@goodandready/dsh-model-sync 0.3.4 → 0.3.6

package/README.md

@@ -1,85 +1,175 @@
-# @goodandready/dsh-model-sync
-
-Automatic model catalog synchronization for API-key providers in DeepSeek Harness.
-
-The plugin is designed for:
-
-- API-key providers already configured in DSH;
-- API-key providers available in the built-in DSH catalog but not configured yet;
-- future custom API-key provider routes.
-
-OAuth, subscription, and tool-only routes are intentionally outside the scope.
-
-## Status
-
-Runtime, scheduler, HTTP API, reconciliation, model allowlists, and Settings UI are implemented. The release workflow includes staging and production verification before publication.
-
-The Settings UI groups synchronization controls, run summaries, configured providers, diagnostics, notifications, history, and editors into responsive cards that follow the DSH theme.
-Each configured provider row shows `enabled/total` model counts and a temporary `+N` discovery suffix for newly found models; the manual model-selection and policy editors expand directly below that row.
-
-## Development
-
-See the project documentation under ``docs/`` for architecture, research, testing, and release notes.
-
-## API and safety
-
-`GET /dsh-model-sync/status` reports provider state without secrets.
-
-`POST /dsh-model-sync/run` defaults to a read-only dry-run:
-
-```json
-{ "provider": "openai", "dryRun": true }
+# 📦 @goodandready/dsh-model-sync
+
+<div align="center">
+
+<h3>Dynamic LLM Model Catalog Synchronization & Automated Balance Monitoring for DeepSeek Harness</h3>
+
+<p align="center">
+  <a href="https://www.npmjs.com/package/@goodandready/dsh-model-sync"><img src="https://img.shields.io/npm/v/@goodandready/dsh-model-sync.svg?style=for-the-badge&color=6366f1&labelColor=1e1b4b" alt="npm version"></a>
+  <a href="LICENSE"><img src="https://img.shields.io/badge/License-MIT-10b981.svg?style=for-the-badge&color=10b981&labelColor=064e3b" alt="license"></a>
+  <a href="https://github.com/topics/dsh-plugin"><img src="https://img.shields.io/badge/DSH-Plugin-8b5cf6.svg?style=for-the-badge&labelColor=2e1065" alt="DSH Plugin"></a>
+  <a href="https://nodejs.org"><img src="https://img.shields.io/badge/Node-20%2B-f59e0b.svg?style=for-the-badge&labelColor=451a03" alt="Node version"></a>
+</p>
+
+<p align="center">
+  <a href="https://goodandready.app/"><img src="https://img.shields.io/badge/All_Author_Projects-goodandready.app-ff4500.svg?style=for-the-badge&logo=rocket&logoColor=white&labelColor=1a1a2e" alt="All Author Projects"></a>
+</p>
+
+<p align="center">
+  <a href="README.md"><b>🇬🇧 English</b></a> •
+  <a href="README.ru.md"><b>🇷🇺 Русский</b></a> •
+  <a href="README.zh.md"><b>🇨🇳 中文说明</b></a>
+</p>
+
+</div>
+
+---
+
+## ⚡ Overview
+
+**`dsh-model-sync`** keeps your **DeepSeek Harness** model catalog up-to-date with upstream AI providers. 
+
+Instead of manually editing YAML configs whenever a provider launches a new model, adjusts context window limits, or changes pricing, `dsh-model-sync` automatically discovers new releases, updates capability metadata (`vision`, `tools`, `reasoning`, `embeddings`), tracks account balances, and reconciles models on the fly without server restarts.
+
+```mermaid
+graph LR
+    subgraph Trigger [Scheduling & Manual Trigger]
+        Cron[⏰ Background Polling Scheduler] --> Engine[dsh-model-sync Core Engine]
+        WebUI[🖥️ Settings: Sync Now Button] --> Engine
+    end
+
+    subgraph Providers [25+ Upstream Providers]
+        Engine --> Registry{Adapter Registry}
+        Registry -->|Bearer Auth| P1[OpenAI / DeepSeek / OpenRouter / Groq]
+        Registry -->|x-api-key Auth| P2[Anthropic Claude / Custom Gateways]
+        Registry -->|query-key Auth| P3[Google Gemini]
+        Registry -->|Local Probe| P4[Local Ollama / vLLM / SGLang]
+    end
+
+    subgraph Reconcile [Catalog Reconciler & Auditing]
+        P1 --> Normalizer[Model Normalizer & Capabilities Tagger]
+        P2 --> Normalizer
+        P3 --> Normalizer
+        P4 --> Normalizer
+        Normalizer --> Diff[Sync Diff Tracker: Added / Deprecated]
+        Diff --> Catalog[Active DSH Models Catalog]
+    end
+
+    style Trigger fill:#1e1e2e,stroke:#89b4fa,stroke-width:2px,color:#cdd6f4
+    style Providers fill:#181825,stroke:#cba6f7,stroke-width:2px,color:#cdd6f4
+    style Reconcile fill:#11111b,stroke:#a6e3a1,stroke-width:2px,color:#cdd6f4
 ```
 
-Apply is explicit:
-
-```json
-{ "provider": "openai", "dryRun": false, "removeMissing": false }
+---
+
+## ✨ Key Features
+
+* 🔄 **Automated Catalog Discovery**: Syncs model lists, aliases, context windows, and capability flags (`vision`, `tools`, `reasoning`, `embeddings`) across 25+ providers.
+* 🌐 **25+ Built-in Provider Adapters**: Pre-configured discovery for OpenAI, Anthropic, Google, DeepSeek, xAI, OpenRouter, Groq, Mistral, Cerebras, Fireworks, HuggingFace, Moonshot, NVIDIA, Qwen, Together, Xiaomi MiMo, SiliconFlow, and local Ollama.
+* 🔌 **Generic / Custom Adapter Support**: Connect arbitrary OpenAI-compatible `/v1/models` endpoints with configurable auth (`bearer`, `x-api-key`, `query-key`, `none`).
+* 📊 **Balance & Quota Monitoring**: Queries upstream billing endpoints (where supported) to report account credit balances and prevent unexpected exhaustion.
+* 📜 **Sync History & Diff Auditing**: Tracks all newly discovered models, deprecated IDs, and modified capabilities with timestamped diff logs in the Web UI.
+* 🖥️ **Full Web GUI Panel (**Settings → Model Synchronization**)**:
+  * Live status cards per provider with active model counters;
+  * Instant "Sync Now" button for on-demand catalog refreshes;
+  * Provider enable/disable toggles and custom endpoint inputs.
+
+---
+
+## 🛠️ Supported Discovery Providers
+
+| Provider Identifier | Authentication Format | Capabilities Auto-Detection |
+|---|---|---|
+| `openai` | Bearer Token | `vision`, `tools`, `reasoning`, `embeddings` |
+| `anthropic` | `x-api-key` header | `vision`, `tools`, `reasoning` |
+| `google` | Query parameter / Key | `vision`, `tools`, `reasoning`, `embeddings` |
+| `deepseek` | Bearer Token | `tools`, `reasoning` |
+| `xai` | Bearer Token | `vision`, `tools`, `reasoning` |
+| `openrouter` | Bearer Token | Multi-vendor model catalog with pricing & context limits |
+| `groq` | Bearer Token | Ultra-fast inference catalog |
+| `mistral` | Bearer Token | `tools`, `reasoning`, `vision` |
+| `fireworks` | Bearer Token | Open-weights inference endpoints |
+| `huggingface` | Bearer Token | Serverless Inference API catalog |
+| `together` | Bearer Token | Open-source model catalog |
+| `ollama` | Local HTTP (`none`) | Local offline model discovery (`/api/tags`) |
+| `custom` | Configurable | Any custom OpenAI or Google compatible gateway |
+
+---
+
+## 📦 Quick Installation
+
+```bash
+dsh plugin --profile web add @goodandready/dsh-model-sync
 ```
 
-Omit `provider` to discover all enabled API-key providers in one run. The Settings UI exposes this as **Refresh all**.
-
-`POST /dsh-model-sync/selection` stores and applies a per-provider model allowlist:
-
-```json
-{ "provider": "openai", "models": ["gpt-5", "gpt-5-mini"] }
+> [!IMPORTANT]
+> Restart DSH Web UI after installation (`systemctl --user restart dsh-web`) to activate background catalog synchronization.
+
+---
+
+## 🛠️ Enhancements & Fixes in v0.3.6
+
+* 🔑 **Universal Credential Resolution (`credentialRef` & `apiKeyRef`)**:
+  * Added full support for modern DSH `credentialRef` and `apiKeyRef` references across inventory detection, provider adapters, and OpenAI-compatible generic routes.
+  * Providers configured via DSH Credentials Service are now accurately discovered and authenticated without requiring legacy `apiKeyEnv`.
+* ♻️ **Lifecycle Cleanup for Deprecated Models**:
+  * Fixed an issue where models previously marked as `deprecated` would remain indefinitely in settings when removed by the upstream provider. When `removeMissing: true` is enabled, missing deprecated models now cleanly transition to `removed` status once the grace period expires.
+* ⏰ **Accurate Scheduler Inactive Status**:
+  * Corrected `scheduler.status()` to return `nextRunAt: null` when background synchronization is disabled, avoiding misleading run times in the Web UI.
+* 🛡️ **Web UI Resilience & Big Catalog Scalability**:
+  * Eliminated potential UI crashes during model sorting by safely handling missing model display names.
+  * Replaced spread operations on model arrays with iterative aggregations, preventing call stack overflow (`RangeError`) on massive catalogs (10,000+ models).
+  * Surfaced detailed probe error messages in the UI when testing individual models via the `▶` button.
+  * Localized cache expiration and offline status strings in Russian and English.
+* ⚡ **Deterministic Model Diffing & HTTP Stream Safety**:
+  * Implemented stable key-sorted serialization in `diffModels` and `sameModel` to eliminate false-positive diffs caused by non-deterministic JSON key order.
+  * Added early stream destruction on oversized HTTP request bodies to prevent memory bloat.
+
+---
+
+## 🚀 Enhancements in v0.3.5
+
+* ⏱️ **Adaptive Rate Limit & Retry-After Handling**:
+  * Automatically parses HTTP headers `Retry-After` (in seconds or RFC dates) and `x-ratelimit-reset`.
+  * Exponential backoff prioritizes the exact delay requested by upstream providers, eliminating unnecessary throttling.
+  * Re-classified HTTP 429 errors under a dedicated `rate-limit` diagnostic label.
+* 🔄 **Optimistic Concurrency & Conflict Auto-Recovery**:
+  * Added automated retry loop for `SETTINGS_CONFLICT` (HTTP 409). When concurrent settings updates occur in DSH, the plugin seamlessly re-fetches the latest revision and reapplies the catalog mutations.
+* 🎯 **Advanced Web UI Picker & Filtering**:
+  * **Tokenized Smart Search**: Multi-word queries (`claude 3.5 sonnet`) match across model IDs, display names, tags, and descriptions simultaneously.
+  * **Quick Filter Chips**: One-click toggles for `Context ≥ 100k` and `Cheap ≤ $1/1M` in addition to capability toggles (`Vision`, `Tools`, `Reasoning`, `Embeddings`).
+  * **Batch Selection Controls**: One-click buttons to *Select all shown*, *Deselect all shown*, and *Invert selection*.
+  * **Chunked Rendering / Pagination**: Renders models in batches of 50 to maintain smooth 60fps UI performance even with catalogs containing 1,000+ models.
+  * **Deprecation Badges**: Models with scheduled retirement dates display an explicit countdown badge (`Deprecates in Nd`).
+* 💾 **Storage & Metadata Optimizations**:
+  * Robust context limit parser supporting suffix multipliers (`128k`, `1M`, `128,000`).
+  * Automatic history compaction: historical snapshots older than the 5 latest sync runs have heavy before/after dumps pruned while retaining summary diffs, keeping `settings.yaml` ultra-compact.
+
+---
+
+## ⚙️ Configuration Reference (`settings.yaml`)
+
+```yaml
+dsh-model-sync:
+  enabled: true
+  syncIntervalMinutes: 60
+  autoReconcile: true
+  providers:
+    openrouter:
+      enabled: true
+      keyEnv: OPENROUTER_API_KEY
+    deepseek:
+      enabled: true
+      keyEnv: DEEPSEEK_API_KEY
+    custom:
+      enabled: false
+      baseURL: http://127.0.0.1:8000/v1
+      authType: bearer
+      keyEnv: CUSTOM_API_KEY
 ```
 
-An empty `models` array means all models in the latest available catalog. The selected catalog is written to DSH's `llm-pi-ai` settings, so the standard DSH model picker refreshes from the resulting list.
-
-A dry-run renders a per-provider diff preview (added, removed, and metadata-changed models). Use the explicit **Apply preview** action to write the catalog; stale-model removal still requires its confirmation.
-
-The model picker exposes a capability filter based only on explicit normalized metadata (`vision`, `tools`, `reasoning`, and `embeddings`). Multiple selected capabilities use an all-of match, and filtering does not discard hidden selections.
-
-The settings card uses the same spacing, typography, borders, and theme variables as the built-in DSH plugin cards in both light and dark themes.
-
-
-The plugin-owned `adapterRegistry` setting adds an opt-in declarative adapter without copying code. A provider entry can specify `endpoint`, `auth` (`bearer`, `x-api-key`, `query-key`, or `none`), `parser`, `modelsPath`, field paths, and explicit capability mappings. Endpoints must be HTTPS except loopback HTTP; credential-like static headers and URL query/fragment data are rejected. Existing built-in descriptors and OpenAI-compatible routes remain the default. Runtime integrations can register an explicit adapter implementation through the registry escape hatch, provided it implements `discover` and `health`.
-
-The full catalog discovered by an applied run is cached in the plugin's own settings
-separately from the model allowlist. After a restart, the model chooser can show the
-full last-known catalog while the standard DSH picker still exposes only the selected
-models. Dry-run remains read-only and does not persist this cache.
-
-
-`GET /dsh-model-sync/report` returns the latest stable aggregate report. It distinguishes success, partial success, failure, and empty runs, includes provider/model counts, and labels whether the source was manual or scheduled. `GET /dsh-model-sync/notifications` returns the bounded in-app notification ledger; repeated identical failures update an occurrence counter instead of creating noise. `POST /dsh-model-sync/notifications/read` and `POST /dsh-model-sync/notifications/acknowledge` accept `{ "id": "sync-notice-..." }` and persist the corresponding state for applied runs. Report and notification text is redacted before it leaves the synchronizer; no webhook or MCP dependency is required.
-
-`GET /dsh-model-sync/credentials` returns bounded diagnostics for configured provider credential references. It calls the DSH credentials `describe` seam only, reports masked reference labels, source/configured state, rotation order, and the last safe request outcome. It never resolves or returns a credential value. `POST /dsh-model-sync/credentials/check` runs the health probe and returns the same diagnostics in one response; a failed reference is recorded per provider without hiding the remaining rotation entries.
-
-`GET /dsh-model-sync/history` returns the bounded history of applied runs. Use
-`?provider=<id>&details=true` for one provider's full snapshots and metadata diff.
-`POST /dsh-model-sync/history/rollback` accepts `{ "historyId": "sync-…", "provider": "openai" }`
-and restores only that provider's catalog; it never changes the model allowlist.
-History is written only by non-dry-run synchronization, is retained according to
-`historyLimit` (1–200, default 50), and redacts credential-shaped error values.
-Rename detection requires an explicit provider `aliases` or `previousIds` field;
-model-id similarity is never guessed.
+---
 
+## 📄 License
 
-Lifecycle protection keeps a model `stale` after a missing discovery and only
-moves it to `removed` after the configured `staleGraceRuns` and an explicit
-removal confirmation. Provider-declared `deprecated`/`deprecationDate`/
-`expiration_date` metadata is retained as an audit status. Repeated transport or
-credential failures do not advance lifecycle counters. The full catalog cache
-retains stale entries until the user confirms removal; the model picker defaults
-to active models while explicit selections may still name a retained model.
+MIT © [GooDAnDReaDY](https://github.com/GooDAnDReaDY)

package/README.ru.md

@@ -0,0 +1,152 @@
+# 📦 @goodandready/dsh-model-sync
+
+<div align="center">
+
+<h3>Динамическая синхронизация каталогов моделей и автоматический мониторинг баланса для DeepSeek Harness</h3>
+
+<p align="center">
+  <a href="https://www.npmjs.com/package/@goodandready/dsh-model-sync"><img src="https://img.shields.io/npm/v/@goodandready/dsh-model-sync.svg?style=for-the-badge&color=6366f1&labelColor=1e1b4b" alt="npm version"></a>
+  <a href="LICENSE"><img src="https://img.shields.io/badge/License-MIT-10b981.svg?style=for-the-badge&color=10b981&labelColor=064e3b" alt="license"></a>
+  <a href="https://github.com/topics/dsh-plugin"><img src="https://img.shields.io/badge/DSH-Plugin-8b5cf6.svg?style=for-the-badge&labelColor=2e1065" alt="DSH Plugin"></a>
+  <a href="https://nodejs.org"><img src="https://img.shields.io/badge/Node-20%2B-f59e0b.svg?style=for-the-badge&labelColor=451a03" alt="Node version"></a>
+</p>
+
+<p align="center">
+  <a href="https://goodandready.app/"><img src="https://img.shields.io/badge/Все_проекты_автора-goodandready.app-ff4500.svg?style=for-the-badge&logo=rocket&logoColor=white&labelColor=1a1a2e" alt="Все проекты автора"></a>
+</p>
+
+<p align="center">
+  <a href="README.md"><b>🇬🇧 English</b></a> •
+  <a href="README.ru.md"><b>🇷🇺 Русский</b></a> •
+  <a href="README.zh.md"><b>🇨🇳 中文说明</b></a>
+</p>
+
+</div>
+
+---
+
+## ⚡ Обзор
+
+**`dsh-model-sync`** обеспечивает автоматическую актуализацию каталога моделей **DeepSeek Harness** напрямую от подключенных провайдеров.
+
+Вместо ручного редактирования YAML-файлов при выходе новых моделей, смене лимитов контекста или цен, плагин автоматически обнаруживает релизы, обновляет флаги возможностей (`vision`, `tools`, `reasoning`, `embeddings`), отслеживает остатки баланса и обновляет каталог на лету без перезапуска сервера.
+
+```mermaid
+graph LR
+    subgraph Trigger [Планировщик и ручной запуск]
+        Cron[⏰ Фоновый планировщик опроса] --> Engine[Ядро dsh-model-sync]
+        WebUI[🖥️ Кнопка «Синхронизировать сейчас»] --> Engine
+    end
+
+    subgraph Providers [25+ Внешних провайдеров]
+        Engine --> Registry{Реестр адаптеров}
+        Registry -->|Авторизация Bearer| P1[OpenAI / DeepSeek / OpenRouter / Groq]
+        Registry -->|Заголовок x-api-key| P2[Anthropic Claude / Кастомные шлюзы]
+        Registry -->|Ключ query-key| P3[Google Gemini]
+        Registry -->|Локальный опрос| P4[Локальная Ollama / vLLM / SGLang]
+    end
+
+    subgraph Reconcile [Сверка каталога и аудит]
+        P1 --> Normalizer[Нормализация и разметка возможностей]
+        P2 --> Normalizer
+        P3 --> Normalizer
+        P4 --> Normalizer
+        Normalizer --> Diff[Журнал изменений: Добавлено / Устарело]
+        Diff --> Catalog[Активный каталог моделей DSH]
+    end
+
+    style Trigger fill:#1e1e2e,stroke:#89b4fa,stroke-width:2px,color:#cdd6f4
+    style Providers fill:#181825,stroke:#cba6f7,stroke-width:2px,color:#cdd6f4
+    style Reconcile fill:#11111b,stroke:#a6e3a1,stroke-width:2px,color:#cdd6f4
+```
+
+---
+
+## ✨ Ключевые возможности
+
+* 🔄 **Автоматическое обнаружение моделей**: синхронизация списков моделей, контекстных окон и возможностей (`vision`, `tools`, `reasoning`, `embeddings`) для 25+ провайдеров.
+* 🌐 **25+ встроенных адаптеров**: готовая поддержка OpenAI, Anthropic, Google, DeepSeek, xAI, OpenRouter, Groq, Mistral, Cerebras, Fireworks, HuggingFace, Moonshot, NVIDIA, Qwen, Together, Xiaomi MiMo, SiliconFlow и локальной Ollama.
+* 🔌 **Универсальные кастомные адаптеры**: подключение любых сторонних OpenAI-совместимых эндпоинтов `/v1/models` (`bearer`, `x-api-key`, `query-key`, `none`).
+* 📊 **Мониторинг баланса и квот**: опрос биллинга провайдеров (где доступно) для контроля остатка кредитов.
+* 📜 **Журнал изменений (Diff Log)**: фиксация добавленных, удаленных и обновленных моделей с отметками времени.
+* 🖥️ **Панель управления в Web GUI (**Настройки → Синхронизация моделей**)**:
+  * Статус-карточки со счетчиками активных моделей по каждому провайдеру;
+  * Кнопка «Синхронизировать сейчас» для мгновенного обновления;
+  * Переключатели провайдеров и ввод пользовательских адресов.
+
+---
+
+## 📦 Быстрая установка
+
+```bash
+dsh plugin --profile web add @goodandready/dsh-model-sync
+```
+
+---
+
+## 🛠️ Улучшения и исправления в v0.3.6
+
+* 🔑 **Универсальное разрешение учетных данных (`credentialRef` и `apiKeyRef`)**:
+  * Добавлена полная поддержка ссылок `credentialRef` и `apiKeyRef` в модулях инвентаря, реестре адаптеров и generic-маршрутах OpenAI-совместимых провайдеров.
+  * Провайдеры, настроенные через DSH Credentials Service, теперь корректно обнаруживаются и аутентифицируются без необходимости обязательного наличия `apiKeyEnv`.
+* ♻️ **Очистка устаревших моделей в жизненном цикле**:
+  * Исправлена ошибка, из-за которой модели в статусе `deprecated` оставались в конфигурации бессрочно после их удаления провайдером. При флаге `removeMissing: true` такие модели теперь гарантированно переводятся в статус `removed` по истечении grace-периода.
+* ⏰ **Точный статус неактивного планировщика**:
+  * Исправлен `scheduler.status()`: при отключённой фоновой синхронизации `nextRunAt` теперь возвращает `null`, исключая вводящие в заблуждение даты следующего запуска в Web UI.
+* 🛡️ **Надежность и масштабируемость Web UI**:
+  * Защита от сбоев сортировки при отсутствии поля `name` у модели.
+  * Замена spread-операций на итеративную агрегацию при вычислении экстремумов контекста и цены, что предотвращает переполнение стека (`RangeError`) на огромных каталогах (10 000+ моделей).
+  * Детальный вывод текста ошибки при тестировании модели по кнопке `▶`.
+  * Локализация срока давности кеша в карточке провайдера.
+* ⚡ **Детерминированное сравнение каталогов и безопасность потоков**:
+  * Реализована стабильная сортировка ключей перед сравнением моделей, исключающая фиктивные diff-записи из-за произвольного порядка ключей в JSON.
+  * Добавлено прерывание сетевого потока при превышении лимита размера тела HTTP-запроса для защиты от утечек памяти.
+
+---
+
+## 🚀 Новые возможности в v0.3.5
+
+* ⏱️ **Адаптивная обработка Rate Limit и Retry-After**:
+  * Автоматический парсинг HTTP-заголовков `Retry-After` (в секундах и формате RFC-дат) и `x-ratelimit-reset`.
+  * При повторах задержка выставляется строго по требованию провайдера.
+  * Ошибки 429 классифицируются в отдельную категорию `rate-limit`.
+* 🔄 **Автоматическое разрешение конфликтов настроек**:
+  * Добавлен цикл повторов для `SETTINGS_CONFLICT` (HTTP 409). При параллельных правках конфигурации DSH плагин автоматически перечитывает ревизию и накладывает изменения без ошибок.
+* 🎯 **Оптимизация панели выбора моделей в Web UI**:
+  * **Токенизированный умный поиск**: Слова поискового запроса (`claude 3.5 sonnet`) сопоставляются независимо по ID, названиям, тегам и описаниям моделей.
+  * **Быстрые фильтр-чипы**: Фильтры `Контекст ≥ 100k` и `Дешевые ≤ $1/1M` в дополнение к фильтрам возможностей (`Зрение`, `Инструменты`, `Рассуждение`, `Эмбеддинги`).
+  * **Массовый выбор**: Кнопки *Выбрать все показанные*, *Снять выбор с показанных*, *Инвертировать выбор*.
+  * **Порционный рендеринг (пагинация)**: Отображение порциями по 50 моделей с кнопкой «Показать ещё» для плавной работы интерфейса в каталогах с 1000+ моделями.
+  * **Бейджи устаревания моделей**: Для моделей с датой вывода из эксплуатации выводится обратный отсчет (`Устареет через N дн.`).
+* 💾 **Хранение и парсинг метаданных**:
+  * Корректный разбор числовых значений с запятыми и суффиксами (`128k`, `1M`, `128,000`).
+  * Автоматическая компактификация: снимки каталогов старше 5 последних запусков очищаются от тяжелых дампов, сохраняя diff-аудит и компактность `settings.yaml`.
+
+---
+
+## ⚙️ Пример конфигурации (`settings.yaml`)
+
+```yaml
+dsh-model-sync:
+  enabled: true
+  syncIntervalMinutes: 60
+  autoReconcile: true
+  providers:
+    openrouter:
+      enabled: true
+      keyEnv: OPENROUTER_API_KEY
+    deepseek:
+      enabled: true
+      keyEnv: DEEPSEEK_API_KEY
+    custom:
+      enabled: false
+      baseURL: http://127.0.0.1:8000/v1
+      authType: bearer
+      keyEnv: CUSTOM_API_KEY
+```
+
+---
+
+## 📄 Лицензия
+
+MIT © [GooDAnDReaDY](https://github.com/GooDAnDReaDY)

package/README.zh.md

@@ -0,0 +1,75 @@
+# 📦 @goodandready/dsh-model-sync
+
+<div align="center">
+
+<h3>DeepSeek Harness 服务商模型目录动态同步与账户余额监控插件</h3>
+
+<p align="center">
+  <a href="https://www.npmjs.com/package/@goodandready/dsh-model-sync"><img src="https://img.shields.io/npm/v/@goodandready/dsh-model-sync.svg?style=for-the-badge&color=6366f1&labelColor=1e1b4b" alt="npm version"></a>
+  <a href="LICENSE"><img src="https://img.shields.io/badge/License-MIT-10b981.svg?style=for-the-badge&color=10b981&labelColor=064e3b" alt="license"></a>
+  <a href="https://github.com/topics/dsh-plugin"><img src="https://img.shields.io/badge/DSH-Plugin-8b5cf6.svg?style=for-the-badge&labelColor=2e1065" alt="DSH Plugin"></a>
+  <a href="https://nodejs.org"><img src="https://img.shields.io/badge/Node-20%2B-f59e0b.svg?style=for-the-badge&labelColor=451a03" alt="Node version"></a>
+</p>
+
+<p align="center">
+  <a href="https://goodandready.app/"><img src="https://img.shields.io/badge/作者全部项目-goodandready.app-ff4500.svg?style=for-the-badge&logo=rocket&logoColor=white&labelColor=1a1a2e" alt="作者全部项目"></a>
+</p>
+
+<p align="center">
+  <a href="README.md"><b>🇬🇧 English</b></a> •
+  <a href="README.ru.md"><b>🇷🇺 Русский</b></a> •
+  <a href="README.zh.md"><b>🇨🇳 中文说明</b></a>
+</p>
+
+</div>
+
+---
+
+## ⚡ 插件概览
+
+**`dsh-model-sync`** 保持 **DeepSeek Harness** 模型选单与上游大模型服务商实时同步。
+
+每当服务商上线新模型、扩充上下文长度或调整定价时，插件自动拉取最新清单，同步模型能力标签（`vision` 视觉、`tools` 工具调用、`reasoning` 深度思考、`embeddings` 嵌入），并在无需重启服务的前提下热重载目录。
+
+```mermaid
+graph LR
+    subgraph Trigger [调度与手动触发]
+        Cron[⏰ 后台定时轮询调度器] --> Engine[dsh-model-sync 核心引擎]
+        WebUI[🖥️ 设置面板: 立即同步按钮] --> Engine
+    end
+
+    subgraph Providers [25+ 上游服务商]
+        Engine --> Registry{适配器注册表}
+        Registry -->|Bearer 鉴权| P1[OpenAI / DeepSeek / OpenRouter / Groq]
+        Registry -->|x-api-key 鉴权| P2[Anthropic Claude / 自定义网关]
+        Registry -->|query-key 鉴权| P3[Google Gemini]
+        Registry -->|本地探针| P4[本地 Ollama / vLLM / SGLang]
+    end
+
+    subgraph Reconcile [目录对齐与审计]
+        P1 --> Normalizer[模型归一化与能力特征打标]
+        P2 --> Normalizer
+        P3 --> Normalizer
+        P4 --> Normalizer
+        Normalizer --> Diff[差异审计流: 新增 / 废弃模型]
+        Diff --> Catalog[DSH 当前生效模型选单]
+    end
+
+    style Trigger fill:#1e1e2e,stroke:#89b4fa,stroke-width:2px,color:#cdd6f4
+    style Providers fill:#181825,stroke:#cba6f7,stroke-width:2px,color:#cdd6f4
+    style Reconcile fill:#11111b,stroke:#a6e3a1,stroke-width:2px,color:#cdd6f4
+```
+
+---
+
+## 📦 安装指南
+
+```bash
+dsh plugin --profile web add @goodandready/dsh-model-sync
+```
+
+---
+
+## 📄 开源协议
+
+MIT © [GooDAnDReaDY](https://github.com/GooDAnDReaDY)

package/lib/adapter-registry.js

@@ -121,7 +121,8 @@ export function normalizeAdapterConfig(config, profile = {}) {
 }
 
 async function requestSpecific(profile, descriptor, { resolveCredential, fetchImpl, signal }) {
-  const apiKey = profile.apiKeyEnv ? await resolveCredential(profile.apiKeyEnv) : ''
+  const keyRef = profile.apiKeyEnv || profile.credentialRef || profile.apiKeyRef
+  const apiKey = keyRef ? await resolveCredential(keyRef) : ''
   const url = new URL(descriptor.endpoint)
   const headers = { Accept: 'application/json', ...(descriptor.headers ?? {}) }
   if (descriptor.auth === 'bearer' && apiKey) headers.Authorization = 'Bearer ' + apiKey
@@ -133,8 +134,8 @@ async function requestSpecific(profile, descriptor, { resolveCredential, fetchIm
     throw error
   }
   if (!response.ok) {
-    const error = catalogRequestError(response.status)
-    error.adapterCode = [401, 403].includes(response.status) ? 'auth' : 'http'
+    const error = catalogRequestError(response.status, response)
+    error.adapterCode = [401, 403].includes(response.status) ? 'auth' : response.status === 429 ? 'rate-limit' : 'http'
     throw error
   }
   return response
@@ -144,6 +145,12 @@ async function fetchSpecific(profile, descriptor, options) {
   let payload
   try {
     const response = await requestSpecific(profile, descriptor, options)
+    const declared = Number(response.headers?.get?.('content-length') ?? NaN)
+    if (Number.isFinite(declared) && declared > MAX_RESPONSE_BYTES) {
+      const error = new Error('model catalog response is too large')
+      error.adapterCode = 'schema'
+      throw error
+    }
     const text = await response.text()
     if (Buffer.byteLength(text, 'utf8') > MAX_RESPONSE_BYTES) {
       const error = new Error('model catalog response is too large')

package/lib/client.js

@@ -41,10 +41,10 @@ window.__ModuleLoader__.load({
       document.head.appendChild(tag)
     }
     const en = {
-      title: 'Model synchronization', showSettings: 'Show settings', hideSettings: 'Hide settings', desc: 'Refresh model catalogs for API-key providers. A dry-run is the default; applying changes is explicit.', loading: 'Loading…', refresh: 'Refresh status', refreshAll: 'Refresh all', run: 'Discover', apply: 'Apply changes', dryRun: 'Preview only (dry-run)', all: 'All API-key providers', ok: 'ready', dormant: 'not configured', running: 'running…', noProviders: 'No API-key providers found.', lastRun: 'Last run', previewTitle: 'Changes preview', previewNone: 'No changes detected.', previewAdded: 'Added', previewRemoved: 'Removed', previewChanged: 'Changed', previewApply: 'Apply preview', previewDismiss: 'Close preview', selectModels: 'Choose models', saveSelection: 'Save and apply selection', selectAtLeastOne: 'Select at least one model or use All models.', cancel: 'Cancel', selected: 'selected', allModels: 'All models', noModels: 'No models in the current catalog.', capabilityFilter: 'Filter by capabilities', capabilityCount: 'shown', noCapabilities: 'No capability metadata in this catalog.', pickerNote: 'The selection is applied to the DSH model picker immediately.', health: 'Check availability', healthOk: 'available', healthError: 'unavailable', healthDormant: 'not configured', policy: 'Manual model selection', savePolicy: 'Save policy', policyInclude: 'Include patterns (one per line)', policyExclude: 'Exclude patterns (one per line)', policyRequire: 'Required capabilities (comma-separated)', policyDeny: 'Denied capabilities (comma-separated)', policyNote: 'Patterns match model id, name, and tags. Apply changes to update the DSH picker.', scheduler: 'Scheduler', nextRun: 'next run', schedulerOff: 'off', schedulerOn: 'on', removeMissing: 'Confirm stale removal', stale: 'stale', deprecated: 'deprecated', history: 'Synchronization history', historyRun: 'Run', diffAdded: 'added', diffRemoved: 'removed', diffRenamed: 'renamed', diffChanged: 'metadata changed', rollback: 'Rollback catalog', rollbackConfirm: 'Rollback this provider catalog to the selected snapshot?', credentials: 'Credential diagnostics', checkCredentials: 'Check credentials', missingCredential: 'not configured', credentialReady: 'configured', report: 'Run report', success: 'success', partial: 'partial success', failure: 'failed', empty: 'no configured providers', notifications: 'Notifications', markRead: 'Mark read', acknowledge: 'Acknowledge', noNotifications: 'No notifications.',
+      title: 'Model synchronization', showSettings: 'Show settings', hideSettings: 'Hide settings', desc: 'Refresh model catalogs for API-key providers. A dry-run is the default; applying changes is explicit.', loading: 'Loading…', refresh: 'Refresh status', refreshAll: 'Refresh all', run: 'Discover', apply: 'Apply changes', dryRun: 'Preview only (dry-run)', all: 'All API-key providers', ok: 'ready', dormant: 'not configured', running: 'running…', noProviders: 'No API-key providers found.', lastRun: 'Last run', previewTitle: 'Changes preview', previewNone: 'No changes detected.', previewAdded: 'Added', previewRemoved: 'Removed', previewChanged: 'Changed', previewApply: 'Apply preview', previewDismiss: 'Close preview', selectModels: 'Choose models', saveSelection: 'Save and apply selection', selectAtLeastOne: 'Select at least one model or use All models.', cancel: 'Cancel', selected: 'selected', allModels: 'All models', noModels: 'No models in the current catalog.', capabilityFilter: 'Filter by capabilities', capabilityCount: 'shown', noCapabilities: 'No capability metadata in this catalog.', pickerNote: 'The selection is applied to the DSH model picker immediately.', health: 'Check availability', healthOk: 'available', healthError: 'unavailable', healthDormant: 'not configured', policy: 'Manual model selection', savePolicy: 'Save policy', policyInclude: 'Include patterns (one per line)', policyExclude: 'Exclude patterns (one per line)', policyRequire: 'Required capabilities (comma-separated)', policyDeny: 'Denied capabilities (comma-separated)', policyNote: 'Patterns match model id, name, and tags. Apply changes to update the DSH picker.', scheduler: 'Scheduler', nextRun: 'next run', schedulerOff: 'off', schedulerOn: 'on', removeMissing: 'Confirm stale removal', stale: 'stale', deprecated: 'deprecated', history: 'Synchronization history', historyRun: 'Run', diffAdded: 'added', diffRemoved: 'removed', diffRenamed: 'renamed', diffChanged: 'metadata changed', rollback: 'Rollback catalog', rollbackConfirm: 'Rollback this provider catalog to the selected snapshot?', credentials: 'Credential diagnostics', checkCredentials: 'Check credentials', missingCredential: 'not configured', credentialReady: 'configured', report: 'Run report', success: 'success', partial: 'partial success', failure: 'failed', empty: 'no configured providers', notifications: 'Notifications', markRead: 'Mark read', acknowledge: 'Acknowledge', noNotifications: 'No notifications.', selectAllFiltered: 'Select all shown', deselectAllFiltered: 'Deselect all shown', invertFiltered: 'Invert selection', filterContext100k: 'Context ≥ 100k', filterCheap: 'Cheap ≤ $1/1M', showMore: 'Show more (%d)', showingCount: 'Showing %d of %d models', deprecatesIn: 'Deprecates in %dd', cacheHoursAgo: 'cache %dh ago',
     }
     const ru = {
-      title: 'Синхронизация моделей', showSettings: 'Показать настройки', hideSettings: 'Скрыть настройки', desc: 'Обновляет каталоги моделей у провайдеров с API-ключом. По умолчанию только dry-run; применение запускается явно.', loading: 'Загрузка…', refresh: 'Обновить статус', refreshAll: 'Обновить все', run: 'Проверить', apply: 'Применить изменения', dryRun: 'Только просмотр (dry-run)', all: 'Все API-key провайдеры', ok: 'готов', dormant: 'не настроен', running: 'выполняется…', noProviders: 'API-key провайдеры не найдены.', lastRun: 'Последний запуск', previewTitle: 'Предпросмотр изменений', previewNone: 'Изменений не обнаружено.', previewAdded: 'Добавлено', previewRemoved: 'Удалено', previewChanged: 'Изменено', previewApply: 'Применить изменения', previewDismiss: 'Закрыть предпросмотр', selectModels: 'Выбрать модели', saveSelection: 'Сохранить и применить выбор', selectAtLeastOne: 'Выберите хотя бы одну модель или нажмите «Все модели».', cancel: 'Отмена', selected: 'выбрано', allModels: 'Все модели', noModels: 'В текущем каталоге нет моделей.', capabilityFilter: 'Фильтр по возможностям', capabilityCount: 'показано', noCapabilities: 'В этом каталоге нет данных о возможностях.', pickerNote: 'Выбор сразу применяется к пикеру моделей DSH.', health: 'Проверить доступность', healthOk: 'доступен', healthError: 'недоступен', healthDormant: 'не настроен', policy: 'Ручной выбор моделей', savePolicy: 'Сохранить политику', policyInclude: 'Включать (по одному шаблону в строке)', policyExclude: 'Исключать (по одному шаблону в строке)', policyRequire: 'Обязательные возможности (через запятую)', policyDeny: 'Запрещённые возможности (через запятую)', policyNote: 'Шаблоны ищутся в id, названии и тегах модели. Нажмите «Применить», чтобы обновить пикер DSH.', scheduler: 'Планировщик', nextRun: 'следующий запуск', schedulerOff: 'выключен', schedulerOn: 'включён', removeMissing: 'Подтвердить удаление устаревших', stale: 'устарела', deprecated: 'deprecated', history: 'История синхронизаций', historyRun: 'Запуск', diffAdded: 'добавлено', diffRemoved: 'удалено', diffRenamed: 'переименовано', diffChanged: 'изменена метаинформация', rollback: 'Откатить каталог', rollbackConfirm: 'Откатить каталог этого провайдера к выбранному снимку?', credentials: 'Диагностика ключей', checkCredentials: 'Проверить ключи', missingCredential: 'не настроен', credentialReady: 'настроен', report: 'Итог запуска', success: 'успех', partial: 'частичный успех', failure: 'ошибка', empty: 'нет настроенных провайдеров', notifications: 'Уведомления', markRead: 'Прочитано', acknowledge: 'Подтвердить', noNotifications: 'Новых уведомлений нет.',
+      title: 'Синхронизация моделей', showSettings: 'Показать настройки', hideSettings: 'Скрыть настройки', desc: 'Обновляет каталоги моделей у провайдеров с API-ключом. По умолчанию только dry-run; применение запускается явно.', loading: 'Загрузка…', refresh: 'Обновить статус', refreshAll: 'Обновить все', run: 'Проверить', apply: 'Применить изменения', dryRun: 'Только просмотр (dry-run)', all: 'Все API-key провайдеры', ok: 'готов', dormant: 'не настроен', running: 'выполняется…', noProviders: 'API-key провайдеры не найдены.', lastRun: 'Последний запуск', previewTitle: 'Предпросмотр изменений', previewNone: 'Изменений не обнаружено.', previewAdded: 'Добавлено', previewRemoved: 'Удалено', previewChanged: 'Изменено', previewApply: 'Применить изменения', previewDismiss: 'Закрыть предпросмотр', selectModels: 'Выбрать модели', saveSelection: 'Сохранить и применить выбор', selectAtLeastOne: 'Выберите хотя бы одну модель или нажмите «Все модели».', cancel: 'Отмена', selected: 'выбрано', allModels: 'Все модели', noModels: 'В текущем каталоге нет моделей.', capabilityFilter: 'Фильтр по возможностям', capabilityCount: 'показано', noCapabilities: 'В этом каталоге нет данных о возможностях.', pickerNote: 'Выбор сразу применяется к пикеру моделей DSH.', health: 'Проверить доступность', healthOk: 'доступен', healthError: 'недоступен', healthDormant: 'не настроен', policy: 'Ручной выбор моделей', savePolicy: 'Сохранить политику', policyInclude: 'Включать (по одному шаблону в строке)', policyExclude: 'Исключать (по одному шаблону в строке)', policyRequire: 'Обязательные возможности (через запятую)', policyDeny: 'Запрещённые возможности (через запятую)', policyNote: 'Шаблоны ищутся в id, названии и тегах модели. Нажмите «Применить», чтобы обновить пикер DSH.', scheduler: 'Планировщик', nextRun: 'следующий запуск', schedulerOff: 'выключен', schedulerOn: 'включён', removeMissing: 'Подтвердить удаление устаревших', stale: 'устарела', deprecated: 'deprecated', history: 'История синхронизаций', historyRun: 'Запуск', diffAdded: 'добавлено', diffRemoved: 'удалено', diffRenamed: 'переименовано', diffChanged: 'изменена метаинформация', rollback: 'Откатить каталог', rollbackConfirm: 'Откатить каталог этого провайдера к выбранному снимку?', credentials: 'Диагностика ключей', checkCredentials: 'Проверить ключи', missingCredential: 'не настроен', credentialReady: 'настроен', report: 'Итог запуска', success: 'успех', partial: 'частичный успех', failure: 'ошибка', empty: 'нет настроенных провайдеров', notifications: 'Уведомления', markRead: 'Прочитано', acknowledge: 'Подтвердить', noNotifications: 'Новых уведомлений нет.', selectAllFiltered: 'Выбрать все показанные', deselectAllFiltered: 'Снять выбор с показанных', invertFiltered: 'Инвертировать выбор', filterContext100k: 'Контекст ≥ 100k', filterCheap: 'Дешевые ≤ $1/1M', showMore: 'Показать ещё (%d)', showingCount: 'Показано %d из %d моделей', deprecatesIn: 'Устареет через %d дн.', cacheHoursAgo: 'кеш %dч назад',
     }
     const API = '/dsh-model-sync'
     const CAPABILITY_LABELS = { vision: { en: 'Vision', ru: 'Зрение' }, tools: { en: 'Tools', ru: 'Инструменты' }, reasoning: { en: 'Reasoning', ru: 'Рассуждение' }, embeddings: { en: 'Embeddings', ru: 'Эмбеддинги' } }
@@ -58,8 +58,9 @@ window.__ModuleLoader__.load({
       const locale = useLocale(ctx); const t = locale === 'ru' ? ru : en
       const [status, setStatus] = React.useState(null); const [provider, setProvider] = React.useState(''); const [dryRun, setDryRun] = React.useState(true); const [removeMissing, setRemoveMissing] = React.useState(false); const [busy, setBusy] = React.useState(false); const [error, setError] = React.useState(''); const [credentialStatus, setCredentialStatus] = React.useState(null)
       const [selectionProvider, setSelectionProvider] = React.useState(''); const [selectionDraft, setSelectionDraft] = React.useState([]); const [capabilityFilter, setCapabilityFilter] = React.useState([]); const [searchQuery, setSearchQuery] = React.useState(''); const [sortBy, setSortBy] = React.useState('name'); const [preview, setPreview] = React.useState(null); const [policyProvider, setPolicyProvider] = React.useState(''); const [policyDraft, setPolicyDraft] = React.useState({ include: '', exclude: '', requireCapabilities: '', denyCapabilities: '' })
+      const [pageSize, setPageSize] = React.useState(50); const [quickFilter, setQuickFilter] = React.useState({ context100k: false, cheap: false })
       const [tryState, setTryState] = React.useState({})
-      const doTry = (provider, modelId) => { setTryState((s)=>({ ...s, [provider+':'+modelId]: { loading: true } })); fetch(`${API}/try`, { method: 'POST', headers: { 'content-type': 'application/json' }, body: JSON.stringify({ provider, model: modelId }) }).then(async r=>{ const b=await r.json(); if(!r.ok) throw new Error(b?.error?.message||`HTTP ${r.status}`); return b }).then(b=>setTryState((s)=>({ ...s, [provider+':'+modelId]: { ok: b.ok, latency: b.latencyMs, status: b.status } }))).catch(e=>setTryState((s)=>({ ...s, [provider+':'+modelId]: { error: String(e.message||e) } }))) }
+      const doTry = (provider, modelId) => { setTryState((s)=>({ ...s, [provider+':'+modelId]: { loading: true } })); fetch(`${API}/try`, { method: 'POST', headers: { 'content-type': 'application/json' }, body: JSON.stringify({ provider, model: modelId }) }).then(async r=>{ const b=await r.json(); if(!r.ok) throw new Error(b?.error?.message||`HTTP ${r.status}`); return b }).then(b=>setTryState((s)=>({ ...s, [provider+':'+modelId]: { ok: b.ok, latency: b.latencyMs, status: b.status, error: b.error } }))).catch(e=>setTryState((s)=>({ ...s, [provider+':'+modelId]: { error: String(e.message||e) } }))) }
       const load = React.useCallback(() => { setError(''); return fetch(`${API}/status`).then((r) => r.json()).then(setStatus).catch((e) => setError(String(e?.message ?? e))) }, [])
       const loadCredentials = React.useCallback(() => { setError(''); return fetch(`${API}/credentials`).then(async (r) => { const body = await r.json(); if (!r.ok) throw new Error(body?.error?.message ?? `HTTP ${r.status}`); return body }).then(setCredentialStatus).catch((e) => setError(String(e?.message ?? e))) }, [])
       React.useEffect(() => { ensureStyles(); load(); loadCredentials(); const timer = setInterval(load, 15000); return () => clearInterval(timer) }, [load, loadCredentials])
@@ -67,8 +68,8 @@ window.__ModuleLoader__.load({
       const checkHealth = (targetProvider) => { setBusy(true); setError(''); fetch(`${API}/health`, { method: 'POST', headers: { 'content-type': 'application/json' }, body: JSON.stringify(targetProvider ? { provider: targetProvider } : {}) }).then(async (r) => { const body = await r.json(); if (!r.ok) throw new Error(body?.error?.message ?? `HTTP ${r.status}`); return body }).then(setStatus).catch((e) => setError(String(e?.message ?? e))).finally(() => setBusy(false)) }
       const checkCredentials = (targetProvider) => { setBusy(true); setError(''); fetch(`${API}/credentials/check`, { method: 'POST', headers: { 'content-type': 'application/json' }, body: JSON.stringify(targetProvider ? { provider: targetProvider } : {}) }).then(async (r) => { const body = await r.json(); if (!r.ok) throw new Error(body?.error?.message ?? `HTTP ${r.status}`); return body }).then((body) => { setCredentialStatus(body.credentials); setStatus((current) => ({ ...(current ?? {}), lastHealth: body.health })) }).catch((e) => setError(String(e?.message ?? e))).finally(() => setBusy(false)) }
       const updateNotification = (id, action) => { setBusy(true); setError(''); fetch(API + '/notifications/' + action, { method: 'POST', headers: { 'content-type': 'application/json' }, body: JSON.stringify({ id }) }).then(async (r) => { const body = await r.json(); if (!r.ok) throw new Error(body?.error?.message ?? ('HTTP ' + r.status)); return body }).then(() => load()).catch((e) => setError(String(e?.message ?? e))).finally(() => setBusy(false)) }
-      const openSelection = (row) => { const available = row.availableModels ?? row.models ?? []; const defaults = row.selectableModels ?? available; closePolicy(); setSelectionProvider(row.provider); setSelectionDraft(row.selectedModels?.length ? row.selectedModels : defaults.map((model) => model.id)); setCapabilityFilter([]); setSearchQuery(''); setSortBy('name') }
-      const closeSelection = () => { setSelectionProvider(''); setSelectionDraft([]); setCapabilityFilter([]); setSearchQuery(''); setSortBy('name') }
+      const openSelection = (row) => { const available = row.availableModels ?? row.models ?? []; const defaults = row.selectableModels ?? available; closePolicy(); setSelectionProvider(row.provider); setSelectionDraft(row.selectedModels?.length ? row.selectedModels : defaults.map((model) => model.id)); setCapabilityFilter([]); setSearchQuery(''); setSortBy('name'); setPageSize(50); setQuickFilter({ context100k: false, cheap: false }) }
+      const closeSelection = () => { setSelectionProvider(''); setSelectionDraft([]); setCapabilityFilter([]); setSearchQuery(''); setSortBy('name'); setPageSize(50); setQuickFilter({ context100k: false, cheap: false }) }
       const openPolicy = (row) => { const policy = status?.policies?.[row.provider] ?? row.policy ?? {}; closeSelection(); setPolicyProvider(row.provider); setPolicyDraft({ include: (policy.include ?? []).join('\n'), exclude: (policy.exclude ?? []).join('\n'), requireCapabilities: Object.keys(policy.requireCapabilities ?? {}).join(', '), denyCapabilities: Object.keys(policy.denyCapabilities ?? {}).join(', ') }) }
       const closePolicy = () => { setPolicyProvider(''); setPolicyDraft({ include: '', exclude: '', requireCapabilities: '', denyCapabilities: '' }) }
       const splitValues = (value) => [...new Set(String(value ?? '').split(/[,\n]+/).map((item) => item.trim()).filter(Boolean))]
@@ -78,8 +79,10 @@ window.__ModuleLoader__.load({
       const rollback = (entry, targetProvider) => { if (!window.confirm(t.rollbackConfirm)) return; setBusy(true); setError(''); fetch(`${API}/history/rollback`, { method: 'POST', headers: { 'content-type': 'application/json' }, body: JSON.stringify({ historyId: entry.id, provider: targetProvider }) }).then(async (r) => { const body = await r.json(); if (!r.ok) throw new Error(body?.error?.message ?? `HTTP ${r.status}`); return body }).then(() => load()).catch((e) => setError(String(e?.message ?? e))).finally(() => setBusy(false)) }
       const isNewModel = (model, row) => { const added = status?.lastRun?.results?.find(r=>r.provider===row.provider)?.diff?.added||[]; const isAdded = added.some(m=>m.id===model.id); if(!isAdded) return false; const finishedAt = status?.lastRun?.finishedAt; if(!finishedAt) return false; return (Date.now()-new Date(finishedAt).getTime()) < 7*24*60*60*1000; };
       const rows = (status?.providers ?? []).filter((row) => row.configured); const selectedRow = rows.find((row) => row.provider === selectionProvider); const choices = selectedRow?.availableModels ?? selectedRow?.models ?? []; const filteredChoicesBase = filterModelsByCapabilities(choices, capabilityFilter);
-      const searchedChoices = searchQuery ? filteredChoicesBase.filter((m) => { const q = searchQuery.toLowerCase(); return m.id.toLowerCase().includes(q) || (m.name||'').toLowerCase().includes(q) || (m.aliases||[]).some(a=>a.toLowerCase().includes(q)) || (m.tags||[]).some(t=>t.toLowerCase().includes(q)) || (m.description||'').toLowerCase().includes(q); }) : filteredChoicesBase;
-      const filteredChoices = [...searchedChoices].sort((a,b) => { if (sortBy==='price') { const pa=a.pricing?.inputPerToken??Infinity; const pb=b.pricing?.inputPerToken??Infinity; return pa-pb; } if (sortBy==='context') return (b.contextWindow??0)-(a.contextWindow??0); if (sortBy==='date') return (b.firstSeenAt??0)-(a.firstSeenAt??0); return a.name.localeCompare(b.name); }); const availableCapabilities = [...new Set(choices.flatMap((model) => Object.entries(model?.capabilities ?? {}).filter(([, enabled]) => enabled === true).map(([key]) => key)))].sort(); const button = (label, onClick) => React.createElement('button', { type: 'button', disabled: busy, onClick, className: 'dms-btn' }, label)
+      const tokens = searchQuery.trim().toLowerCase().split(/\s+/).filter(Boolean);
+      const searchedChoices = tokens.length ? filteredChoicesBase.filter((m) => { const haystack = (m.id + ' ' + (m.name||'') + ' ' + (m.aliases||[]).join(' ') + ' ' + (m.tags||[]).join(' ') + ' ' + (m.description||'')).toLowerCase(); return tokens.every((tok) => haystack.includes(tok)); }) : filteredChoicesBase;
+      const quickFilteredChoices = searchedChoices.filter((m) => { if (quickFilter.context100k && (!m.contextWindow || m.contextWindow < 100000)) return false; if (quickFilter.cheap && (!m.pricing?.inputPerToken || (m.pricing.inputPerToken * 1000000) > 1.0)) return false; return true; });
+      const filteredChoices = [...quickFilteredChoices].sort((a,b) => { if (sortBy==='price') { const pa=a.pricing?.inputPerToken??Infinity; const pb=b.pricing?.inputPerToken??Infinity; return pa-pb; } if (sortBy==='context') return (b.contextWindow??0)-(a.contextWindow??0); if (sortBy==='date') return (b.firstSeenAt??0)-(a.firstSeenAt??0); return (a.name || a.id || '').localeCompare(b.name || b.id || ''); }); const maxContext = filteredChoices.reduce((max, m) => Math.max(max, m.contextWindow || 0), 0); const minPrice = filteredChoices.reduce((min, m) => m.pricing?.inputPerToken !== undefined ? Math.min(min, m.pricing.inputPerToken) : min, Infinity); const availableCapabilities = [...new Set(choices.flatMap((model) => Object.entries(model?.capabilities ?? {}).filter(([, enabled]) => enabled === true).map(([key]) => key)))].sort(); const visibleChoices = filteredChoices.slice(0, pageSize); const button = (label, onClick) => React.createElement('button', { type: 'button', disabled: busy, onClick, className: 'dms-btn' }, label)
       const renderNotifications = () => {
         const entries = status?.notifications ?? []
         if (entries.length === 0) return React.createElement('small', { style: { display: 'block', marginTop: 6 } }, t.noNotifications)
@@ -121,9 +124,9 @@ window.__ModuleLoader__.load({
           const healthEntry = status?.lastHealth?.results?.find((item) => item.provider === row.provider)
           const latencyText = healthEntry?.latencyMs ? ` · ${healthEntry.latencyMs}ms` : ''
           const isOffline = latestRun?.status !== 'ok' && modelCount > 0
-          const offlineText = isOffline ? ` · кеш ${Math.max(0, Math.round((Date.now() - (status?.lastRun?.finishedAt ? new Date(status.lastRun.finishedAt).getTime() : Date.now()))/3600000))}ч назад` : ''
+          const offlineText = isOffline ? ' · ' + t.cacheHoursAgo.replace('%d', Math.max(0, Math.round((Date.now() - (status?.lastRun?.finishedAt ? new Date(status.lastRun.finishedAt).getTime() : Date.now()))/3600000))) : ''
           const providerCard = React.createElement('div', { className: 'dms-provider' }, React.createElement('span', null, `${row.provider}: ${row.configured ? t.ok : t.dormant} · ${modelSummary}${lifecycleText}${healthText}${latencyText}${offlineText}`), button(t.selectModels, () => openSelection(row)), button(t.policy, () => openPolicy(row)))
-          const selectionEditor = selectionProvider === row.provider ? React.createElement('div', { className: 'dms-editor' }, React.createElement('strong', null, `${t.selectModels}: ${selectionProvider}`), React.createElement('div', { style: { display: 'flex', gap: 8, margin: '8px 0', flexWrap: 'wrap' } }, React.createElement('input', { placeholder: 'Поиск id/name/tag', value: searchQuery, onChange: (e)=>setSearchQuery(e.target.value), style: { flex: '1 1 160px', padding: '6px 8px', borderRadius: 6, border: '1px solid var(--dsw-alias-border-l2)', background: 'var(--dsw-alias-bg-layer-2)' } }), React.createElement('select', { value: sortBy, onChange: (e)=>setSortBy(e.target.value), style: { padding: '6px 8px', borderRadius: 6 } }, React.createElement('option', {value:'name'}, 'Имя A→Z'), React.createElement('option', {value:'price'}, 'Цена ↑'), React.createElement('option', {value:'context'}, 'Контекст ↓'), React.createElement('option', {value:'date'}, 'Новые ↓'))), React.createElement('div', { style: { margin: '8px 0' } }, choices.length === 0 ? React.createElement('small', null, t.noModels) : React.createElement(React.Fragment, null, availableCapabilities.length > 0 ? React.createElement('div', { style: { marginBottom: 8 } }, React.createElement('strong', null, t.capabilityFilter), availableCapabilities.map((capability) => React.createElement('label', { key: capability, style: { display: 'inline-block', marginLeft: 8 } }, React.createElement('input', { type: 'checkbox', checked: capabilityFilter.includes(capability), onChange: (e) => setCapabilityFilter((current) => e.target.checked ? [...current, capability] : current.filter((item) => item !== capability)) }), ' ' + (CAPABILITY_LABELS[capability]?.[locale] ?? capability)))) : React.createElement('small', { style: { display: 'block', marginBottom: 8 } }, t.noCapabilities), React.createElement('small', { style: { display: 'block', marginBottom: 8 } }, filteredChoices.length + '/' + choices.length + ' ' + t.capabilityCount), filteredChoices.map((model) => React.createElement('label', { key: model.id, style: { display: 'block', margin: '4px 0', padding: '4px 6px', borderRadius: 6, background: isNewModel(model,row) ? 'var(--dsw-alias-bg-layer-2)' : 'transparent' } }, React.createElement('input', { type: 'checkbox', checked: selectionDraft.includes(model.id), onChange: (e) => setSelectionDraft((current) => e.target.checked ? [...current, model.id] : current.filter((id) => id !== model.id)) }), ` ${model.name ?? model.id} (${model.id})`, isNewModel(model,row) ? React.createElement('span', { style: { marginLeft: 6, fontSize: 11, padding: '1px 5px', borderRadius: 4, background: 'var(--dsw-alias-state-business-primary)', color: 'white' } }, 'NEW') : null, (model.contextWindow && model.contextWindow >= Math.max(...filteredChoices.map(m=>m.contextWindow||0))) ? React.createElement('span', { style: { marginLeft: 4, fontSize: 10, padding: '1px 4px', borderRadius: 3, background: 'var(--dsw-alias-state-warning-primary)', color: 'white' } }, '★') : null, (model.pricing && model.pricing.inputPerToken === Math.min(...filteredChoices.filter(m=>m.pricing?.inputPerToken!==undefined).map(m=>m.pricing.inputPerToken))) ? React.createElement('span', { style: { marginLeft: 4, fontSize: 10, padding: '1px 4px', borderRadius: 3, background: 'var(--dsw-alias-state-success-primary)', color: 'white' } }, '$') : null, model.tags && model.tags.length ? React.createElement('small', { style: { marginLeft: 6, color: 'var(--dsw-alias-label-secondary)' } }, model.tags.join(', ')) : null, model.pricing ? React.createElement('small', { style: { marginLeft: 6, color: 'var(--dsw-alias-label-secondary)', fontSize: 11 } }, `in $${(model.pricing.inputPerToken*1000000).toFixed(2)}/1M` + (model.pricing.cacheReadPerToken!==undefined ? ` / cache $${(model.pricing.cacheReadPerToken*1000000).toFixed(2)}` : '') + (model.pricing.outputPerToken!==undefined ? ` / out $${(model.pricing.outputPerToken*1000000).toFixed(2)}` : '')) : null, model.description ? React.createElement('small', { title: model.description, style: { display: 'block', marginLeft: 22, color: 'var(--dsw-alias-label-secondary)', fontSize: 11 } }, model.description.slice(0,120)) : null, row.baseURL || row.api ? React.createElement('a', { href: (row.baseURL||'') + '/models/'+encodeURIComponent(model.id), target: '_blank', rel: 'noopener', style: { marginLeft: 8, fontSize: 11 } }, 'docs') : null, React.createElement('button', { type: 'button', onClick: (e)=>{e.preventDefault(); doTry(row.provider, model.id);}, style: { marginLeft: 8, fontSize: 11, padding: '1px 6px', borderRadius: 4 } }, (tryState[row.provider+':'+model.id] && tryState[row.provider+':'+model.id].loading) ? '…' : '▶'), tryState[row.provider+':'+model.id] && !tryState[row.provider+':'+model.id].loading ? React.createElement('small', { style: { marginLeft: 4, color: tryState[row.provider+':'+model.id].ok ? 'var(--dsw-alias-state-success-primary)' : 'var(--dsw-alias-state-error-primary)', fontSize: 11 } }, tryState[row.provider+':'+model.id].ok ? `${tryState[row.provider+':'+model.id].latency}ms` : (tryState[row.provider+':'+model.id].error||tryState[row.provider+':'+model.id].status)) : null)))), React.createElement('small', null, t.pickerNote), React.createElement('div', { style: { marginTop: 8 } }, button(t.allModels, () => setSelectionDraft(choices.map((model) => model.id))), button(t.saveSelection, saveSelection), button(t.cancel, closeSelection))) : null
+          const selectionEditor = selectionProvider === row.provider ? React.createElement('div', { className: 'dms-editor' }, React.createElement('strong', null, `${t.selectModels}: ${selectionProvider}`), React.createElement('div', { style: { display: 'flex', gap: 8, margin: '8px 0', flexWrap: 'wrap' } }, React.createElement('input', { placeholder: 'Поиск id/name/tag', value: searchQuery, onChange: (e)=>setSearchQuery(e.target.value), style: { flex: '1 1 160px', padding: '6px 8px', borderRadius: 6, border: '1px solid var(--dsw-alias-border-l2)', background: 'var(--dsw-alias-bg-layer-2)' } }), React.createElement('select', { value: sortBy, onChange: (e)=>setSortBy(e.target.value), style: { padding: '6px 8px', borderRadius: 6 } }, React.createElement('option', {value:'name'}, 'Имя A→Z'), React.createElement('option', {value:'price'}, 'Цена ↑'), React.createElement('option', {value:'context'}, 'Контекст ↓'), React.createElement('option', {value:'date'}, 'Новые ↓'))), React.createElement('div', { style: { display: 'flex', gap: 6, margin: '6px 0', flexWrap: 'wrap' } }, React.createElement('button', { type: 'button', className: 'dms-btn', style: quickFilter.context100k ? { borderColor: 'var(--dsw-alias-brand-primary)', background: 'var(--dsw-alias-bg-layer-2)' } : {}, onClick: () => setQuickFilter((f) => ({ ...f, context100k: !f.context100k })) }, (quickFilter.context100k ? '✓ ' : '') + t.filterContext100k), React.createElement('button', { type: 'button', className: 'dms-btn', style: quickFilter.cheap ? { borderColor: 'var(--dsw-alias-brand-primary)', background: 'var(--dsw-alias-bg-layer-2)' } : {}, onClick: () => setQuickFilter((f) => ({ ...f, cheap: !f.cheap })) }, (quickFilter.cheap ? '✓ ' : '') + t.filterCheap)), React.createElement('div', { style: { margin: '8px 0' } }, choices.length === 0 ? React.createElement('small', null, t.noModels) : React.createElement(React.Fragment, null, availableCapabilities.length > 0 ? React.createElement('div', { style: { marginBottom: 8 } }, React.createElement('strong', null, t.capabilityFilter), availableCapabilities.map((capability) => React.createElement('label', { key: capability, style: { display: 'inline-block', marginLeft: 8 } }, React.createElement('input', { type: 'checkbox', checked: capabilityFilter.includes(capability), onChange: (e) => setCapabilityFilter((current) => e.target.checked ? [...current, capability] : current.filter((item) => item !== capability)) }), ' ' + (CAPABILITY_LABELS[capability]?.[locale] ?? capability)))) : React.createElement('small', { style: { display: 'block', marginBottom: 8 } }, t.noCapabilities), React.createElement('div', { style: { display: 'flex', gap: 6, margin: '6px 0', flexWrap: 'wrap' } }, button(t.selectAllFiltered, () => { const ids = new Set(selectionDraft); for (const m of filteredChoices) ids.add(m.id); setSelectionDraft([...ids]); }), button(t.deselectAllFiltered, () => { const filterSet = new Set(filteredChoices.map((m) => m.id)); setSelectionDraft(selectionDraft.filter((id) => !filterSet.has(id))); }), button(t.invertFiltered, () => { const current = new Set(selectionDraft); for (const m of filteredChoices) { if (current.has(m.id)) current.delete(m.id); else current.add(m.id); } setSelectionDraft([...current]); })), React.createElement('small', { style: { display: 'block', marginBottom: 8 } }, filteredChoices.length + '/' + choices.length + ' ' + t.capabilityCount), visibleChoices.map((model) => { const deprecationDays = (() => { const d = model.lifecycle?.deprecationDate; if (!d) return null; const diff = new Date(d).getTime() - Date.now(); return diff > 0 ? Math.ceil(diff / (1000 * 60 * 60 * 24)) : 0; })(); return React.createElement('label', { key: model.id, style: { display: 'block', margin: '4px 0', padding: '4px 6px', borderRadius: 6, background: isNewModel(model,row) ? 'var(--dsw-alias-bg-layer-2)' : 'transparent' } }, React.createElement('input', { type: 'checkbox', checked: selectionDraft.includes(model.id), onChange: (e) => setSelectionDraft((current) => e.target.checked ? [...current, model.id] : current.filter((id) => id !== model.id)) }), ` ${model.name ?? model.id} (${model.id})`, isNewModel(model,row) ? React.createElement('span', { style: { marginLeft: 6, fontSize: 11, padding: '1px 5px', borderRadius: 4, background: 'var(--dsw-alias-state-business-primary)', color: 'white' } }, 'NEW') : null, deprecationDays !== null ? React.createElement('span', { style: { marginLeft: 6, fontSize: 10, padding: '1px 5px', borderRadius: 4, background: 'var(--dsw-alias-state-warning-primary)', color: 'white' } }, deprecationDays > 0 ? t.deprecatesIn.replace('%d', deprecationDays) : t.deprecated) : null, (model.contextWindow && maxContext > 0 && model.contextWindow >= maxContext) ? React.createElement('span', { style: { marginLeft: 4, fontSize: 10, padding: '1px 4px', borderRadius: 3, background: 'var(--dsw-alias-state-warning-primary)', color: 'white' } }, '★') : null, (model.pricing && minPrice !== Infinity && model.pricing.inputPerToken === minPrice) ? React.createElement('span', { style: { marginLeft: 4, fontSize: 10, padding: '1px 4px', borderRadius: 3, background: 'var(--dsw-alias-state-success-primary)', color: 'white' } }, '$') : null, model.tags && model.tags.length ? React.createElement('small', { style: { marginLeft: 6, color: 'var(--dsw-alias-label-secondary)' } }, model.tags.join(', ')) : null, model.pricing ? React.createElement('small', { style: { marginLeft: 6, color: 'var(--dsw-alias-label-secondary)', fontSize: 11 } }, `in $${(model.pricing.inputPerToken*1000000).toFixed(2)}/1M` + (model.pricing.cacheReadPerToken!==undefined ? ` / cache $${(model.pricing.cacheReadPerToken*1000000).toFixed(2)}` : '') + (model.pricing.outputPerToken!==undefined ? ` / out $${(model.pricing.outputPerToken*1000000).toFixed(2)}` : '')) : null, model.description ? React.createElement('small', { title: model.description, style: { display: 'block', marginLeft: 22, color: 'var(--dsw-alias-label-secondary)', fontSize: 11 } }, model.description.slice(0,120)) : null, row.baseURL || row.api ? React.createElement('a', { href: (row.baseURL||'') + '/models/'+encodeURIComponent(model.id), target: '_blank', rel: 'noopener', style: { marginLeft: 8, fontSize: 11 } }, 'docs') : null, React.createElement('button', { type: 'button', onClick: (e)=>{e.preventDefault(); doTry(row.provider, model.id);}, style: { marginLeft: 8, fontSize: 11, padding: '1px 6px', borderRadius: 4 } }, (tryState[row.provider+':'+model.id] && tryState[row.provider+':'+model.id].loading) ? '…' : '▶'), tryState[row.provider+':'+model.id] && !tryState[row.provider+':'+model.id].loading ? React.createElement('small', { style: { marginLeft: 4, color: tryState[row.provider+':'+model.id].ok ? 'var(--dsw-alias-state-success-primary)' : 'var(--dsw-alias-state-error-primary)', fontSize: 11 } }, tryState[row.provider+':'+model.id].ok ? `${tryState[row.provider+':'+model.id].latency}ms` : (tryState[row.provider+':'+model.id].error||tryState[row.provider+':'+model.id].status)) : null); }), filteredChoices.length > pageSize ? React.createElement('div', { style: { display: 'flex', alignItems: 'center', gap: 8, margin: '8px 0' } }, button(t.showMore.replace('%d', Math.min(50, filteredChoices.length - pageSize)), () => setPageSize((s) => s + 50)), React.createElement('small', { style: { color: 'var(--dsw-alias-label-secondary)' } }, t.showingCount.replace('%d', visibleChoices.length).replace('%d', filteredChoices.length))) : null)), React.createElement('small', null, t.pickerNote), React.createElement('div', { style: { marginTop: 8 } }, button(t.allModels, () => setSelectionDraft(choices.map((model) => model.id))), button(t.saveSelection, saveSelection), button(t.cancel, closeSelection))) : null
           const policyEditor = policyProvider === row.provider ? React.createElement('div', { className: 'dms-editor' }, React.createElement('strong', null, `${t.policy}: ${policyProvider}`), React.createElement('label', { style: { display: 'block', marginTop: 8 } }, t.policyInclude, React.createElement('textarea', { rows: 3, value: policyDraft.include, onChange: (e) => setPolicyDraft((current) => ({ ...current, include: e.target.value })), style: { display: 'block', width: '100%' } })), React.createElement('label', { style: { display: 'block', marginTop: 8 } }, t.policyExclude, React.createElement('textarea', { rows: 3, value: policyDraft.exclude, onChange: (e) => setPolicyDraft((current) => ({ ...current, exclude: e.target.value })), style: { display: 'block', width: '100%' } })), React.createElement('label', { style: { display: 'block', marginTop: 8 } }, t.policyRequire, React.createElement('input', { value: policyDraft.requireCapabilities, onChange: (e) => setPolicyDraft((current) => ({ ...current, requireCapabilities: e.target.value })), style: { display: 'block', width: '100%' } })), React.createElement('label', { style: { display: 'block', marginTop: 8 } }, t.policyDeny, React.createElement('input', { value: policyDraft.denyCapabilities, onChange: (e) => setPolicyDraft((current) => ({ ...current, denyCapabilities: e.target.value })), style: { display: 'block', width: '100%' } })), React.createElement('small', null, t.policyNote), React.createElement('div', { style: { marginTop: 8 } }, button(t.savePolicy, savePolicy), button(t.cancel, closePolicy))) : null
           return React.createElement('div', { key: row.provider, className: 'dms-provider-wrap' }, providerCard, selectionEditor, policyEditor)
         }),

package/lib/generic-adapter.js

@@ -38,9 +38,10 @@ function requestHeaders(apiKey) {
 }
 
 async function resolveGenericRequest(profile, { resolveCredential, fetchImpl, signal }) {
-  const apiKey = profile.apiKeyEnv ? await resolveCredential(profile.apiKeyEnv) : ''
+  const keyRef = profile.apiKeyEnv || profile.credentialRef || profile.apiKeyRef
+  const apiKey = keyRef ? await resolveCredential(keyRef) : ''
   const response = await fetchImpl(endpointFor(profile.baseURL), { headers: requestHeaders(apiKey), signal })
-  if (!response.ok) throw catalogRequestError(response.status)
+  if (!response.ok) throw catalogRequestError(response.status, response)
   return response
 }
 

package/lib/history.js

@@ -80,11 +80,29 @@ export function createHistoryEntry(result, { version = 1, maxModels = MAX_HISTOR
   }
 }
 
-export function appendHistory(history, entry, limit = DEFAULT_HISTORY_LIMIT) {
+export const DEFAULT_SNAPSHOT_RETENTION = 5
+
+export function compactHistoryEntry(entry) {
+  if (!entry || typeof entry !== 'object') return entry
+  const providers = (entry.providers ?? []).map((provider) => {
+    const copy = { ...provider }
+    delete copy.before
+    delete copy.after
+    return copy
+  })
+  return { ...entry, providers }
+}
+
+export function appendHistory(history, entry, limit = DEFAULT_HISTORY_LIMIT, { retainSnapshots = DEFAULT_SNAPSHOT_RETENTION } = {}) {
   const rows = Array.isArray(history) ? history.filter((row) => row && typeof row === 'object') : []
   const next = [...rows, structuredClone(entry)]
   next.sort((left, right) => Number(left.version ?? 0) - Number(right.version ?? 0))
-  return next.slice(-historyLimit(limit))
+  const bounded = next.slice(-historyLimit(limit))
+  const cutoff = Math.max(0, bounded.length - Math.max(1, Number(retainSnapshots) || DEFAULT_SNAPSHOT_RETENTION))
+  for (let i = 0; i < cutoff; i++) {
+    bounded[i] = compactHistoryEntry(bounded[i])
+  }
+  return bounded
 }
 
 function briefModel(model) {

package/lib/http.js

@@ -19,14 +19,25 @@ function trusted(req) {
 function readJson(req) {
   return new Promise((resolve, reject) => {
     let raw = ''
+    let failed = false
     req.on('data', (chunk) => {
+      if (failed) return
       raw += chunk
-      if (raw.length > 64 * 1024) reject(new Error('request body is too large'))
+      if (raw.length > 64 * 1024) {
+        failed = true
+        const err = new Error('request body is too large')
+        if (typeof req.destroy === 'function') req.destroy(err)
+        reject(err)
+      }
     })
     req.on('end', () => {
+      if (failed) return
       try { resolve(JSON.parse(raw || '{}')) } catch (error) { reject(error) }
     })
-    req.on('error', reject)
+    req.on('error', (err) => {
+      if (failed) return
+      reject(err)
+    })
   })
 }
 

package/lib/index.js

@@ -142,9 +142,9 @@ export function apply(ctx, config) {
   const scheduler = createSyncScheduler(synchronizer, { getConfig: () => liveConfig })
   reconfigure = () => scheduler.reconfigure()
   scheduler.start()
+  ctx.effect(() => () => scheduler.stop(), 'dsh-model-sync: scheduler cleanup')
   ctx.effect(() => {
     const registrations = registerHttpApi(ctx, synchronizer, { getSchedulerStatus: () => scheduler.status() })
-  ctx.effect(() => () => scheduler.stop(), 'dsh-model-sync: scheduler cleanup')
     return () => {
       for (const dispose of registrations) if (typeof dispose === 'function') dispose()
     }

package/lib/inventory.js

@@ -52,7 +52,7 @@ function liveProviders(llm) {
 }
 
 export function isApiKeyProvider(entry, profile) {
-  if (profile?.apiKeyEnv) return true
+  if (profile?.apiKeyEnv || profile?.credentialRef || profile?.apiKeyRef) return true
   return entry?.declared === false && API_KEY_PROVIDER_SET.has(entry.provider)
 }
 
@@ -74,7 +74,8 @@ export function listApiKeyProviders(llm, settings) {
       declared: entry.declared === true,
       configured: Object.prototype.hasOwnProperty.call(profiles, entry.provider),
       live: live.has(entry.provider),
-      apiKeyEnv: profile.apiKeyEnv ?? '',
+      apiKeyEnv: profile.apiKeyEnv ?? profile.credentialRef ?? profile.apiKeyRef ?? '',
+      credentialRef: profile.credentialRef ?? profile.apiKeyRef ?? profile.apiKeyEnv ?? '',
       baseURL: profile.baseURL ?? '',
       api: profile.api ?? '',
       models: normalizeModels(entry.provider, profile.models ?? []),

package/lib/lifecycle.js

@@ -92,7 +92,7 @@ export function updateLifecycle(previous, beforeModels, advertisedModels, {
     record.firstMissingAt ||= now
     record.lastObservedRun = runNumber
     if (record.status !== 'deprecated') record.status = 'stale'
-    if (removeMissing && record.status === 'stale' && record.consecutiveMissing >= staleGraceRuns) {
+    if (removeMissing && (record.status === 'stale' || record.status === 'deprecated') && record.consecutiveMissing >= staleGraceRuns) {
       record.status = 'removed'
       record.removedAt = now
       record.reason = `missing for ${record.consecutiveMissing} discovery runs`

package/lib/models.js

@@ -36,7 +36,21 @@ const PRICING_ALIASES = Object.freeze({
 function numericValue(value, { allowZero = false } = {}) {
   if (typeof value !== 'number' && typeof value !== 'string') return undefined
   if (typeof value === 'string' && value.trim() === '') return undefined
-  const number = Number(value)
+  const clean = typeof value === 'string' ? value.trim().replace(/,/g, '') : value
+  let number = Number(clean)
+  if (typeof clean === 'string' && !Number.isFinite(number)) {
+    const match = clean.match(/^(\d+(?:\.\d+)?)\s*([kmgt])(i?b)?$/i)
+    if (match) {
+      const base = Number(match[1])
+      const unit = match[2].toLowerCase()
+      const isBinary = Boolean(match[3] && match[3].toLowerCase().startsWith('i'))
+      const mult = unit === 'k' ? (isBinary ? 1024 : 1000)
+        : unit === 'm' ? (isBinary ? 1048576 : 1000000)
+        : unit === 'g' ? (isBinary ? 1073741824 : 1000000000)
+        : 1
+      number = Math.round(base * mult)
+    }
+  }
   if (!Number.isFinite(number) || (allowZero ? number < 0 : number <= 0)) return undefined
   return number
 }
@@ -213,6 +227,14 @@ export function normalizeModels(provider, rows) {
   return out
 }
 
+export function stableStringify(value) {
+  if (Array.isArray(value)) return `[${value.map(stableStringify).join(',')}]`
+  if (value && typeof value === 'object') {
+    return `{${Object.keys(value).sort().map((k) => `${JSON.stringify(k)}:${stableStringify(value[k])}`).join(',')}}`
+  }
+  return JSON.stringify(value)
+}
+
 export function diffModels(previous, next) {
   const sortModels = (rows) => normalizeModels('', rows).sort((left, right) => left.id.localeCompare(right.id))
   const beforeRows = sortModels(previous)
@@ -226,7 +248,7 @@ export function diffModels(previous, next) {
   const renamedBefore = new Set()
   for (const [id, model] of after) {
     if (!before.has(id)) continue
-    if (JSON.stringify(before.get(id)) !== JSON.stringify(model)) {
+    if (stableStringify(before.get(id)) !== stableStringify(model)) {
       changed.push({ before: before.get(id), after: model })
     }
   }

package/lib/reconcile.js

@@ -1,7 +1,7 @@
-import { diffModels, normalizeModels } from './models.js'
+import { diffModels, normalizeModels, stableStringify } from './models.js'
 
 function sameModel(left, right) {
-  return JSON.stringify(left) === JSON.stringify(right)
+  return stableStringify(left) === stableStringify(right)
 }
 
 export function reconcileModels(provider, current, discovered, {

package/lib/reliability.js

@@ -16,9 +16,37 @@ export function isRetryableError(error) {
   return status !== undefined && RETRYABLE_STATUSES.has(status)
 }
 
-export function catalogRequestError(status) {
+export function parseRetryAfter(response) {
+  if (!response?.headers?.get) return undefined
+  const retryAfter = response.headers.get('retry-after')
+  if (retryAfter) {
+    const seconds = Number(retryAfter)
+    if (Number.isFinite(seconds) && seconds >= 0) return Math.round(seconds * 1000)
+    const dateMs = Date.parse(retryAfter)
+    if (Number.isFinite(dateMs)) {
+      const diff = dateMs - Date.now()
+      return diff > 0 ? diff : 0
+    }
+  }
+  const reset = response.headers.get('x-ratelimit-reset')
+  if (reset) {
+    const seconds = Number(reset)
+    if (Number.isFinite(seconds) && seconds > 0) {
+      if (seconds > 1000000000) {
+        const diff = (seconds * 1000) - Date.now()
+        return diff > 0 ? diff : 0
+      }
+      return Math.round(seconds * 1000)
+    }
+  }
+  return undefined
+}
+
+export function catalogRequestError(status, response) {
   const error = new Error(`model catalog request failed with HTTP ${status}`)
   error.status = Number(status)
+  const retryAfterMs = parseRetryAfter(response)
+  if (retryAfterMs !== undefined) error.retryAfterMs = retryAfterMs
   return error
 }
 
@@ -89,7 +117,11 @@ export async function retryWithBackoff(operation, {
     } catch (error) {
       lastError = error
       if (attempt >= total || !isRetryableError(error)) throw error
-      const delayMs = Math.min(Math.max(0, Number(maxDelayMs) || 0), Math.max(0, Number(baseDelayMs) || 0) * (2 ** (attempt - 1)))
+      const defaultDelayMs = Math.min(Math.max(0, Number(maxDelayMs) || 0), Math.max(0, Number(baseDelayMs) || 0) * (2 ** (attempt - 1)))
+      const retryAfterMs = Number(error?.retryAfterMs)
+      const delayMs = (Number.isFinite(retryAfterMs) && retryAfterMs > 0)
+        ? Math.min(Math.max(0, Number(maxDelayMs) || 0), retryAfterMs)
+        : defaultDelayMs
       onRetry?.({ attempt, nextAttempt: attempt + 1, delayMs, error })
       await delay(delayMs, signal)
     }

package/lib/scheduler.js

@@ -128,14 +128,14 @@ export function createSyncScheduler(sync, {
     const config = getConfig() ?? {}
     resetStates(config)
     const values = [...state.values()]
-    const nextAt = values.length > 0 ? Math.min(...values.map((row) => row.nextAt)) : null
+    const nextAt = (timer !== null && values.length > 0) ? Math.min(...values.map((row) => row.nextAt)) : null
     const lastRunAt = values.filter((row) => row.lastRunAt).reduce((max, row) => Math.max(max, row.lastRunAt), 0) || null
     return {
       active: timer !== null,
       running,
       lastRunAt,
       nextRunAt: nextAt,
-      providers: Object.fromEntries(values.map((row) => [row.provider ?? '*', { lastRunAt: row.lastRunAt, nextRunAt: row.nextAt, lastStatus: row.lastStatus, ...(row.lastError ? { lastError: row.lastError } : {}) }])),
+      providers: Object.fromEntries(values.map((row) => [row.provider ?? '*', { lastRunAt: row.lastRunAt, nextRunAt: timer !== null ? row.nextAt : null, lastStatus: row.lastStatus, ...(row.lastError ? { lastError: row.lastError } : {}) }])),
     }
   }
 

package/lib/synchronizer.js

@@ -67,15 +67,41 @@ function diagnosticMessage(error) {
 
 function adapterErrorType(error) {
   if (error?.adapterCode === 'auth') return 'auth'
+  if (error?.adapterCode === 'rate-limit') return 'rate-limit'
   if (error?.adapterCode === 'schema') return 'schema'
   if (error?.adapterCode === 'endpoint') return 'endpoint'
   if (error?.code === 'CREDENTIALS_UNAVAILABLE' || error?.code === 'AUTH') return 'credential'
   const status = statusOfError(error)
   if (status === 401 || status === 403) return 'auth'
+  if (status === 429) return 'rate-limit'
   if (status) return 'http'
   return 'endpoint'
 }
 
+async function replaceWithRetry(settings, ns, patchFn, attempts = 3) {
+  let lastError
+  for (let i = 0; i < attempts; i++) {
+    const section = settings?.get?.(ns)
+    const descriptor = descriptorOf(settings, ns)
+    if (!section || !descriptor || typeof settings.replace !== 'function') {
+      throw Object.assign(new Error(`${ns} settings are not writable`), { code: 'SETTINGS_UNAVAILABLE' })
+    }
+    const next = patchFn(structuredClone(section))
+    try {
+      await settings.replace(ns, next, descriptor.revision)
+      return next
+    } catch (err) {
+      lastError = err
+      const isConflict = err?.code === 'SETTINGS_CONFLICT' || err?.status === 409 || /conflict|revision/i.test(err?.message || '')
+      if (isConflict && i < attempts - 1) {
+        continue
+      }
+      throw err
+    }
+  }
+  throw lastError
+}
+
 function rotationRefs(settings, provider) {
   try {
     const rows = settings?.get?.('dsh-key-rotation')?.providers
@@ -417,16 +443,13 @@ export function createModelSynchronizer(ctx, { getConfig = () => ({}), saveConfi
       const resultsToApply = results.filter((row) => row.status === 'ok' && row.changed && (!dryRun || providerAutoApply(row.provider)))
       if (resultsToApply.length > 0) {
         const settings = settingsOf(ctx)
-        const section = settings?.get?.(PIAI_NS)
-        const descriptor = descriptorOf(settings, PIAI_NS)
-        if (!section || !descriptor || typeof settings.replace !== 'function') {
-          throw Object.assign(new Error('llm-pi-ai settings are not writable'), { code: 'SETTINGS_UNAVAILABLE' })
-        }
-        const next = structuredClone(section)
-        for (const row of resultsToApply) {
-          next.providers = catalogPatch(next, row.provider, row.next).providers
-        }
-        await settings.replace(PIAI_NS, next, descriptor.revision)
+        await replaceWithRetry(settings, PIAI_NS, (section) => {
+          let next = section
+          for (const row of resultsToApply) {
+            next = catalogPatch(next, row.provider, row.next)
+          }
+          return next
+        })
         result.applied = true
       }
       if (!dryRun || resultsToApply.length > 0) {
@@ -495,7 +518,14 @@ export function createModelSynchronizer(ctx, { getConfig = () => ({}), saveConfi
     // Use health as lightweight probe; latency is the signal
     const result = await health({ provider })
     const entry = result?.results?.find((r) => r.provider === provider)
-    return { provider, model, ok: entry?.status === 'ok', latencyMs: Date.now() - start, status: entry?.status ?? 'unknown' }
+    return {
+      provider,
+      model,
+      ok: entry?.status === 'ok',
+      latencyMs: Date.now() - start,
+      status: entry?.status ?? 'unknown',
+      ...(entry?.status !== 'ok' ? { error: entry?.message || entry?.status || 'probe failed' } : {}),
+    }
   }
 
   async function rollbackHistory({ historyId, provider } = {}) {
@@ -518,8 +548,7 @@ export function createModelSynchronizer(ctx, { getConfig = () => ({}), saveConfi
     if (JSON.stringify(current) === JSON.stringify(target)) {
       return { historyId, provider, models: target, applied: false, allowlistUntouched: true }
     }
-    const next = catalogPatch(section, provider, target)
-    await settings.replace(PIAI_NS, next, descriptor.revision)
+    await replaceWithRetry(settings, PIAI_NS, (section) => catalogPatch(section, provider, target))
     return { historyId, provider, models: target, applied: true, allowlistUntouched: true }
   }
 
@@ -565,7 +594,6 @@ export function createModelSynchronizer(ctx, { getConfig = () => ({}), saveConfi
     const nextModels = selectedModels.length
       ? selectedModels.map((id) => availableById.get(id))
       : (safeAvailable.length > 0 ? safeAvailable : current)
-    const next = catalogPatch(section, provider, nextModels)
     const previousModelSelections = structuredClone(config.modelSelections ?? {})
     const modelSelections = { ...previousModelSelections, [provider]: selectedModels }
 
@@ -573,7 +601,7 @@ export function createModelSynchronizer(ctx, { getConfig = () => ({}), saveConfi
     // allowlist back so the two namespaces cannot be left disagreeing.
     await saveConfig({ modelSelections })
     try {
-      await settings.replace(PIAI_NS, next, descriptor.revision)
+      await replaceWithRetry(settings, PIAI_NS, (section) => catalogPatch(section, provider, nextModels))
     } catch (error) {
       try {
         await saveConfig({ modelSelections: previousModelSelections })

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@goodandready/dsh-model-sync",
-  "version": "0.3.4",
+  "version": "0.3.6",
   "description": "Automatic model catalog synchronization for API-key DeepSeek Harness providers.",
   "keywords": [
     "deepseek-harness",
@@ -53,4 +53,4 @@
     "@deepseek-ai/schemastery": "^3.18.1",
     "@deepseek-ai/dsh-llm": "^0.1.0-rc.6"
   }
-}
+}