 件の最適な一致を返します。
+  - `+slack send` — ツール名に `slack` が含まれることを必須とし、残りの語で順位付けします。
+- `max_results` (number, optional): 返す結果の最大数。デフォルトは 5。
+
+## 例
+
+### 例 1: 正確な名前で取得
+
+```
+ToolSearch(query="select:WebFetch,WebSearch", max_results=5)
+```
+
+### 例 2: キーワード検索
+
+```
+ToolSearch(query="notebook jupyter", max_results=5)
+```
+
+### 例 3: MCP ツールキット全体を一度にロード
+
+MCP サーバーのすべてのツール（例: computer-use）を一括ロードする場合、一つずつ選択するのではなく単一のキーワード検索を使用してください — サーバー名は部分文字列として、そのサーバー配下のすべてのツールに一致します:
+
+```
+ToolSearch(query="computer-use", max_results=30)
+```
+
+## 注意事項
+
+- 遅延ツールを呼び出す前に、まず `ToolSearch` でそのスキーマを取得する必要があります — パラメータ定義が欠けているため、直接呼び出すと失敗します。
+- ツールキット全体（例: MCP サーバーのすべてのツール）を一括ロードする場合は、ラウンドトリップを減らすために、多数の `select:` 呼び出しよりも一つのキーワード検索を優先してください。
+- スキーマを取得すれば、そのツールは通常のツールと全く同じように動作します。同じツールを再検索しないでください。
+- 結果は `<functions>` ブロックとして返り、各ツールは単一の `<function>{...}</function>` 行になります — トップレベルのツールリストと同じエンコーディングです。

package/concepts/ja/UltraPlan.md

@@ -99,6 +99,7 @@ Your final plan must include the following elements:
 First run `git diff --quiet && git diff --cached --quiet` (or equivalent) to detect whether the working tree actually has non-trivial changes; if there are no real changes (or only whitespace/comment-only edits), skip the UltraReview step.
 Otherwise, if the project is managed with Git:
 Initiate a team (`TeamCreate`), dynamically allocating the number of teammates based on task complexity (5 is recommended);
+If the `TeamCreate` tool is unavailable, spawn teammates directly with the `Agent` tool (note: as teammates, not plain subagents);
 Task: Conduct a Code Review of the current git changes from multiple perspectives;
 Pre-requisites:
 - The git repository may be located in a subdirectory of the current directory; prefer `git rev-parse --show-toplevel` (fall back to recursive lookup) before proceeding;
@@ -115,6 +116,7 @@ Workflow:
 - Distill specific, actionable modification suggestions and annotate them with priority levels (P0/P1/P2/P3);
 - Upon completion, adopt P0 items, and selectively adopt P1 items when they are concrete and low-risk; defer P2/P3 to backlog;
 - After execution is complete, close the team (`TeamDelete`);
+- If the `TeamDelete` tool is unavailable, notify each teammate to exit individually;
 </system-reminder></textarea>
 
 ### Research Expert (researchExpert)

package/concepts/ko/Tool-ToolSearch.md

@@ -0,0 +1,48 @@
+# ToolSearch
+
+"지연 로드 도구"의 전체 스키마 정의를 필요할 때 가져와 호출 가능하게 만듭니다. 많은 도구가 사용 가능한 경우 일부는 미리 로드되지 않고 `<system-reminder>` 메시지 안에 이름만 나타납니다. 스키마를 가져오기 전까지는 이름만 알 수 있고 매개변수 정의가 없으므로 해당 도구를 호출할 수 없습니다. `ToolSearch`는 쿼리를 받아 지연 로드 도구 목록과 대조하고, 일치하는 도구의 전체 JSONSchema 정의를 `<functions>` 블록 안에 반환합니다. 도구의 스키마가 결과에 나타나면 프롬프트 상단에 정의된 도구와 똑같이 호출 가능해집니다.
+
+## 사용 시점
+
+- 지연 로드 도구가 필요한 경우 — 그 이름이 `<system-reminder>`에 나타나지만 최상위 도구 목록에는 매개변수 정의가 없습니다.
+- 필요할 때 로드되는 MCP 서버의 도구(예: Slack, Gmail, computer-use)를 사용하려는 경우.
+- 어떤 기능에 대한 정확한 도구 이름을 확신하지 못해 키워드로 후보를 한 번에 드러내고 싶은 경우.
+
+도구의 스키마가 이미 컨텍스트에 있다면 다시 검색하지 마십시오 — 그냥 호출하십시오.
+
+## 매개변수
+
+- `query` (string, 필수): 지연 로드 도구를 찾는 데 사용하는 쿼리. 세 가지 형식이 지원됩니다:
+  - `select:Read,Edit,Grep` — 이 정확한 이름으로 도구를 가져옵니다.
+  - `notebook jupyter` — 키워드 검색으로, 최대 `max_results`개의 가장 적합한 일치 항목을 반환합니다.
+  - `+slack send` — 도구 이름에 `slack`이 나타나도록 요구한 다음 나머지 용어로 순위를 매깁니다.
+- `max_results` (number, 선택): 반환할 결과의 최대 개수. 기본값은 5입니다.
+
+## 예시
+
+### 예시 1: 정확한 이름으로 가져오기
+
+```
+ToolSearch(query="select:WebFetch,WebSearch", max_results=5)
+```
+
+### 예시 2: 키워드 검색
+
+```
+ToolSearch(query="notebook jupyter", max_results=5)
+```
+
+### 예시 3: MCP 툴킷 전체를 한 번에 로드
+
+MCP 서버의 모든 도구(예: computer-use)를 일괄 로드할 때는 각각을 선택하는 대신 단일 키워드 검색을 사용하십시오 — 서버 이름은 부분 문자열로서 해당 서버 아래의 모든 도구에 일치합니다:
+
+```
+ToolSearch(query="computer-use", max_results=30)
+```
+
+## 참고사항
+
+- 지연 로드 도구를 호출하기 전에 먼저 `ToolSearch`로 그 스키마를 가져와야 합니다 — 매개변수 정의가 없으므로 직접 호출하면 실패합니다.
+- 툴킷 전체(예: MCP 서버의 모든 도구)를 일괄 로드할 때는 왕복 횟수를 줄이기 위해 여러 번의 `select:` 호출보다 한 번의 키워드 검색을 선호하십시오.
+- 스키마를 가져오면 도구는 일반 도구와 똑같이 동작합니다. 같은 도구를 다시 검색하지 마십시오.
+- 결과는 `<functions>` 블록으로 돌아오며, 각 도구는 하나의 `<function>{...}</function>` 줄입니다 — 최상위 도구 목록과 동일한 인코딩입니다.

package/concepts/ko/UltraPlan.md

@@ -99,6 +99,7 @@ Your final plan must include the following elements:
 First run `git diff --quiet && git diff --cached --quiet` (or equivalent) to detect whether the working tree actually has non-trivial changes; if there are no real changes (or only whitespace/comment-only edits), skip the UltraReview step.
 Otherwise, if the project is managed with Git:
 Initiate a team (`TeamCreate`), dynamically allocating the number of teammates based on task complexity (5 is recommended);
+If the `TeamCreate` tool is unavailable, spawn teammates directly with the `Agent` tool (note: as teammates, not plain subagents);
 Task: Conduct a Code Review of the current git changes from multiple perspectives;
 Pre-requisites:
 - The git repository may be located in a subdirectory of the current directory; prefer `git rev-parse --show-toplevel` (fall back to recursive lookup) before proceeding;
@@ -115,6 +116,7 @@ Workflow:
 - Distill specific, actionable modification suggestions and annotate them with priority levels (P0/P1/P2/P3);
 - Upon completion, adopt P0 items, and selectively adopt P1 items when they are concrete and low-risk; defer P2/P3 to backlog;
 - After execution is complete, close the team (`TeamDelete`);
+- If the `TeamDelete` tool is unavailable, notify each teammate to exit individually;
 </system-reminder></textarea>
 
 ### Research Expert (researchExpert)