clauderipple 0.9.1 → 0.9.3

package/CHANGELOG.md

@@ -1,5 +1,47 @@
 # Changelog
 
+## 0.9.3 — 2026-10-14
+
+### Fixed
+
+- **Claude workers offer X-High.** Claude's effort ladder stopped at high and max, so X-High could
+  not be picked for Claude in the mod or the dashboard though GPT had it. Opus, Sonnet and Haiku 5.5
+  all take it (#56).
+- **A model ticked after the mod's worker list was saved is usable.** It used to stay off and show
+  as unavailable in the mod (GPT-6 Astra in #56). A worker the list has not seen now starts on.
+  Settings saved before this keep the old reading until the list is changed once.
+- **The plan limit shows in the band on Claude chats.** The desktop Code tab left the mod's rate
+  limits empty, so the limit never appeared even with it switched on. The mod now falls back to the
+  router's figure for the same Claude login, and hides it when that figure is over 15 minutes old.
+- `/ripple-bar` no longer says the band is shown above the prompt when the layout is Minimal, where
+  only the short line under the prompt is drawn (#56).
+- The close button lights as one disc under the pointer instead of drawing a second box around it.
+
+## 0.9.2 — 2026-10-14
+
+### Added
+
+- **The mod shows when keep-warm held your cache.** A successful ping used to be silent, so it was
+  easy to think keep-warm never ran. When you come back, a card now says how many pings kept how
+  many tokens in cache and how long is left. It goes when you send a prompt or close it, and the
+  cold-cache card takes over if the cache expires after all.
+- **Keep-warm per chat.** The band's flame now turns keep-warm on or off for that chat only: a full
+  flame where it is on, an empty one where it is off. Settings is what a new chat starts with.
+  After a restart or resume, a chat follows Settings again.
+
+- **Grok through a Grok subscription.** A new "Grok subscription" provider uses Grok 4.7, 4.7
+  Fast, 4.6 and 4.5 with the Grok CLI's own sign-in (`grok login`), with no API key. It reads the
+  CLI's session and, when the session is about to expire, has the CLI refresh it. Tools,
+  thinking and effort (low…xhigh, per model) work as on the other translated providers. This
+  is the channel the Grok CLI itself talks to, not a published xAI API, so a CLI update can
+  change it. Pay-per-token access with an xAI API key stays under the xAI preset. Thanks @nakjiya (#52).
+
+### Fixed
+
+- An OpenAI-compatible provider's error that echoed the bare token, without its `Bearer` scheme,
+  could show it in the error message. It is now redacted like the full header value (#52).
+- On a two-line mod card, 리플이 sat on the title's line; it now sits at the card's middle.
+
 ## 0.9.1 — 2026-10-14
 
 ### Added

package/README.es.md

@@ -552,6 +552,7 @@ compatible con Anthropic que hayas configurado está disponible de la misma form
 |---|---|---|---|---|
 | Suscripción de ChatGPT | Backend de Codex | varios inicios de sesión, con cambio cuando uno se agota (también reutiliza el de Codex) | leída de tu suscripción (GPT-6 Sol, Luna, Astra, GPT-5.6 …) | esfuerzo low…max (Luna: ultra), caché de prompts 94–99 % |
 | Google Gemini | API de Gemini | clave de API de AI Studio o inicio de sesión con cuenta de Google (Antigravity, consulta las preguntas frecuentes) | descubierta | traducido (Messages ⇄ generateContent), las firmas de pensamiento se conservan entre turnos |
+| Suscripción de Grok | proxy de chat de la CLI de Grok | el inicio de sesión de la CLI de Grok (`grok login`), renovado por la CLI | leída de tu suscripción (Grok 4.7, 4.7 Fast, 4.6, 4.5) | traducido (Chat Completions), esfuerzo low…xhigh; canal propio de la CLI, no una API publicada ([§4e](docs/ARCHITECTURE.md#4e-grok-providers-implemented-2026-10-08)) |
 | OpenRouter | Compatible con Anthropic | clave de API | más de 400, descubierta | compatibilidad de esfuerzo por modelo leída de la API |
 | DeepSeek, Kimi, Z.ai GLM, MiniMax, Qwen (intl / cn) | Compatible con Anthropic | clave de API | predefinida | verificado con la documentación del proveedor |
 | xAI Grok, Mistral, Groq, Together, Fireworks | Compatible con OpenAI | clave de API | descubierta | traducido (Chat Completions / Responses) |

package/README.ja.md

@@ -321,6 +321,7 @@ ClaudeのモデルはCodexのモデル一覧にそれぞれの名前で表示さ
 |---|---|---|---|---|
 | ChatGPTサブスクリプション | Codexバックエンド | 複数ログイン、上限で自動切り替え（Codexのログインも再利用） | サブスクリプションから取得（GPT-6 Sol、Luna、Astra、GPT-5.6 …） | 推論強度 low…max（Lunaはultra）、プロンプトキャッシュ 94–99 % |
 | Google Gemini | Gemini API | AI StudioのAPIキー、またはGoogleアカウントでのログイン（Antigravity、FAQ参照） | 自動検出 | 変換（Messages ⇄ generateContent）、thought signatureをターン間で保持 |
+| Grokサブスクリプション | Grok CLIのチャットプロキシ | Grok CLIのログイン（`grok login`）、更新はCLIが行う | サブスクリプションから取得（Grok 4.7、4.7 Fast、4.6、4.5） | 変換（Chat Completions）、推論強度 low…xhigh。公開APIではなくCLI専用の経路（[§4e](docs/ARCHITECTURE.md#4e-grok-providers-implemented-2026-10-08)） |
 | OpenRouter | Anthropic互換 | APIキー | 400以上、自動検出 | モデルごとの推論強度への対応をAPIから取得 |
 | DeepSeek、Kimi、Z.ai GLM、MiniMax、Qwen（国際版 / 中国版） | Anthropic互換 | APIキー | プリセット | ベンダーのドキュメントで確認済み |
 | xAI Grok、Mistral、Groq、Together、Fireworks | OpenAI互換 | APIキー | 自動検出 | 変換（Chat Completions / Responses） |

package/README.ko.md

@@ -449,6 +449,7 @@ Anthropic API 키로 갑니다. 구독 로그인 재사용은 Anthropic 약관
 |---|---|---|---|---|
 | ChatGPT 구독 | Codex 백엔드 | 로그인 여러 개, 한도 차면 자동 전환(Codex 로그인도 재사용) | 구독에서 자동으로 읽음(GPT-6 Sol·Luna·Astra, GPT-5.6 …) | 추론 강도 low…max(Luna는 ultra), 프롬프트 캐시 94~99 % |
 | Google Gemini | Gemini API | AI Studio API 키, 또는 Google 계정 로그인(Antigravity, 위 FAQ 참고) | 자동 검색 | 번역(Messages ⇄ generateContent), 턴 사이 thought signature 유지 |
+| Grok 구독 | Grok CLI 채팅 프록시 | Grok CLI 로그인(`grok login`), 갱신은 CLI가 함 | 구독에서 자동으로 읽음(Grok 4.7, 4.7 Fast, 4.6, 4.5) | 번역(Chat Completions), 추론 강도 low…xhigh. 공개 API가 아닌 CLI 전용 채널([§4e](docs/ARCHITECTURE.md#4e-grok-providers-implemented-2026-10-08)) |
 | OpenRouter | Anthropic 호환 | API 키 | 400+, 자동 검색 | 모델별 추론 강도 지원을 API에서 읽음 |
 | DeepSeek, Kimi, Z.ai GLM, MiniMax, Qwen(국제/중국) | Anthropic 호환 | API 키 | 프리셋 | 벤더 공식 문서로 확인 |
 | xAI Grok, Mistral, Groq, Together, Fireworks | OpenAI 호환 | API 키 | 자동 검색 | 번역(Chat Completions / Responses) |

package/README.md

@@ -527,6 +527,7 @@ provider you configured is available the same way.
 |---|---|---|---|---|
 | ChatGPT subscription | Codex backend | several sign-ins, switching when one runs out (Codex login reused too) | read from your subscription (GPT-6 Sol, Luna, Astra, GPT-5.6 …) | effort low…max (Luna: ultra), prompt cache 94–99 % |
 | Google Gemini | Gemini API | AI Studio API key, or Google account sign-in (Antigravity, see the FAQ) | discovered | translated (Messages ⇄ generateContent), thought signatures kept across turns |
+| Grok subscription | Grok CLI chat proxy | the Grok CLI's sign-in (`grok login`), refreshed through the CLI | read from your subscription (Grok 4.7, 4.7 Fast, 4.6, 4.5) | translated (Chat Completions), effort low…xhigh; the CLI's own channel, not a published API ([§4e](docs/ARCHITECTURE.md#4e-grok-providers-implemented-2026-10-08)) |
 | OpenRouter | Anthropic-compatible | API key | 400+, discovered | per-model effort support read from the API |
 | DeepSeek, Kimi, Z.ai GLM, MiniMax, Qwen (intl / cn) | Anthropic-compatible | API key | preset | verified against vendor docs |
 | xAI Grok, Mistral, Groq, Together, Fireworks | OpenAI-compatible | API key | discovered | translated (Chat Completions / Responses) |

package/README.pt-BR.md

@@ -539,6 +539,7 @@ que você tenha configurado fica disponível da mesma forma.
 |---|---|---|---|---|
 | Assinatura do ChatGPT | backend do Codex | vários logins, trocando quando um se esgota (o login do Codex também é reaproveitado) | lida da sua assinatura (GPT-6 Sol, Luna, Astra, GPT-5.6 …) | esforço low…max (Luna: ultra), cache de prompt 94–99% |
 | Google Gemini | API Gemini | chave de API do AI Studio, ou login com conta do Google (Antigravity, veja as Perguntas frequentes) | descoberta automática | traduzido (Messages ⇄ generateContent), thought signatures mantidas entre turnos |
+| Assinatura do Grok | proxy de chat da CLI do Grok | o login da CLI do Grok (`grok login`), renovado pela CLI | lida da sua assinatura (Grok 4.7, 4.7 Fast, 4.6, 4.5) | traduzido (Chat Completions), esforço low…xhigh; canal próprio da CLI, não uma API publicada ([§4e](docs/ARCHITECTURE.md#4e-grok-providers-implemented-2026-10-08)) |
 | OpenRouter | compatível com Anthropic | chave de API | mais de 400, descoberta automática | suporte a esforço por modelo lido da API |
 | DeepSeek, Kimi, Z.ai GLM, MiniMax, Qwen (intl / cn) | compatível com Anthropic | chave de API | preset | conferido com a documentação do fornecedor |
 | xAI Grok, Mistral, Groq, Together, Fireworks | compatível com OpenAI | chave de API | descoberta automática | traduzido (Chat Completions / Responses) |

package/README.zh-CN.md

@@ -322,6 +322,7 @@ Claude 模型会以各自的名称出现在 Codex 的模型列表中（ClaudeRip
 |---|---|---|---|---|
 | ChatGPT 订阅 | Codex 后端 | 可登录多个账号，一个用尽自动切换（也会复用 Codex 登录） | 从你的订阅读取（GPT-6 Sol、Luna、Astra、GPT-5.6 …） | 推理强度 low…max（Luna：ultra），提示缓存 94–99 % |
 | Google Gemini | Gemini API | AI Studio API 密钥，或 Google 账号登录（Antigravity，见[常见问题](#通过-google-账号使用-gemini-呢)） | 自动发现 | 格式转换（Messages ⇄ generateContent），跨轮次保留思维签名（thought signatures） |
+| Grok 订阅 | Grok CLI 聊天代理 | Grok CLI 的登录（`grok login`），由 CLI 刷新 | 从你的订阅读取（Grok 4.7、4.7 Fast、4.6、4.5） | 格式转换（Chat Completions），推理强度 low…xhigh；CLI 自己的通道，不是公开 API（[§4e](docs/ARCHITECTURE.md#4e-grok-providers-implemented-2026-10-08)） |
 | OpenRouter | Anthropic 兼容 | API 密钥 | 400+，自动发现 | 从 API 读取各模型对推理强度的支持情况 |
 | DeepSeek、Kimi、Z.ai GLM、MiniMax、Qwen（国际版 / 国内版） | Anthropic 兼容 | API 密钥 | 预设 | 已对照厂商文档验证 |
 | xAI Grok、Mistral、Groq、Together、Fireworks | OpenAI 兼容 | API 密钥 | 自动发现 | 格式转换（Chat Completions / Responses） |

package/dist/app/package.json

@@ -1,5 +1,5 @@
 {
   "private": true,
-  "version": "0.9.1",
+  "version": "0.9.3",
   "type": "commonjs"
 }

package/dist/cli/src/index.js

@@ -20,6 +20,7 @@ import { applyAppProxy, caTrusted, currentAppProxy, nssDb, removeAppProxy, trust
 import { runtime } from "./runtime.js";
 import { codexOff, codexOn } from "./codex.js";
 import { ingressModels } from "../../router/src/ingress/models.js";
+import { GROK_DEFAULT_URL } from "../../router/src/providers/grok/catalog.js";
 import { claudeLogin, claudeLogout, desktopClaudeCodeDirs } from "./claude-auth.js";
 import { openBrowser } from "./browser.js";
 import { electronPath, installTrayRuntime, startTray } from "./tray.js";
@@ -315,7 +316,7 @@ async function status() {
     const p = await probe({ host: cfg.listen.host, port: cfg.listen.port, caPem: caPath, upstream: cfg.upstream });
     rows.push(["probe", `${p.ok ? "ok" : "FAIL"}: ${p.detail} (${p.ms}ms)`]);
     for (const [name, p2] of Object.entries(cfg.providers)) {
-        const providerUrl = p2.type === "chatgpt" ? (p2.url ?? "https://chatgpt.com") : p2.type === "anthropic" ? "https://api.anthropic.com" : p2.type === "google" ? (p2.url ?? "https://generativelanguage.googleapis.com") : p2.url;
+        const providerUrl = p2.type === "chatgpt" ? (p2.url ?? "https://chatgpt.com") : p2.type === "anthropic" ? "https://api.anthropic.com" : p2.type === "google" ? (p2.url ?? "https://generativelanguage.googleapis.com") : p2.type === "grok" ? (p2.url ?? GROK_DEFAULT_URL) : p2.url;
         const u = new URL(providerUrl);
         rows.push([`provider ${name}`, await tcpCheck(u.hostname, Number(u.port) || (u.protocol === "https:" ? 443 : 80))]);
     }

package/dist/mod/hooks/i18n.ts

@@ -70,6 +70,7 @@ const EN = {
   'band.hit': 'hit {percent}%',
   'band.effortAutoAlt': 'Effort Auto {state}',
   'band.warmAlt': 'Keep cache warm {state}',
+  'band.warmHere': 'this chat only; new chats follow Settings',
   'band.handoffAdvised': 'Handoff (advised: a fresh chat pays now)',
   'band.closeLog': 'Close the request log',
   'band.openLog': 'Open the request log',
@@ -90,6 +91,9 @@ const EN = {
   // Alerts.
   'alerts.coldTitle': 'The cache went cold',
   'alerts.coldBody': 'It sat {minutes} min, so the next prompt re-reads all {tokens} tokens at full price.',
+  'alerts.keptTitle': 'Cache kept warm while you were away',
+  'alerts.keptBody': '{pings} ping(s) held {tokens} tokens in cache, so your next prompt reads them at the cached rate. {minutes} min left.',
+  'alerts.gotIt': 'Got it',
   'alerts.swampTitle': 'Context {percent}% full',
   'alerts.swampBody': '{tokens} tokens ride along with every prompt: slower, dearer, blurrier.',
   'alerts.hotTitle': '{window} limit at {percent}%',
@@ -347,16 +351,17 @@ const EN = {
   'toasts.workerMissing': '{name} is not installed',
   // Commands (stats and debug labels included).
   'commands.description': 'ClaudeRipple effort, workers, cache and session tools',
-  'commands.argumentHint': 'settings|setup|auto|warm|save|handoff [full]|stats|debug|agents|log|bar|cold|swamp|hot|down',
+  'commands.argumentHint': 'settings|setup|auto|warm|save|handoff [full]|stats|debug|agents|log|bar|cold|swamp|hot|down|kept',
   'commands.logDescription': 'Open or close the ClaudeRipple request log',
   'commands.barDescription': 'Show or hide the ClaudeRipple band',
   'commands.logTitle': 'ClaudeRipple',
   'commands.agentsTitle': 'ClaudeRipple Agents',
-  'commands.help': 'ClaudeRipple: settings, setup, auto, warm, save, handoff [full], stats, debug, agents, log, bar, cold, swamp, hot, down.',
+  'commands.help': 'ClaudeRipple: settings, setup, auto, warm, save, handoff [full], stats, debug, agents, log, bar, cold, swamp, hot, down, kept.',
   'commands.done': 'ClaudeRipple: {command}.',
   'commands.logOpened': 'ClaudeRipple log opened.',
   'commands.logClosed': 'ClaudeRipple log closed.',
   'commands.barShown': 'ClaudeRipple line shown above the prompt.',
+  'commands.barShownMinimal': 'The layout is Minimal, so ClaudeRipple shows as a short line under the prompt. For the full band above the prompt, set Layout to Dashboard in /ripple settings.',
   'commands.barHidden': 'ClaudeRipple line hidden. /ripple-bar shows it again.',
   'commands.statsAuto': 'Auto: {prompts} prompts, {requests} requests; {total} weighted tokens',
   'commands.statsTokens': 'Tokens: input {input}, write {write}, read {read}, output {output}',
@@ -515,6 +520,7 @@ const KO: Record<Key, string> = {
   'band.hit': '적중 {percent}%',
   'band.effortAutoAlt': '추론 강도 자동 {state}',
   'band.warmAlt': '캐시 유지 {state}',
+  'band.warmHere': '이 세션만 적용, 새 세션은 설정을 따릅니다',
   'band.handoffAdvised': '인수인계 (권장: 지금은 새 대화가 더 효율적입니다)',
   'band.closeLog': '요청 로그 닫기',
   'band.openLog': '요청 로그 열기',
@@ -535,6 +541,9 @@ const KO: Record<Key, string> = {
   // Alerts.
   'alerts.coldTitle': '캐시가 만료되었습니다',
   'alerts.coldBody': '{minutes}분 동안 대기하여 캐시가 만료되었습니다. 다음 프롬프트는 {tokens} 토큰 전체를 새로 읽어 비용이 발생합니다.',
+  'alerts.keptTitle': '자리를 비운 동안 캐시를 유지했습니다',
+  'alerts.keptBody': '핑 {pings}회로 {tokens} 토큰을 캐시에 붙잡아 두었습니다. 다음 프롬프트도 캐시로 읽습니다. 만료까지 {minutes}분 남았습니다.',
+  'alerts.gotIt': '확인',
   'alerts.swampTitle': '컨텍스트 {percent}% 사용',
   'alerts.swampBody': '프롬프트마다 {tokens} 토큰이 함께 전송되어 응답 속도와 정확도, 비용에 영향을 줍니다.',
   'alerts.hotTitle': '{window} 한도 {percent}% 도달',
@@ -792,16 +801,17 @@ const KO: Record<Key, string> = {
   'toasts.workerMissing': '{name} 에이전트 미설치',
   // Commands (stats and debug labels included).
   'commands.description': 'ClaudeRipple 추론 강도, 워커, 캐시 및 세션 도구',
-  'commands.argumentHint': 'settings|setup|auto|warm|save|handoff [full]|stats|debug|agents|log|bar|cold|swamp|hot|down',
+  'commands.argumentHint': 'settings|setup|auto|warm|save|handoff [full]|stats|debug|agents|log|bar|cold|swamp|hot|down|kept',
   'commands.logDescription': 'ClaudeRipple 요청 로그 열기/닫기',
   'commands.barDescription': 'ClaudeRipple 상태 바 표시/숨기기',
   'commands.logTitle': 'ClaudeRipple',
   'commands.agentsTitle': 'ClaudeRipple 에이전트',
-  'commands.help': 'ClaudeRipple 하위 명령: settings, setup, auto, warm, save, handoff [full], stats, debug, agents, log, bar, cold, swamp, hot, down.',
+  'commands.help': 'ClaudeRipple 하위 명령: settings, setup, auto, warm, save, handoff [full], stats, debug, agents, log, bar, cold, swamp, hot, down, kept.',
   'commands.done': 'ClaudeRipple: {command} 완료.',
   'commands.logOpened': 'ClaudeRipple 로그를 열었습니다.',
   'commands.logClosed': 'ClaudeRipple 로그를 닫았습니다.',
   'commands.barShown': '프롬프트 상단에 ClaudeRipple 상태 바를 표시합니다.',
+  'commands.barShownMinimal': '레이아웃이 최소화로 되어 있어 ClaudeRipple은 프롬프트 아래 한 줄로만 표시됩니다. 프롬프트 위에 상태 바를 띄우려면 /ripple settings에서 레이아웃을 대시보드로 바꾸십시오.',
   'commands.barHidden': 'ClaudeRipple 상태 바를 숨겼습니다. /ripple-bar로 다시 표시할 수 있습니다.',
   'commands.statsAuto': '자동: 프롬프트 {prompts}개, 요청 {requests}개, 가중 토큰 {total}',
   'commands.statsTokens': '토큰: 입력 {input}, 쓰기 {write}, 읽기 {read}, 출력 {output}',
@@ -958,6 +968,7 @@ const ZH: Record<Key, string> = {
   'band.hit': '命中 {percent}%',
   'band.effortAutoAlt': '推理强度自动{state}',
   'band.warmAlt': '缓存保温{state}',
+  'band.warmHere': '仅限本会话，新会话按设置',
   'band.handoffAdvised': '交接（建议：现在换新对话更划算）',
   'band.closeLog': '关闭请求日志',
   'band.openLog': '打开请求日志',
@@ -978,6 +989,9 @@ const ZH: Record<Key, string> = {
   // Alerts.
   'alerts.coldTitle': '缓存已过期',
   'alerts.coldBody': '闲置了 {minutes} 分钟，下一条提示词会按全价重读全部 {tokens} token。',
+  'alerts.keptTitle': '您离开期间缓存已保持温热',
+  'alerts.keptBody': '保温 ping {pings} 次，把 {tokens} token 留在了缓存里，下一条提示词仍按缓存价读取。还有 {minutes} 分钟过期。',
+  'alerts.gotIt': '知道了',
   'alerts.swampTitle': '上下文已用 {percent}%',
   'alerts.swampBody': '每条提示词都带着 {tokens} token：更慢、更贵、更模糊。',
   'alerts.hotTitle': '{window}限额已达 {percent}%',
@@ -1235,16 +1249,17 @@ const ZH: Record<Key, string> = {
   'toasts.workerMissing': '未安装 {name}',
   // Commands (stats and debug labels included).
   'commands.description': 'ClaudeRipple 推理强度、Worker、缓存与会话工具',
-  'commands.argumentHint': 'settings|setup|auto|warm|save|handoff [full]|stats|debug|agents|log|bar|cold|swamp|hot|down',
+  'commands.argumentHint': 'settings|setup|auto|warm|save|handoff [full]|stats|debug|agents|log|bar|cold|swamp|hot|down|kept',
   'commands.logDescription': '打开或关闭 ClaudeRipple 请求日志',
   'commands.barDescription': '显示或隐藏 ClaudeRipple 状态栏',
   'commands.logTitle': 'ClaudeRipple',
   'commands.agentsTitle': 'ClaudeRipple 代理',
-  'commands.help': 'ClaudeRipple 子命令：settings, setup, auto, warm, save, handoff [full], stats, debug, agents, log, bar, cold, swamp, hot, down.',
+  'commands.help': 'ClaudeRipple 子命令：settings, setup, auto, warm, save, handoff [full], stats, debug, agents, log, bar, cold, swamp, hot, down, kept.',
   'commands.done': 'ClaudeRipple: {command}.',
   'commands.logOpened': '已打开 ClaudeRipple 日志。',
   'commands.logClosed': '已关闭 ClaudeRipple 日志。',
   'commands.barShown': 'ClaudeRipple 状态栏已显示在提示词上方。',
+  'commands.barShownMinimal': '布局是“极简”，所以 ClaudeRipple 只在提示词下方显示一行。要在提示词上方显示完整状态栏，请在 /ripple settings 里把布局改为“仪表盘”。',
   'commands.barHidden': 'ClaudeRipple 状态栏已隐藏，/ripple-bar 可重新显示。',
   'commands.statsAuto': '自动：{prompts} 条提示词，{requests} 次请求；加权 token {total}',
   'commands.statsTokens': 'Token：输入 {input}，写入 {write}，读取 {read}，输出 {output}',
@@ -1401,6 +1416,7 @@ const JA: Record<Key, string> = {
   'band.hit': 'ヒット {percent}%',
   'band.effortAutoAlt': '推論強度の自動 {state}',
   'band.warmAlt': 'キャッシュ保持 {state}',
+  'band.warmHere': 'このチャットだけ。新しいチャットは設定に従います',
   'band.handoffAdvised': '引き継ぎ（推奨: 今は新しい会話のほうが得です）',
   'band.closeLog': 'リクエストログを閉じる',
   'band.openLog': 'リクエストログを開く',
@@ -1421,6 +1437,9 @@ const JA: Record<Key, string> = {
   // Alerts.
   'alerts.coldTitle': 'キャッシュの期限が切れました',
   'alerts.coldBody': '{minutes} 分止まっていたので、次のプロンプトで {tokens} トークンを定価で読み直します。',
+  'alerts.keptTitle': '不在中もキャッシュを温かいまま保ちました',
+  'alerts.keptBody': 'キープウォームが {pings} 回、{tokens} トークンをキャッシュに保持しました。次のプロンプトもキャッシュで読めます。期限まで {minutes} 分です。',
+  'alerts.gotIt': '了解',
   'alerts.swampTitle': 'コンテキスト {percent}% 使用中',
   'alerts.swampBody': 'プロンプトごとに {tokens} トークンが付いていき、遅く・高く・曖昧になります。',
   'alerts.hotTitle': '{window}上限の {percent}% に到達',
@@ -1678,16 +1697,17 @@ const JA: Record<Key, string> = {
   'toasts.workerMissing': '{name} がインストールされていません',
   // Commands (stats and debug labels included).
   'commands.description': 'ClaudeRipple の推論強度、ワーカー、キャッシュ、セッションツール',
-  'commands.argumentHint': 'settings|setup|auto|warm|save|handoff [full]|stats|debug|agents|log|bar|cold|swamp|hot|down',
+  'commands.argumentHint': 'settings|setup|auto|warm|save|handoff [full]|stats|debug|agents|log|bar|cold|swamp|hot|down|kept',
   'commands.logDescription': 'ClaudeRipple のリクエストログを開閉',
   'commands.barDescription': 'ClaudeRipple のバーを表示・非表示',
   'commands.logTitle': 'ClaudeRipple',
   'commands.agentsTitle': 'ClaudeRipple エージェント',
-  'commands.help': 'ClaudeRipple のサブコマンド: settings, setup, auto, warm, save, handoff [full], stats, debug, agents, log, bar, cold, swamp, hot, down.',
+  'commands.help': 'ClaudeRipple のサブコマンド: settings, setup, auto, warm, save, handoff [full], stats, debug, agents, log, bar, cold, swamp, hot, down, kept.',
   'commands.done': 'ClaudeRipple: {command}.',
   'commands.logOpened': 'ClaudeRipple のログを開きました。',
   'commands.logClosed': 'ClaudeRipple のログを閉じました。',
   'commands.barShown': 'ClaudeRipple のバーをプロンプトの上に表示しました。',
+  'commands.barShownMinimal': 'レイアウトがミニマルなので、ClaudeRipple はプロンプトの下に 1 行で表示されます。プロンプトの上にバーを出すには、/ripple settings でレイアウトをダッシュボードにしてください。',
   'commands.barHidden': 'ClaudeRipple のバーを非表示にしました。/ripple-bar で再表示できます。',
   'commands.statsAuto': '自動: プロンプト {prompts} 件、リクエスト {requests} 件、加重トークン {total}',
   'commands.statsTokens': 'トークン: 入力 {input}、書き込み {write}、読み取り {read}、出力 {output}',
@@ -1844,6 +1864,7 @@ const ES: Record<Key, string> = {
   'band.hit': 'acierto {percent}%',
   'band.effortAutoAlt': 'Esfuerzo auto {state}',
   'band.warmAlt': 'Mantener la caché caliente {state}',
+  'band.warmHere': 'solo este chat; los nuevos siguen los Ajustes',
   'band.handoffAdvised': 'Traspaso (recomendado: ahora conviene una conversación nueva)',
   'band.closeLog': 'Cerrar el registro de solicitudes',
   'band.openLog': 'Abrir el registro de solicitudes',
@@ -1864,6 +1885,9 @@ const ES: Record<Key, string> = {
   // Alerts.
   'alerts.coldTitle': 'La caché se enfrió',
   'alerts.coldBody': 'Estuvo {minutes} min parada, así que el próximo prompt relee sus {tokens} tokens a precio completo.',
+  'alerts.keptTitle': 'La caché se mantuvo caliente mientras no estabas',
+  'alerts.keptBody': '{pings} ping(s) mantuvieron {tokens} tokens en caché, así que tu próximo prompt los lee al precio en caché. Quedan {minutes} min.',
+  'alerts.gotIt': 'Entendido',
   'alerts.swampTitle': 'Contexto al {percent}%',
   'alerts.swampBody': '{tokens} tokens viajan con cada prompt: más lento, más caro, más difuso.',
   'alerts.hotTitle': 'Límite {window} al {percent}%',
@@ -2121,16 +2145,17 @@ const ES: Record<Key, string> = {
   'toasts.workerMissing': '{name} no está instalado',
   // Commands (stats and debug labels included).
   'commands.description': 'Esfuerzo, workers, caché y herramientas de sesión de ClaudeRipple',
-  'commands.argumentHint': 'settings|setup|auto|warm|save|handoff [full]|stats|debug|agents|log|bar|cold|swamp|hot|down',
+  'commands.argumentHint': 'settings|setup|auto|warm|save|handoff [full]|stats|debug|agents|log|bar|cold|swamp|hot|down|kept',
   'commands.logDescription': 'Abre o cierra el registro de solicitudes de ClaudeRipple',
   'commands.barDescription': 'Muestra u oculta la barra de ClaudeRipple',
   'commands.logTitle': 'ClaudeRipple',
   'commands.agentsTitle': 'Agentes de ClaudeRipple',
-  'commands.help': 'Subcomandos de ClaudeRipple: settings, setup, auto, warm, save, handoff [full], stats, debug, agents, log, bar, cold, swamp, hot, down.',
+  'commands.help': 'Subcomandos de ClaudeRipple: settings, setup, auto, warm, save, handoff [full], stats, debug, agents, log, bar, cold, swamp, hot, down, kept.',
   'commands.done': 'ClaudeRipple: {command}.',
   'commands.logOpened': 'Registro de ClaudeRipple abierto.',
   'commands.logClosed': 'Registro de ClaudeRipple cerrado.',
   'commands.barShown': 'Barra de ClaudeRipple visible sobre el prompt.',
+  'commands.barShownMinimal': 'La disposición es Mínimo, así que ClaudeRipple aparece como una línea corta bajo el prompt. Para la barra completa sobre el prompt, elige Panel en Disposición en /ripple settings.',
   'commands.barHidden': 'Barra de ClaudeRipple oculta. /ripple-bar la vuelve a mostrar.',
   'commands.statsAuto': 'Auto: {prompts} prompts, {requests} solicitudes; {total} tokens ponderados',
   'commands.statsTokens': 'Tokens: entrada {input}, escritura {write}, lectura {read}, salida {output}',
@@ -2287,6 +2312,7 @@ const PT: Record<Key, string> = {
   'band.hit': 'acerto {percent}%',
   'band.effortAutoAlt': 'Esforço auto {state}',
   'band.warmAlt': 'Manter o cache quente {state}',
+  'band.warmHere': 'só esta conversa; as novas seguem as Configurações',
   'band.handoffAdvised': 'Repasse (recomendado: agora uma conversa nova compensa)',
   'band.closeLog': 'Fechar o log de requisições',
   'band.openLog': 'Abrir o log de requisições',
@@ -2307,6 +2333,9 @@ const PT: Record<Key, string> = {
   // Alerts.
   'alerts.coldTitle': 'O cache esfriou',
   'alerts.coldBody': 'Ficou parada {minutes} min, então o próximo prompt relê os {tokens} tokens pelo preço cheio.',
+  'alerts.keptTitle': 'O cache ficou quente enquanto você esteve fora',
+  'alerts.keptBody': '{pings} ping(s) seguraram {tokens} tokens no cache, então o próximo prompt os lê pelo preço em cache. Faltam {minutes} min.',
+  'alerts.gotIt': 'Entendi',
   'alerts.swampTitle': 'Contexto {percent}% cheio',
   'alerts.swampBody': '{tokens} tokens vão junto com cada prompt: mais lento, mais caro, mais confuso.',
   'alerts.hotTitle': 'Limite {window} em {percent}%',
@@ -2564,16 +2593,17 @@ const PT: Record<Key, string> = {
   'toasts.workerMissing': '{name} não está instalado',
   // Commands (stats and debug labels included).
   'commands.description': 'Esforço, workers, cache e ferramentas de sessão do ClaudeRipple',
-  'commands.argumentHint': 'settings|setup|auto|warm|save|handoff [full]|stats|debug|agents|log|bar|cold|swamp|hot|down',
+  'commands.argumentHint': 'settings|setup|auto|warm|save|handoff [full]|stats|debug|agents|log|bar|cold|swamp|hot|down|kept',
   'commands.logDescription': 'Abre ou fecha o log de requisições do ClaudeRipple',
   'commands.barDescription': 'Mostra ou oculta a barra do ClaudeRipple',
   'commands.logTitle': 'ClaudeRipple',
   'commands.agentsTitle': 'Agentes do ClaudeRipple',
-  'commands.help': 'Subcomandos do ClaudeRipple: settings, setup, auto, warm, save, handoff [full], stats, debug, agents, log, bar, cold, swamp, hot, down.',
+  'commands.help': 'Subcomandos do ClaudeRipple: settings, setup, auto, warm, save, handoff [full], stats, debug, agents, log, bar, cold, swamp, hot, down, kept.',
   'commands.done': 'ClaudeRipple: {command}.',
   'commands.logOpened': 'Log do ClaudeRipple aberto.',
   'commands.logClosed': 'Log do ClaudeRipple fechado.',
   'commands.barShown': 'Barra do ClaudeRipple visível acima do prompt.',
+  'commands.barShownMinimal': 'O layout está em Mínimo, então o ClaudeRipple aparece como uma linha curta abaixo do prompt. Para a barra completa acima do prompt, escolha Painel em Layout no /ripple settings.',
   'commands.barHidden': 'Barra do ClaudeRipple oculta. /ripple-bar mostra de novo.',
   'commands.statsAuto': 'Auto: {prompts} prompts, {requests} requisições; {total} tokens ponderados',
   'commands.statsTokens': 'Tokens: entrada {input}, gravação {write}, leitura {read}, saída {output}',

package/dist/mod/hooks/ladder.js

@@ -150,9 +150,9 @@ export function preset(id, choices) {
     }
     return { rungs, substituted };
 }
-/** The choices a person ticked; an empty list ticks every one. */
-export function tickedChoices(choices, allow) {
-    return choices.filter(choice => isAllowed(candidateKey(choice.target), choice.provider, allow));
+/** The choices a person ticked; an empty list ticks every one, and a key absent from `known` ticks on too. */
+export function tickedChoices(choices, allow, known = []) {
+    return choices.filter(choice => isAllowed(candidateKey(choice.target), choice.provider, allow, known));
 }
 /** Only empty lists need filling; unticked entries stay in order and are skipped at dispatch. */
 export function ladderNeedsFill(rungs, _allow = []) {
@@ -162,8 +162,8 @@ export function ladderNeedsFill(rungs, _allow = []) {
  * Fill empty lists from the ticked, connected choices. Existing ordered lists are never rewritten
  * for temporary availability or shared allow-list changes; dispatch explains each skipped entry.
  */
-export function fillLadder(rungs, allow, choices) {
-    const pick = preset('claudeGpt', tickedChoices(choices, allow));
+export function fillLadder(rungs, allow, choices, known = []) {
+    const pick = preset('claudeGpt', tickedChoices(choices, allow, known));
     const out = {};
     const filled = [];
     for (const rung of RUNGS) {

package/dist/mod/hooks/ladder.ts

@@ -146,9 +146,9 @@ export function preset(id: PresetId, choices: readonly LadderChoice[]): Preset {
   return { rungs, substituted }
 }
 
-/** The choices a person ticked; an empty list ticks every one. */
-export function tickedChoices(choices: readonly LadderChoice[], allow: readonly string[]): LadderChoice[] {
-  return choices.filter(choice => isAllowed(candidateKey(choice.target), choice.provider, allow))
+/** The choices a person ticked; an empty list ticks every one, and a key absent from `known` ticks on too. */
+export function tickedChoices(choices: readonly LadderChoice[], allow: readonly string[], known: readonly string[] = []): LadderChoice[] {
+  return choices.filter(choice => isAllowed(candidateKey(choice.target), choice.provider, allow, known))
 }
 
 /** Only empty lists need filling; unticked entries stay in order and are skipped at dispatch. */
@@ -160,8 +160,8 @@ export function ladderNeedsFill(rungs: Rungs, _allow: readonly string[] = []): b
  * Fill empty lists from the ticked, connected choices. Existing ordered lists are never rewritten
  * for temporary availability or shared allow-list changes; dispatch explains each skipped entry.
  */
-export function fillLadder(rungs: Rungs, allow: readonly string[], choices: readonly LadderChoice[]): { rungs: Record<Rung, Step[]>; filled: Rung[] } {
-  const pick = preset('claudeGpt', tickedChoices(choices, allow))
+export function fillLadder(rungs: Rungs, allow: readonly string[], choices: readonly LadderChoice[], known: readonly string[] = []): { rungs: Record<Rung, Step[]>; filled: Rung[] } {
+  const pick = preset('claudeGpt', tickedChoices(choices, allow, known))
   const out = {} as Record<Rung, Step[]>
   const filled: Rung[] = []
   for (const rung of RUNGS) {