dsh-mcp-panel 0.6.14 → 0.6.15

package/CHANGELOG.md

@@ -2,6 +2,20 @@
 
 All notable changes to this project are documented in this file.
 
+## [0.6.15] - 2026-09-15
+
+### Fixed
+
+- **Issue #27 — `disable` / `enable` / `edit` emitted `- set:` fragments the loader dialect does not implement (silent no-op).** Both renderers (`renderPatchFragment` and the `/mcp <server> disable|enable` suggestion) now emit the loader's real dialect: `- id: <entryId>` + `name:` + `disabled:` (or a full `config:` block for edits). `appendPatchFragment` reads the file back to confirm the appended block survived, and `writePatch` re-verifies every write against the loader's re-applied state before reporting success — a skipped patch (wrong dialect, a row living in a layer a profile patch cannot reach such as `$DSH_HOME/cordis.patch.yml`, or a name mismatch) now fails honestly instead of reading "written". New config: `writeVerifyEnabled` (default true) and `writeVerifyTimeoutMs` (default 3000).
+
+### Added
+
+- **Read-only Resources browser.** The shipped `@deepseek-ai/dsh-mcp-resources` service now bridges MCP resources upstream; the console feature-detects the service + the three registered shared tools and offers list / templates / URI read per server card through the OFFICIAL tool pipeline (results stay panel-only). `callTool` now accepts the three shared resource tools and injects the requested `server` into their arguments (conflicts fail closed). Capability detection no longer probes the never-shipped `mcpCatalog` face; Prompts stays `available: false` until upstream bridges prompt templates.
+
+### Changed
+
+- Compatibility baseline raised to `dsh-v0.1.6-alpha.1`: all `@deepseek-ai/dsh-*` dev/test pins and the runtime `dsh-subprocess` pin moved to `0.1.6-alpha.1`; `dshWorkshop.compatibility.dshVersions` gains `0.1.6-alpha.1`; the five-language READMEs now document the id-override CRUD dialect, the Resources browser, the write-verification hardening, and the corrected capabilities wording (Prompts + resource subscriptions pending).
+
 ## [0.6.14] - 2026-09-12
 
 ### Changed

package/README-es.md

@@ -26,7 +26,7 @@
 
 | Superficie | Estado |
 |---|---|
-| Harness | DeepSeek Harness `dsh-v0.1.5-rc.2` (tag de GitHub, verificado el 2026-09-11) (adaptado el 2026-09-09): el sobre de sesión conserva su campo ignorable solo para compatibilidad de lectura de logs almacenados - Session.append aún no puede estamparlo, por lo que el comportamiento de la puerta no cambia. Verificado el 2026-09-11 contra el checkout master dsh-v0.1.5-rc.2 (cadena completa de puertas + smoke de instalación de perfil). |
+| Harness | DeepSeek Harness `dsh-v0.1.6-alpha.1` (tag de GitHub, verificado el 2026-09-16): cadena completa de puertas (typecheck / typecheck:ci / test / build / verify / package) más smoke de instalación de perfil contra el checkout 0.1.6-alpha.1. El servicio `@deepseek-ai/dsh-mcp-resources` ahora puentea Resources, que la consola detecta y navega en solo lectura. Línea base anterior: `dsh-v0.1.5-rc.2` (verificado el 2026-09-11). |
 | Node | `^22.19.0 \|\| >=24.0.0` |
 | Plataformas | Web GUI (doble cara: host + navegador) |
 | Modelo | Cualquiera (el panel es de solo lectura; solo la salida de `/mcp` es legible por el modelo) |
@@ -40,7 +40,8 @@
 - **`/mcp <servidor> health`** — sugerencias de autorreparación derivadas (ENOENT → dependencia faltante, ECONNREFUSED, timeouts, 401/403/404, DNS, rate limit, reconexión agotada…); código de salida / stderr etiquetados honestamente como *pendiente de soporte upstream*.
 - **`/mcp <servidor> call <tool> [json]`** — llamada de prueba por el **pipeline oficial de herramientas** (`ctx.tools.execute()`); política de permisos pre-ejecución, aprobación, guards y post-ejecución, todo en vigor.
 - **Ajustes → Plugins → MCP** — tarjetas de estado con insignias, diagnósticos y sondas, más el CRUD de servidores y el banco de pruebas de herramientas.
-- **CRUD de servidores** — formularios de alta/edición/borrado → fragmentos `insert`/`set`/`set disabled` → copia al portapapeles o escritura con aprobación y copias de seguridad automáticas.
+- **CRUD de servidores** — formularios de alta/edición/borrado → `insert` para alta y anulaciones dirigidas por id (`- id:` + `name:` + `disabled:`/`config:`) para edición/borrado → copia al portapapeles o escritura con aprobación, copias de seguridad automáticas y re-verificación contra el loader.
+- **Navegación de Resources** — listado de recursos, listado de plantillas y lecturas de URI de solo lectura mediante las herramientas oficiales `list_mcp_resources` / `list_mcp_resource_templates` / `read_mcp_resource` (puenteadas por el servicio `@deepseek-ai/dsh-mcp-resources`); los resultados se muestran solo en la pestaña, nunca en el contexto del modelo.
 - **Banco de pruebas de herramientas** — servidor → herramienta `mcp__*` → argumentos JSON → resultado JSON canónico + contenido renderizado; limitado por `trialMaxResultChars`; solo panel, nunca contexto del modelo.
 
 ## Architecture: official client = bridge, this plugin = console
@@ -78,7 +79,7 @@ La consola **lee** el cliente por su seam de observabilidad `mcp/status` (evento
 |---|---|---|
 | Añadir servidor | Editar YAML, cuidar indentación/comillas | Formulario → fragmento de parche → **copiar** o **escribir** (aprobación + copia de seguridad) |
 | Editar servidor | Editar YAML, reiniciar/recarga en caliente | Formulario precargado de la fila en vivo; los secretos sin cambios conservan su valor en el host |
-| Eliminar servidor | Borrar la fila | Operación `set disabled: true` (el vocabulario de parches no tiene remove) — re-habilitable |
+| Eliminar servidor | Borrar la fila | Anulación `- id:` + `disabled: true` (el vocabulario de parches no tiene remove) — re-habilitable |
 | Ver estado | Leer logs | Insignias + reconexiones + último error, en vivo desde `mcp/status` |
 | Probar una herramienta | Pedírselo al modelo | Banco de pruebas → pipeline oficial `ctx.tools.execute()` (permisos y aprobación en vigor) |
 | Diagnosticar fallos | grep de logs | `/mcp <servidor> health` con sugerencias derivadas |
@@ -132,6 +133,8 @@ Todas las opciones son campos Schemastery `Config` (modificables desde cordis.ym
 | `trialTimeoutMs` | `120000` | Plazo del panel por llamada de prueba en ms |
 | `trialMaxResultChars` | `60000` | Tope del payload de resultado de prueba en caracteres |
 | `writeEnabled` | `true` | Interruptor de seguridad: `false` rechaza toda escritura (copiar sigue funcionando) |
+| `writeVerifyEnabled` | `true` | Re-verifica cada escritura contra el estado re-aplicado del loader antes de reportar éxito |
+| `writeVerifyTimeoutMs` | `3000` | Presupuesto de sondeo para la verificación de escritura en ms |
 | `backupCount` | `5` | Copias de `cordis.patch.yml` retenidas por escritura |
 | `catalogEntries` | `[]` | Superposición de usuario para el directorio recomendado: anexa entradas; una entrada con el mismo `id` reemplaza la integrada |
 
@@ -149,7 +152,9 @@ Todas las opciones son campos Schemastery `Config` (modificables desde cordis.ym
 
 ## Resources & Prompts
 
-El cliente oficial documenta que *"Tools are the only bridged MCP capability"* — Resources y Prompts están diferidos. La consola detecta un seam de catálogo propuesto y mostrará listas de solo lectura el día que se envíe; hasta entonces el tablero de capacidades marca ambos **pendientes de soporte upstream**.
+Resources YA están puenteados upstream: el bundle base monta `@deepseek-ai/dsh-mcp-resources`, el cliente oficial registra el provider de recursos de cada conexión en `ctx.mcpResources`, y ese paquete posee las tres herramientas compartidas (`list_mcp_resources`, `list_mcp_resource_templates`, `read_mcp_resource`). La consola detecta el servicio y las herramientas registradas, y ofrece un navegador de Resources de solo lectura (lista / plantillas / lectura de URI) en cada tarjeta de servidor — cada llamada pasa por el pipeline OFICIAL de herramientas y los resultados nunca entran en el contexto del modelo.
+
+Las **plantillas de prompts** y las **suscripciones a recursos** siguen diferidas upstream; el tablero de capacidades marca Prompts **pendiente de soporte upstream**.
 
 ## Permissions & data
 
@@ -160,14 +165,15 @@ El cliente oficial documenta que *"Tools are the only bridged MCP capability"*
 
 - **El puente sigue siendo el puente.** Sin cambios de transporte, OAuth o protocolo; una fila mcp-client por servidor, exactamente como a mano.
 - **Sin estado falso.** Los campos de conexión sin observaciones upstream leen `unknown` / `—` con `statusSource: 'derived'`; códigos de salida y stderr nunca se inventan.
-- **Escrituras solo-anexar, con aprobación y respaldo.** La consola nunca reescribe `cordis.patch.yml`; anexa operaciones generadas y conserva las `backupCount` copias más recientes.
+- **Escrituras solo-anexar, con aprobación y respaldo.** La consola nunca reescribe `cordis.patch.yml`; anexa operaciones generadas, conserva las `backupCount` copias más recientes y re-verifica cada escritura contra el estado re-aplicado del loader antes de reportar éxito (un parche omitido nunca se muestra como éxito).
 - **Sin inyección de prompts.** El panel no registra secciones de prompt; su único texto visible al modelo son las dos descripciones de herramienta/comando.
 
 ## Known limitations
 
-- **Resources y Prompts** están pendientes de soporte upstream — el cliente oficial solo puentea herramientas.
+- **Plantillas de prompts y suscripciones a recursos** están pendientes de soporte upstream — el cliente oficial puentea herramientas y recursos, pero no prompts ni suscripciones.
 - **Códigos de salida / stderr** se etiquetan *pendiente de soporte upstream* hasta que el cliente los exponga.
 - **Panel de solo lectura** — la consola nunca falsifica un estado de conexión; los campos no observables leen `unknown` / `-1` / `—`.
+- **Las escrituras se verifican, no se garantizan** — la consola re-verifica cada escritura contra el estado re-aplicado del loader y falla honestamente cuando el loader no la aplicó (p. ej. una fila insertada por `$DSH_HOME/cordis.patch.yml`, que un parche de capa de perfil no puede alcanzar).
 
 ## Development
 

package/README-hi.md

@@ -26,7 +26,7 @@
 
 | सतह | स्थिति |
 |---|---|
-| Harness | DeepSeek Harness `dsh-v0.1.5-rc.2` (GitHub tag, 2026-09-11 को सत्यापित) (2026-09-09 को अनुकूलित): सत्र लिफ़ाफ़ा अपना ignorable फ़ील्ड केवल संग्रहीत-लॉग पठन संगतता के लिए रखता है - Session.append अभी भी इसे स्टैम्प नहीं कर सकता, इसलिए गेट व्यवहार अपरिवर्तित है। 2026-09-11 को dsh-v0.1.5-rc.2 master checkout के विरुद्ध सत्यापित (पूर्ण गेट शृंखला + प्रोफ़ाइल इंस्टॉल स्मोक)। |
+| Harness | DeepSeek Harness `dsh-v0.1.6-alpha.1` (GitHub tag, 2026-09-16 को सत्यापित): 0.1.6-alpha.1 checkout के विरुद्ध पूर्ण गेट शृंखला (typecheck / typecheck:ci / test / build / verify / package) + प्रोफ़ाइल इंस्टॉल स्मोक। `@deepseek-ai/dsh-mcp-resources` सेवा अब Resources जोड़ती है, जिसे कंसोल फ़ीचर-डिटेक्ट कर रीड-ओनली ब्राउज़ करता है। पिछली आधार रेखा: `dsh-v0.1.5-rc.2` (2026-09-11 को सत्यापित)। |
 | Node | `^22.19.0 \|\| >=24.0.0` |
 | प्लेटफ़ॉर्म | Web GUI (दोहरा चेहरा: host + browser) |
 | मॉडल | कोई भी (पैनल रीड-ओनली है; केवल `/mcp` आउटपुट मॉडल-पठनीय है) |
@@ -40,7 +40,8 @@
 - **`/mcp <server> health`** — व्युत्पन्न स्व-उपचार सुझाव (ENOENT → अनुपलब्ध निर्भरता, ECONNREFUSED, टाइमआउट, 401/403/404, DNS, दर सीमा, रीकनेक्ट समाप्त…); एग्ज़िट कोड / stderr पूँछ ईमानदारी से *upstream समर्थन की प्रतीक्षा में* चिह्नित।
 - **`/mcp <server> call <tool> [json]`** — **आधिकारिक टूल पाइपलाइन** (`ctx.tools.execute()`) से परीक्षण कॉल; प्री-एग्ज़ीक्यूट अनुमति नीति, अनुमोदन, guards और post-execute सब लागू।
 - **सेटिंग्स → प्लगइन्स → MCP टैब** — बैज, निदान और प्रोब वाले स्थिति कार्ड, साथ में सर्वर CRUD और टूल ट्रायल कंसोल।
-- **सर्वर CRUD** — जोड़/बदल/हटा फ़ॉर्म → `insert`/`set`/`set disabled` अंश → क्लिपबोर्ड कॉपी या अनुमोदित लेखन, स्वचालित बैकअप के साथ।
+- **सर्वर CRUD** — जोड़/बदल/हटा फ़ॉर्म → जोड़ने के लिए `insert` और बदलने/हटाने के लिए id-लक्षित ओवरराइड (`- id:` + `name:` + `disabled:`/`config:`) → क्लिपबोर्ड कॉपी या अनुमोदित लेखन, स्वचालित बैकअप और loader के विरुद्ध पुनः सत्यापन के साथ।
+- **Resources ब्राउज़िंग** — आधिकारिक `list_mcp_resources` / `list_mcp_resource_templates` / `read_mcp_resource` टूल से रीड-ओनली संसाधन सूची, टेम्पलेट सूची और URI पठन (`@deepseek-ai/dsh-mcp-resources` सेवा द्वारा जोड़े गए); परिणाम केवल टैब में दिखते हैं, मॉडल संदर्भ में कभी नहीं।
 - **टूल ट्रायल कंसोल** — सर्वर → `mcp__*` टूल → JSON तर्क → कैनोनिकल JSON परिणाम + रेंडर सामग्री; `trialMaxResultChars` से सीमित; केवल पैनल, मॉडल संदर्भ में कभी नहीं।
 
 ## Architecture: official client = bridge, this plugin = console
@@ -78,7 +79,7 @@
 |---|---|---|
 | सर्वर जोड़ें | YAML संपादित करें, इंडेंट/कोट्स का ध्यान | फ़ॉर्म → patch अंश → **कॉपी** या **लिखें** (अनुमोदन + बैकअप) |
 | सर्वर बदलें | YAML संपादित करें, रीस्टार्ट/हॉट-रीलोड | लाइव पंक्ति से पहले से भरा फ़ॉर्म; बिना बदले सीक्रेट host पर ही रहते हैं |
-| सर्वर हटाएँ | पंक्ति मिटाएँ | `set disabled: true` ऑपरेशन (patch शब्दावली में remove नहीं) — कभी भी फिर से सक्षम करने योग्य |
+| सर्वर हटाएँ | पंक्ति मिटाएँ | `- id:` + `disabled: true` ओवरराइड (patch शब्दावली में remove नहीं) — कभी भी फिर से सक्षम करने योग्य |
 | स्थिति देखें | लॉग पढ़ें | बैज + रीकनेक्ट + अंतिम त्रुटि, `mcp/status` से लाइव |
 | टूल आज़माएँ | मॉडल से कहें | ट्रायल कंसोल → आधिकारिक `ctx.tools.execute()` पाइपलाइन (अनुमतियाँ और अनुमोदन लागू) |
 | विफलताओं का निदान | लॉग grep करें | `/mcp <server> health` व्युत्पन्न सुझावों के साथ |
@@ -132,6 +133,8 @@ dsh --profile web --dump-config | grep -A3 'id: mcp-panel'
 | `trialTimeoutMs` | `120000` | प्रति ट्रायल कॉल पैनल-साइड समय सीमा (ms) |
 | `trialMaxResultChars` | `60000` | ट्रायल परिणाम payload की सीमा (वर्ण) |
 | `writeEnabled` | `true` | कठोर स्विच: `false` हर प्रोफ़ाइल लेखन अस्वीकार करता है (कॉपी फिर भी चलती है) |
+| `writeVerifyEnabled` | `true` | सफलता बताने से पहले हर लेखन को loader के पुनः-लागू अवस्था के विरुद्ध सत्यापित करता है |
+| `writeVerifyTimeoutMs` | `3000` | लेखन सत्यापन के लिए पोलिंग बजट (मि.से.) |
 | `backupCount` | `5` | प्रति लेखन रखे गए `cordis.patch.yml` बैकअप |
 | `catalogEntries` | `[]` | अनुशंसित निर्देशिका के लिए उपयोगकर्ता ओवरले: प्रविष्टियाँ जोड़ता है; समान `id` वाली प्रविष्टि अंतर्निहित को बदल देती है |
 
@@ -149,7 +152,9 @@ dsh --profile web --dump-config | grep -A3 'id: mcp-panel'
 
 ## Resources & Prompts
 
-आधिकारिक क्लाइंट दस्तावेज़ कहता है कि *"Tools are the only bridged MCP capability"* — Resources और Prompts स्थगित हैं। कंसोल प्रस्तावित upstream catalog seam को फ़ीचर-डिटेक्ट करता है और उसके आते ही रीड-ओनली सूचियाँ दिखाएगा; तब तक क्षमता बोर्ड दोनों को **upstream समर्थन की प्रतीक्षा में** दर्शाता है।
+Resources अब upstream जोड़ चुका है: base bundle `@deepseek-ai/dsh-mcp-resources` माउंट करता है, आधिकारिक क्लाइंट हर कनेक्शन का संसाधन provider `ctx.mcpResources` में पंजीकृत करता है, और वह पैकेज तीन साझा टूल (`list_mcp_resources`, `list_mcp_resource_templates`, `read_mcp_resource`) रखता है। कंसोल सेवा और पंजीकृत टूल को फ़ीचर-डिटेक्ट करता है, और हर सर्वर कार्ड पर रीड-ओनली Resources ब्राउज़र (सूची / टेम्पलेट / URI पठन) देता है — हर कॉल आधिकारिक टूल पाइपलाइन से जाती है और परिणाम मॉडल संदर्भ में कभी नहीं जाते।
+
+MCP **prompt टेम्पलेट** और **संसाधन सदस्यताएँ** अभी भी upstream स्थगित हैं; क्षमता बोर्ड Prompts को **upstream समर्थन की प्रतीक्षा में** दर्शाता है।
 
 ## Permissions & data
 
@@ -160,14 +165,15 @@ dsh --profile web --dump-config | grep -A3 'id: mcp-panel'
 
 - **पुल पुल ही रहता है।** ट्रांसपोर्ट, OAuth या प्रोटोकॉल में कोई बदलाव नहीं; प्रति सर्वर एक mcp-client पंक्ति, ठीक हाथ से लिखी हुई।
 - **कोई नकली स्थिति नहीं।** बिना upstream अवलोकन वाले कनेक्शन फ़ील्ड `unknown` / `—` और `statusSource: 'derived'` पढ़ते हैं; एग्ज़िट कोड और stderr पूँछ कभी गढ़े नहीं जाते।
-- **लेखन केवल-जोड़ने वाला, अनुमोदित, बैकअप सहित।** कंसोल कभी `cordis.patch.yml` को दोबारा नहीं लिखता; यह जनरेट किए ऑपरेशन जोड़ता है और नवीनतम `backupCount` बैकअप रखता है।
+- **लेखन केवल-जोड़ने वाला, अनुमोदित, बैकअप सहित।** कंसोल कभी `cordis.patch.yml` को दोबारा नहीं लिखता; यह जनरेट किए ऑपरेशन जोड़ता है, नवीनतम `backupCount` बैकअप रखता है, और सफलता बताने से पहले हर लेखन को loader के पुनः-लागू अवस्था के विरुद्ध सत्यापित करता है (छोड़ा गया patch कभी सफल नहीं दिखता)।
 - **कोई प्रॉम्प्ट इंजेक्शन नहीं।** पैनल कोई प्रॉम्प्ट अनुभाग पंजीकृत नहीं करता; मॉडल को दिखने वाला उसका एकमात्र पाठ दो टूल/कमांड विवरण हैं।
 
 ## Known limitations
 
-- **Resources और Prompts** upstream समर्थन की प्रतीक्षा में — आधिकारिक क्लाइंट केवल टूल जोड़ता है।
+- **prompt टेम्पलेट और संसाधन सदस्यताएँ** upstream समर्थन की प्रतीक्षा में — आधिकारिक क्लाइंट टूल और संसाधन जोड़ता है, पर prompts या सदस्यताएँ नहीं।
 - **एग्ज़िट कोड / stderr पूँछ** तब तक *upstream समर्थन की प्रतीक्षा में* चिह्नित रहते हैं जब तक क्लाइंट उन्हें उजागर न करे।
 - **रीड-ओनली पैनल** — कंसोल कभी कनेक्शन स्थिति नहीं गढ़ता; अप्रेक्षणीय फ़ील्ड `unknown` / `-1` / `—` पढ़ते हैं।
+- **लेखन सत्यापित होते हैं, गारंटी नहीं** — कंसोल हर लेखन को loader के पुनः-लागू अवस्था के विरुद्ध सत्यापित करता है और जब loader उसे लागू न करे (जैसे `$DSH_HOME/cordis.patch.yml` द्वारा डाली गई पंक्ति, जिसे प्रोफ़ाइल-परत patch नहीं पहुँच सकता) तो ईमानदारी से विफल होता है।
 
 ## Development
 

package/README-pt.md

@@ -26,7 +26,7 @@
 
 | Superfície | Status |
 |---|---|
-| Harness | DeepSeek Harness `dsh-v0.1.5-rc.2` (tag do GitHub, verificado em 2026-09-11) (adaptado em 2026-09-09): o envelope de sessão mantém seu campo ignorable apenas para compatibilidade de leitura de logs armazenados - o Session.append ainda não consegue estampá-lo, então o comportamento da porta não muda. Verificado em 2026-09-11 contra o checkout master dsh-v0.1.5-rc.2 (cadeia completa de portas + smoke de instalação de perfil). |
+| Harness | DeepSeek Harness `dsh-v0.1.6-alpha.1` (tag do GitHub, verificado em 2026-09-16): cadeia completa de portas (typecheck / typecheck:ci / test / build / verify / package) mais smoke de instalação de perfil contra o checkout 0.1.6-alpha.1. O serviço `@deepseek-ai/dsh-mcp-resources` agora pontua Resources, que o console detecta e navega somente leitura. Linha de base anterior: `dsh-v0.1.5-rc.2` (verificado em 2026-09-11). |
 | Node | `^22.19.0 \|\| >=24.0.0` |
 | Plataformas | Web GUI (duas faces: host + navegador) |
 | Modelo | Qualquer (o painel é somente leitura; só a saída de `/mcp` é legível pelo modelo) |
@@ -40,7 +40,8 @@ O `dsh-mcp-panel` é a camada de experiência sobre o cliente MCP oficial: uma v
 - **`/mcp <servidor> health`** — sugestões de autorreparo derivadas (ENOENT → dependência ausente, ECONNREFUSED, timeouts, 401/403/404, DNS, rate limit, reconexão esgotada…); código de saída / cauda de stderr rotulados honestamente como *aguardando suporte upstream*.
 - **`/mcp <servidor> call <tool> [json]`** — chamada de teste pelo **pipeline oficial de ferramentas** (`ctx.tools.execute()`); política de permissão pré-execução, aprovação, guards e pós-execução, tudo em vigor.
 - **Configurações → Plugins → MCP** — cartões de status com selos, diagnósticos e sondas, mais o CRUD de servidores e o banco de testes de ferramentas.
-- **CRUD de servidores** — formulários de adicionar/editar/remover → fragmentos `insert`/`set`/`set disabled` → cópia para a área de transferência ou gravação com aprovação e backups automáticos.
+- **CRUD de servidores** — formulários de adicionar/editar/remover → `insert` para adicionar e anulações dirigidas por id (`- id:` + `name:` + `disabled:`/`config:`) para editar/remover → cópia para a área de transferência ou gravação com aprovação, backups automáticos e re-verificação contra o loader.
+- **Navegação de Resources** — listagem de recursos, listagem de templates e leituras de URI somente leitura pelas ferramentas oficiais `list_mcp_resources` / `list_mcp_resource_templates` / `read_mcp_resource` (ponteadas pelo serviço `@deepseek-ai/dsh-mcp-resources`); os resultados aparecem apenas na aba, nunca no contexto do modelo.
 - **Banco de testes de ferramentas** — servidor → ferramenta `mcp__*` → argumentos JSON → resultado JSON canônico + conteúdo renderizado; limitado por `trialMaxResultChars`; somente painel, nunca contexto do modelo.
 
 ## Architecture: official client = bridge, this plugin = console
@@ -78,7 +79,7 @@ O console **lê** o cliente pelo seu seam de observabilidade `mcp/status` (event
 |---|---|---|
 | Adicionar servidor | Editar YAML, cuidar indentação/aspas | Formulário → fragmento de patch → **copiar** ou **gravar** (aprovação + backup) |
 | Editar servidor | Editar YAML, reiniciar/recarga a quente | Formulário pré-preenchido da linha ao vivo; segredos inalterados preservam o valor no host |
-| Remover servidor | Apagar a linha | Operação `set disabled: true` (o vocabulário de patches não tem remove) — re-habilitável |
+| Remover servidor | Apagar a linha | Anulação `- id:` + `disabled: true` (o vocabulário de patches não tem remove) — re-habilitável |
 | Ver status | Ler logs | Selos + reconexões + último erro, ao vivo do `mcp/status` |
 | Testar uma ferramenta | Pedir ao modelo | Banco de testes → pipeline oficial `ctx.tools.execute()` (permissões e aprovação em vigor) |
 | Diagnosticar falhas | grep de logs | `/mcp <servidor> health` com sugestões derivadas |
@@ -132,6 +133,8 @@ Todas as opções são campos Schemastery `Config` (modificáveis a partir do co
 | `trialTimeoutMs` | `120000` | Prazo do painel por chamada de teste em ms |
 | `trialMaxResultChars` | `60000` | Teto do payload do resultado de teste em caracteres |
 | `writeEnabled` | `true` | Interruptor de segurança: `false` rejeita toda gravação (copiar continua funcionando) |
+| `writeVerifyEnabled` | `true` | Re-verifica cada gravação contra o estado re-aplicado do loader antes de reportar sucesso |
+| `writeVerifyTimeoutMs` | `3000` | Orçamento de sondagem para a verificação de gravação em ms |
 | `backupCount` | `5` | Backups de `cordis.patch.yml` retidos por gravação |
 | `catalogEntries` | `[]` | Sobreposição do usuário para o diretório recomendado: anexa entradas; uma entrada com o mesmo `id` substitui a integrada |
 
@@ -149,7 +152,9 @@ Todas as opções são campos Schemastery `Config` (modificáveis a partir do co
 
 ## Resources & Prompts
 
-O cliente oficial documenta que *"Tools are the only bridged MCP capability"* — Resources e Prompts estão adiados. O console detecta um seam de catálogo proposto e exibirá listas somente leitura no dia em que for enviado; até lá o painel de capacidades marca ambos **aguardando suporte upstream**.
+Resources JÁ são ponteados upstream: o bundle base monta `@deepseek-ai/dsh-mcp-resources`, o cliente oficial registra o provider de recursos de cada conexão em `ctx.mcpResources`, e esse pacote possui as três ferramentas compartilhadas (`list_mcp_resources`, `list_mcp_resource_templates`, `read_mcp_resource`). O console detecta o serviço e as ferramentas registradas, e oferece um navegador de Resources somente leitura (lista / templates / leitura de URI) em cada cartão de servidor — cada chamada passa pelo pipeline OFICIAL de ferramentas e os resultados nunca entram no contexto do modelo.
+
+**Templates de prompts** e **assinaturas de recursos** permanecem adiados upstream; o painel de capacidades marca Prompts **aguardando suporte upstream**.
 
 ## Permissions & data
 
@@ -160,14 +165,15 @@ O cliente oficial documenta que *"Tools are the only bridged MCP capability"*
 
 - **A ponte continua sendo a ponte.** Sem mudanças de transporte, OAuth ou protocolo; uma linha mcp-client por servidor, exatamente como à mão.
 - **Sem status falso.** Campos de conexão sem observações upstream leem `unknown` / `—` com `statusSource: 'derived'`; códigos de saída e stderr nunca são inventados.
-- **Gravações somente-anexar, com aprovação e backup.** O console nunca reescreve o `cordis.patch.yml`; anexa operações geradas e conserva os `backupCount` backups mais recentes.
+- **Gravações somente-anexar, com aprovação e backup.** O console nunca reescreve o `cordis.patch.yml`; anexa operações geradas, conserva os `backupCount` backups mais recentes e re-verifica cada gravação contra o estado re-aplicado do loader antes de reportar sucesso (um patch omitido nunca aparece como sucesso).
 - **Sem injeção de prompts.** O painel não registra seções de prompt; seu único texto visível ao modelo são as duas descrições de ferramenta/comando.
 
 ## Known limitations
 
-- **Resources e Prompts** aguardam suporte upstream — o cliente oficial só pontua ferramentas.
+- **Templates de prompts e assinaturas de recursos** aguardam suporte upstream — o cliente oficial pontua ferramentas e recursos, mas não prompts nem assinaturas.
 - **Códigos de saída / caudas de stderr** são rotulados *aguardando suporte upstream* até o cliente expô-los.
 - **Painel somente leitura** — o console nunca falsifica um estado de conexão; campos não observáveis leem `unknown` / `-1` / `—`.
+- **Gravações são verificadas, não garantidas** — o console re-verifica cada gravação contra o estado re-aplicado do loader e falha honestamente quando o loader não a aplicou (ex.: uma linha inserida por `$DSH_HOME/cordis.patch.yml`, que um patch de camada de perfil não alcança).
 
 ## Development
 

package/README-zh.md

@@ -28,7 +28,7 @@
 
 | 维度 | 状态 |
 |---|---|
-| Harness | DeepSeek Harness `dsh-v0.1.5-rc.2`（GitHub tag，2026-09-11 已核验）（2026-09-09 已适配）：会话信封保留 ignorable 字段但仅用于存量日志读取兼容——Session.append 仍无法盖章，门控行为不变。已于 2026-09-11 对照 dsh-v0.1.5-rc.2 master checkout 核验（全量门禁链 + profile 安装冒烟）。 |
+| Harness | DeepSeek Harness `dsh-v0.1.6-alpha.1`（GitHub tag，2026-09-16 已核验）：全量门禁链（typecheck / typecheck:ci / test / build / verify / package）+ 对照 0.1.6-alpha.1 checkout 的 profile 安装冒烟。上游 `@deepseek-ai/dsh-mcp-resources` 现已桥接 Resources，控制台特征探测后提供只读浏览。上一基线：`dsh-v0.1.5-rc.2`（2026-09-11 已核验）。 |
 | Node | `^22.19.0 \|\| >=24.0.0` |
 | 平台 | Web GUI（双面：Host + 浏览器） |
 | 模型 | 任意（面板只读；仅 `/mcp` 输出对模型可见） |
@@ -42,7 +42,8 @@
 - **`/mcp <server> health`** —— 派生自愈建议（ENOENT → 依赖缺失、ECONNREFUSED、超时、401/403/404、DNS、限流、重连耗尽…）；退出码 / stderr 尾部如实标注*待官方支持*。
 - **`/mcp <server> call <tool> [json]`** —— 经**官方工具管线**（`ctx.tools.execute()`）试用调用；pre-execute 权限策略、审批、guard、post-execute 全部生效。
 - **设置 → 插件 → MCP 页** —— 状态卡片（徽章、诊断、探测），加 server CRUD 与工具试用台。
-- **Server CRUD** —— 增删改表单 → `insert`/`set`/`set disabled` 片段 → 剪贴板复制或审批写入，自动备份。
+- **Server CRUD** —— 增删改表单 → 添加用 `insert`，编辑/删除用 id 定向覆盖（`- id:` + `name:` + `disabled:`/`config:`）→ 剪贴板复制或审批写入，自动备份，并在报成功前对照 loader 复验。
+- **Resources 浏览** —— 经官方 `list_mcp_resources` / `list_mcp_resource_templates` / `read_mcp_resource` 工具只读列出资源、模板与 URI 读取（由上游 `@deepseek-ai/dsh-mcp-resources` 服务桥接）；结果只显示在本页、绝不进入模型上下文。
 - **工具试用台** —— 选 server → 选 `mcp__*` 工具 → JSON 填参 → 规范 JSON 结果 + render 内容；按 `trialMaxResultChars` 截断；仅面板可见、永不进入模型上下文。
 
 ## Architecture: official client = bridge, this plugin = console
@@ -80,7 +81,7 @@
 |---|---|---|
 | 添加服务器 | 改 YAML，注意缩进与引号 | 表单 → patch 片段 → **一键复制**或**写入**（审批 + 自动备份） |
 | 修改服务器 | 改 YAML，重启/热重载 | 表单预填当前行；未改动的密钥在 host 侧保留原值 |
-| 删除服务器 | 删掉该行 | 追加 `set disabled: true` 操作（patch 词汇表没有 remove）——可随时重新启用 |
+| 删除服务器 | 删掉该行 | 追加 `- id:` + `disabled: true` 覆盖操作（patch 词汇表没有 remove）——可随时重新启用 |
 | 查看状态 | 翻日志 | 徽章 + 重连次数 + 最近错误，来自 `mcp/status` seam 实时数据 |
 | 试用工具 | 让模型调用 | 试用台 → 官方 `ctx.tools.execute()` 管线（权限与审批全程生效） |
 | 排查故障 | grep 日志 | `/mcp <server> health` 派生自愈建议 |
@@ -134,6 +135,8 @@ dsh --profile web --dump-config | grep -A3 'id: mcp-panel'
 | `trialTimeoutMs` | `120000` | 每次试用调用的面板侧截止时间（ms） |
 | `trialMaxResultChars` | `60000` | 试用结果载荷上限（字符） |
 | `writeEnabled` | `true` | 总开关：`false` 拒绝一切 profile 写入（仍可复制片段） |
+| `writeVerifyEnabled` | `true` | 每次写入在报成功前对照 loader 重放后的状态复验 |
+| `writeVerifyTimeoutMs` | `3000` | 写入复验的轮询预算（毫秒） |
 | `backupCount` | `5` | 每次写入保留的 `cordis.patch.yml` 备份数 |
 | `catalogEntries` | `[]` | 推荐服务器目录的用户覆盖：追加条目，同 `id` 的条目替换内置条目 |
 
@@ -151,7 +154,9 @@ dsh --profile web --dump-config | grep -A3 'id: mcp-panel'
 
 ## Resources & Prompts
 
-官方 client 明确记载 *"Tools 是当前唯一桥接的 MCP 能力"*——Resources 与 Prompts 处于 deferred 状态。控制台特征探测了上游提案中的 catalog seam，一旦落地即可只读展示列表；在此之前，能力一览中二者均标注**待官方支持**。
+Resources 已由上游桥接：base bundle 默认挂载 `@deepseek-ai/dsh-mcp-resources`，官方 client 把每个连接的资源 provider 注册进 `ctx.mcpResources`，该包拥有三个共享工具（`list_mcp_resources`、`list_mcp_resource_templates`、`read_mcp_resource`）。控制台特征探测该服务与已注册工具，并在每个 server 卡上提供只读 Resources 浏览（列表 / 模板 / URI 读取）——每次调用都走**官方工具管线**，结果只显示在本页、绝不进入模型上下文。
+
+MCP **prompt 模板**与**资源订阅**仍待上游；能力一览中 Prompts 标注**待官方支持**。
 
 ## Permissions & data
 
@@ -162,14 +167,15 @@ dsh --profile web --dump-config | grep -A3 'id: mcp-panel'
 
 - **桥接层仍是桥接层。** 不改传输、OAuth、协议；每个 server 一行 mcp-client，与你手写完全一致。
 - **不伪造状态。** 无上游观测时连接字段显示 `unknown` / `—` 并标注 `statusSource: 'derived'`；退出码与 stderr 尾部绝不臆造。
-- **写入只追加、走审批、先备份。** 控制台从不改写 `cordis.patch.yml`：只追加生成的操作，并保留最新 `backupCount` 份备份。
+- **写入只追加、走审批、先备份。** 控制台从不改写 `cordis.patch.yml`：只追加生成的操作，保留最新 `backupCount` 份备份，并在报成功前对照 loader 重放后的状态复验（被跳过的补丁绝不显示为成功）。
 - **零提示词注入。** 本插件不注册任何提示词段落；对模型可见的文本只有两个工具/命令描述。
 
 ## Known limitations
 
-- **Resources 与 Prompts** 待官方支持——官方 client 仅桥接工具。
+- **prompt 模板与资源订阅** 待官方支持——官方 client 桥接工具与资源，但未桥接 prompts 与订阅。
 - **退出码 / stderr 尾部** 在 client 暴露前如实标注*待官方支持*。
 - **只读面板** —— 控制台从不伪造连接状态；不可观测字段显示 `unknown` / `-1` / `—`。
+- **写入复验而非保证** —— 控制台对照 loader 重放后的状态复验每次写入；loader 未应用时（例如该行由 `$DSH_HOME/cordis.patch.yml` 插入、profile 层补丁够不着）如实报失败。
 
 ## Development
 

package/README.md

@@ -29,7 +29,7 @@
 
 | Surface | Status |
 |---|---|
-| Harness | DeepSeek Harness `dsh-v0.1.5-rc.2` (GitHub tag, verified 2026-09-11) (adapted 2026-09-09): the session envelope keeps its ignorable field for stored-log read compatibility only - Session.append still cannot stamp it, so audit-gate behavior is unchanged. Verified 2026-09-11 against the dsh-v0.1.5-rc.2 master checkout (full gate chain + profile install smoke). |
+| Harness | DeepSeek Harness `dsh-v0.1.6-alpha.1` (GitHub tag, verified 2026-09-16): full gate chain (typecheck / typecheck:ci / test / build / verify / package) plus a profile install smoke against the 0.1.6-alpha.1 checkout. The shipped `@deepseek-ai/dsh-mcp-resources` service now bridges Resources, which the console feature-detects and browses read-only. Previous baseline: `dsh-v0.1.5-rc.2` (verified 2026-09-11). |
 | Node | `^22.19.0 \|\| >=24.0.0` |
 | Platforms | Web GUI (dual-face: host + browser) |
 | Model | Any (the panel is read-only; only `/mcp` output is model-readable) |
@@ -43,7 +43,8 @@
 - **`/mcp <server> health`** — derived self-heal suggestions (ENOENT → missing dependency, ECONNREFUSED, timeouts, 401/403/404, DNS, rate limit, reconnect exhaustion…); exit code / stderr tail honestly labeled *pending upstream support* until the client exposes them.
 - **`/mcp <server> call <tool> [json]`** — trial-call through the **official tool pipeline** (`ctx.tools.execute()`); pre-execute permission policy, approval, guards, and post-execute all apply.
 - **Settings → Plugins → MCP tab** — status cards with badges, diagnostics, and probes, plus the server CRUD and the tool trial console.
-- **Server CRUD** — add/edit/remove forms → `insert`/`set`/`set disabled` fragments → clipboard copy or approval-gated write with automatic backups.
+- **Server CRUD** — add/edit/remove forms → `insert` for add and id-targeted overrides (`- id:` + `name:` + `disabled:`/`config:`) for edit/remove → clipboard copy or approval-gated write with automatic backups and loader re-verification.
+- **Resources browse** — read-only resource listing, template listing, and URI reads through the official `list_mcp_resources` / `list_mcp_resource_templates` / `read_mcp_resource` tools (bridged by the shipped `@deepseek-ai/dsh-mcp-resources` service); results are shown in the tab only, never in model context.
 - **Recommended directory** — a built-in community MCP server catalog (filesystem, git, github, fetch, playwright, …) served in the snapshot; `catalogEntries` appends/overrides entries, and `catalogToConfigInput` turns one into a one-click add.
 - **Config import/export** — `exportConfigs()` serializes the server rows to a versioned JSON document (a `!!js` row exports as `null` with a reason), and `importPreview()` parses an export back into per-server `add` patch fragments for review.
 - **Tool trial console** — server → `mcp__*` tool → JSON args → canonical JSON result + rendered content; capped by `trialMaxResultChars`; panel-only, never model context.
@@ -83,7 +84,7 @@ The console **reads** the client through its proposed `mcp/status` observability
 |---|---|---|
 | Add a server | Edit YAML, mind indent/quoting | Form → patch fragment → **copy** or **write** (approval + auto backup) |
 | Edit a server | Edit YAML, restart/hot-reload | Form pre-filled from the live row; unchanged secrets keep their raw values host-side |
-| Remove a server | Delete the row | `set disabled: true` operation (the patch vocabulary has no remove) — re-enableable anytime |
+| Remove a server | Delete the row | `- id:` + `disabled: true` override (the patch vocabulary has no remove) — re-enableable anytime |
 | See status | Read logs | Badges + reconnects + last error, live from the `mcp/status` seam |
 | Try a tool | Ask the model to call it | Trial console → official `ctx.tools.execute()` pipeline (permission & approval stay in force) |
 | Diagnose failures | Grep logs | `/mcp <server> health` with derived self-heal suggestions |
@@ -137,6 +138,8 @@ All tunables are Schemastery `Config` fields (changeable from cordis.yml). `cord
 | `trialTimeoutMs` | `120000` | Panel-side deadline per trial call in ms |
 | `trialMaxResultChars` | `60000` | Cap on the trial result payload in chars |
 | `writeEnabled` | `true` | Kill switch: `false` rejects every profile write (copy still works) |
+| `writeVerifyEnabled` | `true` | Re-verify each write against the loader's re-applied state before reporting success |
+| `writeVerifyTimeoutMs` | `3000` | Polling budget for write verification in ms |
 | `backupCount` | `5` | `cordis.patch.yml` backups retained per write |
 | `catalogEntries` | `[]` | User overlay for the recommended server directory: entries append, an entry with the same `id` replaces the built-in one |
 
@@ -154,7 +157,9 @@ All tunables are Schemastery `Config` fields (changeable from cordis.yml). `cord
 
 ## Resources & Prompts
 
-The official client documents that *"Tools are the only bridged MCP capability"* — Resources and Prompts are deferred. The console feature-detects a proposed upstream catalog seam and will show read-only lists the day it ships; until then the capabilities board marks both **pending upstream support**.
+Resources ARE bridged upstream: the base bundle mounts `@deepseek-ai/dsh-mcp-resources`, the official client registers each connection's resource provider into `ctx.mcpResources`, and that package owns the three shared tools (`list_mcp_resources`, `list_mcp_resource_templates`, `read_mcp_resource`). The console feature-detects the service and the registered tools, and offers a read-only Resources browser (list / templates / URI read) on every server card — each call runs through the OFFICIAL tool pipeline, and results never enter model context.
+
+MCP **prompt templates** and **resource subscriptions** remain deferred upstream; the capabilities board marks Prompts **pending upstream support**.
 
 ## Permissions & data
 
@@ -165,14 +170,15 @@ The official client documents that *"Tools are the only bridged MCP capability"*
 
 - **The bridge stays the bridge.** No transport, OAuth, or protocol changes; one mcp-client row per server, exactly as hand-written.
 - **No fake status.** Connection fields without upstream observations read `unknown` / `—` with `statusSource: 'derived'`; exit codes and stderr tails are never invented.
-- **Writes are append-only, approval-gated, and backed up.** The console never rewrites `cordis.patch.yml`; it appends generated operations and keeps the newest `backupCount` backups.
+- **Writes are append-only, approval-gated, and backed up.** The console never rewrites `cordis.patch.yml`; it appends generated operations, keeps the newest `backupCount` backups, and re-verifies every write against the loader's re-applied state before reporting success (a skipped patch never reads as success).
 - **No prompt injection.** The panel registers **no prompt sections**; its only model-facing text is the two tool/command descriptions.
 
 ## Known limitations
 
-- **Resources & Prompts** are pending upstream support — the official client bridges tools only.
+- **Prompt templates & resource subscriptions** are pending upstream support — the official client bridges tools and resources, but not prompts or subscriptions.
 - **Exit codes / stderr tails** are labeled *pending upstream support* until the client exposes them.
 - **Read-only panel** — the console never fakes a connection state; unobservable fields read `unknown` / `-1` / `—`.
+- **Writes are verified, not guaranteed** — the console re-verifies each write against the loader's re-applied state and fails honestly when the loader did not apply it (e.g. a row inserted by `$DSH_HOME/cordis.patch.yml`, which a profile-layer patch cannot reach).
 
 ## Development
 

package/cordis.patch.yml

@@ -32,5 +32,10 @@
         # Kill switch for profile-patch writes: false rejects every write
         # (fragments stay copyable).
         # writeEnabled: true
+        # Verify each write against the loader's re-applied state before
+        # reporting success (a skipped patch must not read as success).
+        # writeVerifyEnabled: true
+        # Polling budget for write verification in milliseconds.
+        # writeVerifyTimeoutMs: 3000
         # Number of cordis.patch.yml backups retained per write.
         # backupCount: 5

package/docs/upstream-proposal.md

@@ -167,3 +167,12 @@ both before and after this proposal lands.
   `PerryLink/deepseek-harness:feat/mcp-client-status-observability-seam`.
 - Everything else in this document stands as written.
 
+## Status check 2026-09-16 (0.1.6-alpha.1)
+
+> Measured read-only against `origin/master` = `0d1f5000` (`0.1.6-alpha.1`, a 666-commit interval since the previous baseline).
+
+- **The `mcp/status` seam is still missing.** `mcp/status` / `mcpStatus` / `McpStatus` -> 0 hits across `packages`; the mcp-client still exposes no per-server status/observability surface, so the panel keeps its feature-detected, derived-status fallback. The proposal and its tripwire remain unchanged.
+- **Resources shipped differently than proposed.** The upstream bridge now provides Resources through a NEW package, `@deepseek-ai/dsh-mcp-resources` (mounted by `packages/bundle/base/cordis.patch.yml:471-472`): the client registers each connection's provider into `ctx.mcpResources`, and the package owns three shared tools (`list_mcp_resources`, `list_mcp_resource_templates`, `read_mcp_resource`). The console's old "catalog seam" idea (`mcpCatalog`) is therefore retired — the panel feature-detects `ctx.mcpResources` + the registered tools and browses resources through the OFFICIAL tools (`src/upstream.ts`, `src/service.ts`, `src/trial.ts`).
+- **Prompts and resource subscriptions remain deferred** (`packages/mcp/mcp-client/README.md:209` documents prompts as unsupported); the capabilities board keeps Prompts `available: false`.
+- The `dsh-mcp-panel` package now runs its full gate chain against `0.1.6-alpha.1` (compatibility baseline raised from `0.1.5-rc.2`).
+