openclacky 1.5.1 → 1.5.3

data/lib/clacky/default_extensions/ext-studio/panels/studio/view.js

@@ -51,6 +51,7 @@
       "err.version_conflict": "Version {{ver}} has already been published. Please enter a higher version number above and try again.",
       "err.invalid_version": "Invalid version. Use format like 1.0.0.",
       "err.name_taken": "The name \"{{id}}\" is already taken by another user.",
+      "err.id_folder_mismatch": "Extension ID mismatch: the folder is named \"{{folder}}\" but ext.yml declares id \"{{id}}\". Rename the folder or the id so they match before publishing.",
       "rename.label": "Choose a new ID for your extension:",
       "rename.placeholder": "e.g. my-ext-yourname",
       "rename.hint": "Only lowercase letters, digits and hyphens. This will rename the local folder and update ext.yml.",
@@ -235,6 +236,7 @@
       "err.version_conflict": "版本 {{ver}} 已发布过，请在上方输入更高的版本号后重试。",
       "err.invalid_version": "版本号格式不对，请使用如 1.0.0 的格式。",
       "err.name_taken": "扩展名「{{id}}」已被其他用户占用。",
+      "err.id_folder_mismatch": "扩展 ID 不一致：文件夹名为「{{folder}}」，但 ext.yml 声明的 id 是「{{id}}」。请先把文件夹名或 id 改成一致再发布。",
       "rename.label": "请为你的扩展选择一个新 ID：",
       "rename.placeholder": "例如 my-ext-yourname",
       "rename.hint": "只能使用小写字母、数字和连字符。",
@@ -726,6 +728,7 @@
         } catch (e) {
           const msg = e.message || "";
           const isConflict = /must be greater than/i.test(msg);
+          const mismatch = msg.match(/folder is '([^']+)' but ext\.yml declares id '([^']+)'/);
           if (isConflict) {
             setProgress(t("err.version_conflict", { ver }), true);
             verInput.classList.add("studio-input-error");
@@ -733,6 +736,8 @@
             verInput.focus();
             verInput.select();
             verInput.addEventListener("input", () => verInput.classList.remove("studio-input-error"), { once: true });
+          } else if (mismatch) {
+            setProgress(t("err.id_folder_mismatch", { folder: mismatch[1], id: mismatch[2] }), true);
           } else {
             setProgress(t("err.generic", { msg }), true);
           }
@@ -1952,7 +1957,7 @@
     .studio-form-status { font-size: 12px; color: var(--color-text-tertiary); }
     .studio-btn-sm { padding: 4px 10px; font-size: 12px; }
     .studio-readme-overlay { align-items: center; justify-content: center; }
-    .studio-readme-modal { width: min(1200px, calc(100vw - 48px)); height: calc(100vh - 80px); max-width: calc(100vw - 48px); border-radius: var(--radius-md, 8px); margin: 0; padding: 24px 28px 20px; display: flex; flex-direction: column; }
+    .studio-readme-modal { max-width: 1280px; width: 98%; height: 90vh; max-height: 960px; border-radius: var(--radius-md, 8px); margin: 0; padding: 24px 28px 20px; display: flex; flex-direction: column; }
     .studio-readme-modal .studio-modal-title { margin-bottom: 16px; flex-shrink: 0; }
     .studio-readme-split { display: flex; gap: 14px; flex: 1; min-height: 0; margin-bottom: 18px; }
     .studio-readme-pane { flex: 1; display: flex; flex-direction: column; min-width: 0; }
@@ -1960,13 +1965,20 @@
     .studio-readme-textarea { flex: 1; resize: none; font-family: monospace; font-size: 12.5px; line-height: 1.65; border-radius: var(--radius-sm); }
     .studio-readme-preview { flex: 1; overflow-y: auto; overflow-x: hidden; border: 1px solid var(--color-border-primary); border-radius: var(--radius-sm); padding: 12px 16px; background: var(--color-bg-secondary); font-size: 13px; line-height: 1.75; }
     .studio-readme-preview > *:first-child { margin-top: 0; }
-    .studio-readme-preview img { max-width: 100%; height: auto; display: block; border-radius: var(--radius-sm); }
     .studio-readme-screenshots { border-top: 1px solid var(--color-border-primary); padding-top: 14px; margin-bottom: 6px; }
     .studio-readme-screenshots-row { display: flex; align-items: center; gap: 10px; }
     .studio-readme-screenshots-hint { font-size: 11px; color: var(--color-text-muted); }
     .studio-readme-card { margin-top: 8px; }
     .studio-readme-card-preview { max-height: 200px; overflow-y: auto; font-size: 13px; }
-    .extension-readme-body img { max-width: min(800px, 100%); height: auto; }
+    .extension-readme-body img { max-width: min(800px, 100%); height: auto; display: inline; vertical-align: middle; }
+    .studio-readme-preview img { max-width: 100%; height: auto; display: inline; vertical-align: middle; }
+    .studio-readme-preview hr, .extension-readme-body hr { border: none; border-top: 1px solid var(--color-border-primary); margin: 1.25em 0; }
+    .studio-readme-preview table, .extension-readme-body table { border-collapse: collapse; width: 100%; margin: 0.75em 0; font-size: 0.8125rem; }
+    .studio-readme-preview th, .studio-readme-preview td, .extension-readme-body th, .extension-readme-body td { border: 1px solid var(--color-border-primary); padding: 0.375em 0.75em; text-align: left; }
+    .studio-readme-preview th, .extension-readme-body th { background: rgba(128,128,128,0.15); font-weight: 600; }
+    .studio-readme-preview pre, .extension-readme-body pre { background: rgba(128,128,128,0.15); border-radius: 6px; padding: 0.75em 1em; overflow-x: auto; margin: 0.625em 0; }
+    .studio-readme-preview code, .extension-readme-body code { font-family: monospace; font-size: 0.8125em; background: rgba(128,128,128,0.15); border-radius: 4px; padding: 0.15em 0.4em; }
+    .studio-readme-preview pre code, .extension-readme-body pre code { background: none; padding: 0; }
     .studio-link-btn { background: none; border: none; padding: 0; font-size: 12px; color: var(--color-accent-primary); cursor: pointer; text-decoration: underline; text-underline-offset: 2px; }
     .studio-link-btn:hover { color: var(--color-accent-hover); }
   `;

data/lib/clacky/extension/packager.rb

@@ -3,6 +3,7 @@
 require "fileutils"
 require "tmpdir"
 require "open-uri"
+require "psych"
 require "zip"
 
 module Clacky
@@ -38,6 +39,7 @@ module Clacky
           raise Error, "no container found at #{container_dir} (expected #{MANIFEST})"
         end
 
+        refuse_id_folder_mismatch!(slug, container_dir)
         verify_container!(slug, container_dir)
         refuse_protected!(slug, container_dir)
 
@@ -96,6 +98,21 @@ module Clacky
         Clacky::ExtensionLoader.invalidate_cache!
       end
 
+      # The container's identity must be the same on both sides of the publish
+      # boundary: the client keys extensions by folder name, the marketplace
+      # keys them by ext.yml `id`. If they diverge, a rename on one side alone
+      # lets a container masquerade under a name the uniqueness check never saw.
+      private def refuse_id_folder_mismatch!(slug, container_dir)
+        manifest = File.join(container_dir, MANIFEST)
+        data     = Psych.safe_load(File.read(manifest), permitted_classes: [], aliases: true) || {}
+        declared = data["id"].to_s.strip
+        return if declared == slug
+
+        raise Error,
+              "extension id mismatch: folder is '#{slug}' but #{MANIFEST} declares id '#{declared}'. " \
+              "Rename the folder or the id so they match before packing."
+      end
+
       # Packing is for authored (open) containers. Anything already protected or
       # carrying encrypted skills is produced by the platform pipeline; refuse
       # to re-pack it so we never smuggle marketplace artifacts out of band.

data/lib/clacky/locales/en.rb

@@ -9,11 +9,12 @@ module Clacky
       "llm.error.403.model_not_allowed" => "This model is not available on your current plan",
       "llm.error.403.api_key_revoked"   => "API key has been revoked, please generate a new one",
       "llm.error.403.api_key_expired"   => "API key has expired, please generate a new one",
-      "llm.error.403.quota_exceeded"    => "Quota exceeded, please upgrade your plan",
+      "llm.error.403.quota_exceeded"    => "API key quota exceeded. Please update the quota or disable the limit in the admin console",
       "llm.error.403.access_denied"     => "Access denied, please check your API key permissions",
       "llm.error.403.default"           => "Access denied",
       "llm.error.endpoint_not_found"    => "API endpoint not found, please check your service URL",
       "llm.error.rate_limit_429"        => "Rate limit exceeded, please wait a moment",
+      "llm.error.quota_exhausted"       => "API quota exhausted (key usage limit reached). Please raise the quota or upgrade your plan in the console",
       "llm.error.server_error"          => "Service temporarily unavailable (%<status>d), retrying...",
       "llm.error.unexpected"            => "Unexpected error (%<status>d)",
       "llm.error.html_response"         => "Service temporarily unavailable (received HTML error page), retrying...",
@@ -21,6 +22,7 @@ module Clacky
       "llm.error.request_timeout"       => "Request timed out after %<retries>d retries",
       "llm.error.network_failed"        => "Network connection failed after %<retries>d retries",
       "llm.error.service_unavailable"   => "Service unavailable after %<retries>d retries",
+      "llm.warn.switching_to_fallback_url" => "Primary endpoint unreachable. Switching to fallback gateway: %<url>s",
       "platform.error.invalid_proof"        => "Invalid license key — please check and try again.",
       "platform.error.invalid_signature"    => "Invalid request signature.",
       "platform.error.nonce_replayed"       => "Duplicate request detected. Please try again.",
@@ -28,7 +30,7 @@ module Clacky
       "platform.error.license_revoked"      => "This license has been revoked. Please contact support.",
       "platform.error.license_expired"      => "This license has expired. Please renew to continue.",
       "platform.error.device_limit_reached" => "Device limit reached for this license.",
-      "platform.error.device_revoked"       => "This device has been revoked from the license.",
+      "platform.error.device_revoked"       => "This device has been revoked from the license. To re-activate, please wait 15 minutes and try again.",
       "platform.error.invalid_license"      => "License key not found. Please verify the key.",
       "platform.error.device_not_found"     => "Device not registered. Please re-activate.",
       "platform.error.contributor_required" => "Publishing extensions requires becoming a contributor. Sign in, open \"My Extensions\", and click \"Become a contributor\" (no review needed).",

data/lib/clacky/locales/zh.rb

@@ -9,11 +9,12 @@ module Clacky
       "llm.error.403.model_not_allowed" => "当前模型不支持免费试用，请升级套餐或切换其他模型",
       "llm.error.403.api_key_revoked"   => "API 密钥已被撤销，请前往控制台重新生成",
       "llm.error.403.api_key_expired"   => "API 密钥已过期，请前往控制台重新生成",
-      "llm.error.403.quota_exceeded"    => "配额已用完，请升级套餐",
+      "llm.error.403.quota_exceeded"    => "API Key 配额已超限，请在管理后台调整配额或关闭限制",
       "llm.error.403.access_denied"     => "访问被拒绝，请检查 API 密钥权限",
       "llm.error.403.default"           => "访问被拒绝",
       "llm.error.endpoint_not_found"    => "API 端点不存在，请检查服务地址配置",
-      "llm.error.rate_limit_429"        => "请求过于频繁，请稍候重试",
+      "llm.error.rate_limit_429"        => "请求过于频繁，请稍后重试",
+      "llm.error.quota_exhausted"       => "API 配额已用尽（已达 Key 的用量上限），请前往控制台提升配额或升级套餐后继续",
       "llm.error.server_error"          => "服务暂时不可用（%<status>d），正在重试...",
       "llm.error.unexpected"            => "请求失败（%<status>d）",
       "llm.error.html_response"         => "服务暂时不可用（收到 HTML 错误页），正在重试...",
@@ -21,6 +22,7 @@ module Clacky
       "llm.error.request_timeout"       => "请求超时（已重试 %<retries>d 次）",
       "llm.error.network_failed"        => "网络连接失败（已重试 %<retries>d 次）",
       "llm.error.service_unavailable"   => "服务暂时不可用（已重试 %<retries>d 次）",
+      "llm.warn.switching_to_fallback_url" => "主节点无法连接，正在切换到备用节点：%<url>s",
       "platform.error.invalid_proof"        => "许可证密钥无效，请检查后重试。",
       "platform.error.invalid_signature"    => "请求签名无效。",
       "platform.error.nonce_replayed"       => "检测到重复请求，请重试。",
@@ -28,7 +30,7 @@ module Clacky
       "platform.error.license_revoked"      => "该许可证已被吊销，请联系客服。",
       "platform.error.license_expired"      => "该许可证已过期，请续订后继续。",
       "platform.error.device_limit_reached" => "该许可证的设备数量已达上限。",
-      "platform.error.device_revoked"       => "该设备已从许可证中移除。",
+      "platform.error.device_revoked"       => "该设备已从许可证中移除。如需重新绑定，请等待 15 分钟后再试。",
       "platform.error.invalid_license"      => "未找到许可证密钥，请核对后重试。",
       "platform.error.device_not_found"     => "设备未注册，请重新激活。",
       "platform.error.contributor_required" => "发布扩展需要先成为扩展贡献者。请登录平台，打开「我的扩展」页面，点击「成为扩展贡献者」即可开通（无需审核）。",

data/lib/clacky/message_format/anthropic.rb

@@ -66,6 +66,7 @@ module Clacky
         system_text = system_messages.map { |m| extract_text(m[:content]) }.join("\n\n")
 
         api_messages = regular_messages.map { |msg| to_api_message(msg, caching_enabled) }
+        api_messages = merge_consecutive_tool_results(api_messages)
         api_tools    = tools&.map { |t| to_api_tool(t) }
 
         if caching_enabled && api_tools&.any?
@@ -76,7 +77,33 @@ module Clacky
         body[:system] = system_text unless system_text.empty?
         body[:tools]  = api_tools   if api_tools&.any?
 
-        if (effort = normalized_effort(reasoning_effort))
+        # Kimi (Moonshot Coding Plan) models routed through the Anthropic
+        # /v1/messages format expect thinking:{type:"enabled"} — the native
+        # Anthropic "adaptive" type is Claude-specific and is silently ignored
+        # by the Kimi backend, so thinking never actually activates on K3.
+        #
+        # K3 additionally supports a "max" effort level (strongest reasoning)
+        # beyond the standard low/medium/high triple. The generic
+        # normalized_effort() helper drops "max" because it only whitelists
+        # the standard three levels, making K3's signature strongest-reasoning
+        # mode unreachable through this format adapter.
+        #
+        # This branch only changes K3's wire format; every other model keeps
+        # using the adaptive path below unchanged. When reasoning_effort is
+        # nil/empty we leave the body untouched (provider default), and we
+        # never emit output_config for "on" so the backend can pick its own
+        # default — both behaviours match the official kimi CLI.
+        if model.to_s.match?(/\A(kimi-)?k3/i)
+          effort_str = reasoning_effort.to_s
+          k3_effort =
+            case effort_str
+            when "max", "xhigh"              then "max"
+            when "high", "medium", "low"     then effort_str
+            else                                  nil
+            end
+          body[:thinking] = { type: "enabled" }
+          body[:output_config] = { effort: k3_effort } if k3_effort
+        elsif (effort = normalized_effort(reasoning_effort))
           body[:thinking] = { type: "adaptive" }
           body[:output_config] = { effort: effort }
         end
@@ -90,6 +117,34 @@ module Clacky
         %w[low medium high].include?(s) ? s : nil
       end
 
+      # Merge consecutive tool_result user messages into a single user message.
+      # Anthropic requires all tool_results for an assistant turn to appear in
+      # a SINGLE user message immediately following the assistant. When an
+      # assistant calls multiple tools, the canonical format produces one
+      # role:"tool" message per result, which to_api_message converts into
+      # separate role:"user" messages — breaking the Anthropic contract.
+      private_class_method def self.merge_consecutive_tool_results(api_messages)
+        result = []
+        pending = nil
+
+        api_messages.each do |msg|
+          if msg[:role] == "user" && msg[:content].is_a?(Array) &&
+             msg[:content].any? { |b| b.is_a?(Hash) && b[:type] == "tool_result" }
+            if pending
+              pending[:content] += msg[:content]
+            else
+              pending = msg
+              result << pending
+            end
+          else
+            pending = nil
+            result << msg
+          end
+        end
+
+        result
+      end
+
       # ── Response parsing ──────────────────────────────────────────────────────
 
       # Parse Anthropic API response into canonical internal format.

data/lib/clacky/message_format/open_ai.rb

@@ -47,7 +47,7 @@ module Clacky
       def build_request_body(messages, model, tools, max_tokens, caching_enabled, vision_supported: true, reasoning_effort: nil)
         api_messages = messages.map { |msg| normalize_message_content(msg, vision_supported: vision_supported) }
 
-        body = { model: model, max_tokens: max_tokens, messages: api_messages }
+        body = { model: model, token_field_for(model) => max_tokens, messages: api_messages }
 
         if tools&.any?
           if caching_enabled
@@ -59,9 +59,7 @@ module Clacky
           end
         end
 
-        if reasoning_effort && !reasoning_effort.to_s.empty?
-          body[:reasoning_effort] = reasoning_effort.to_s
-        end
+        apply_reasoning_params(body, model, reasoning_effort)
 
         body
       end
@@ -196,6 +194,73 @@ module Clacky
 
       # ── Private helpers ───────────────────────────────────────────────────────
 
+      # Returns the token-limit field name for the given model.
+      # Most OpenAI-compatible APIs use :max_tokens; MiMo-V2.5 (Xiaomi) uses
+      # :max_completion_tokens to cap the combined thinking + answer length.
+      private_class_method def self.token_field_for(model)
+        model.to_s.match?(/^mimo-v2/i) ? :max_completion_tokens : :max_tokens
+      end
+
+      # Apply model-specific reasoning / thinking parameters to the request body.
+      #
+      # Different model families expose extended reasoning through different API
+      # fields. We map the shared internal reasoning_effort value to each
+      # family's native parameters in-place.
+      #
+      # Zero side-effects: when reasoning_effort is nil/empty the body is
+      # unchanged, preserving the provider default for all models.
+      private_class_method def self.apply_reasoning_params(body, model, reasoning_effort)
+        effort_str = reasoning_effort.to_s
+
+        if model.to_s.match?(/^glm-[45]/i)
+          # GLM (Zhipu / Z.ai) supports a native top-level "thinking" field
+          # ({type: "enabled"|"disabled"}) plus a restricted reasoning_effort
+          # that only accepts "max" or "high". Other effort levels collapse
+          # to "high". Send both fields so GLM activates thinking correctly.
+          if %w[off nothink disabled].include?(effort_str)
+            body[:thinking] = { type: "disabled" }
+          elsif !effort_str.empty?
+            glm_effort =
+              case effort_str
+              when "max", "xhigh" then "max"
+              when "high"          then "high"
+              when "medium", "low" then "high"   # GLM collapses these to "high"
+              else                      "max"
+              end
+            body[:thinking] = { type: "enabled" }
+            body[:reasoning_effort] = glm_effort
+          end
+        elsif model.to_s.match?(/^kimi-k3/i)
+          # Kimi K3 reasoning_effort only accepts low/high/max (default max).
+          # Thinking is always enabled and cannot be turned off; "off" maps
+          # to the lowest intensity "low". Unsupported effort levels (medium,
+          # xhigh) would be rejected or ignored by the server.
+          if %w[off nothink disabled].include?(effort_str)
+            body[:reasoning_effort] = "low"
+          elsif !effort_str.empty?
+            body[:reasoning_effort] =
+              case effort_str
+              when "max", "xhigh"  then "max"
+              when "high"          then "high"
+              when "medium", "low" then "low"
+              else                      "max"
+              end
+          end
+        elsif model.to_s.match?(/^mimo-v2/i)
+          # MiMo-V2.5 (Xiaomi) controls reasoning via a top-level
+          # thinking:{type:...} field rather than reasoning_effort. Map the
+          # internal reasoning_effort value to thinking on/off so MiMo does
+          # not receive an unsupported parameter.
+          if %w[off nothink disabled].include?(effort_str)
+            body[:thinking] = { type: "disabled" }
+          elsif !effort_str.empty?
+            body[:thinking] = { type: "enabled" }
+          end
+        elsif reasoning_effort && !effort_str.empty?
+          body[:reasoning_effort] = effort_str
+        end
+      end
+
       private_class_method def self.parse_tool_calls(raw)
         return nil if raw.nil? || raw.empty?
 

data/lib/clacky/message_history.rb

@@ -10,7 +10,7 @@ module Clacky
     # Fields that are internal to the agent and must not be sent to the API.
     INTERNAL_FIELDS = %i[
       task_id created_at system_injected session_context memory_update
-      subagent_instructions subagent_result token_usage
+      subagent_instructions subagent_result subagent_transcript token_usage
       compressed_summary chunk_path truncated transient
       chunk_index chunk_count
     ].freeze
@@ -71,6 +71,20 @@ module Clacky
       self
     end
 
+    # Attach a key/value onto the most recent tool result message matching the
+    # given tool_call_id. Handles both OpenAI-style (role:"tool", tool_call_id)
+    # and Anthropic-style (role:"user" with tool_result blocks) messages.
+    # No-op if no matching message is found.
+    def attach_to_tool_result(tool_call_id, key, value)
+      msg = @messages.reverse.find do |m|
+        (m[:role] == "tool" && m[:tool_call_id] == tool_call_id) ||
+          (m[:role] == "user" && m[:content].is_a?(Array) &&
+            m[:content].any? { |b| b.is_a?(Hash) && b[:type] == "tool_result" && b[:tool_use_id] == tool_call_id })
+      end
+      msg[key] = value if msg
+      self
+    end
+
     # Mutate the last message matching the predicate lambda in-place.
     # Used by execute_skill_with_subagent to update instruction messages.
     def mutate_last_matching(predicate, &block)

data/lib/clacky/providers.rb

@@ -148,6 +148,18 @@ module Clacky
           "abs-claude-fable-5"    => "abs-claude-opus-4-8",
           "abs-claude-sonnet-4-6" => "abs-claude-sonnet-4-5"
         },
+        # Secondary gateway URL used when the primary base_url is unreachable
+        # after max retries. The model name stays the same — only the endpoint
+        # changes. Nil / absent means no URL fallback for this provider.
+        "fallback_base_url" => "https://llm.1024code.com",
+        # Two selectable endpoints exposed in the Base URL dropdown:
+        #   Primary   — global CDN, lowest latency for most regions.
+        #   Secondary — China-optimised relay (1024code.com), useful when
+        #               the primary is unreachable from mainland China.
+        "endpoint_variants" => [
+          { "label" => "Primary (Global)",  "label_key" => "settings.models.baseurl.variant.openclacky_primary",   "base_url" => "https://api.openclacky.com" }.freeze,
+          { "label" => "Secondary (China)", "label_key" => "settings.models.baseurl.variant.openclacky_secondary", "base_url" => "https://llm.1024code.com"   }.freeze
+        ].freeze,
         "website_url" => "https://www.openclacky.com/ai-keys"
       }.freeze,
 
@@ -336,13 +348,30 @@ module Clacky
         "base_url" => "https://api.xiaomimimo.com/v1",
         "api" => "openai-completions",
         "default_model" => "mimo-v2.5-pro",
-        "models" => ["mimo-v2.5-pro", "mimo-v2-pro", "mimo-v2-omni"],
-        # MiMo-V2-Pro is text-only; MiMo-V2-Omni supports vision (omni = multimodal).
+        # The MiMo-V2 family (mimo-v2-pro / mimo-v2-omni) was retired on
+        # 2026-06-30 and the model ids are no longer accepted by the API. The
+        # current lineup is the V2.5 series:
+        #   - mimo-v2.5-pro: text reasoning flagship, no vision
+        #   - mimo-v2.5: native omni-modal (image/video/audio/text), vision-capable
+        # Source: https://platform.xiaomimimo.com/docs/zh-CN/model
+        "models" => ["mimo-v2.5-pro", "mimo-v2.5"],
         "capabilities" => { "vision" => false }.freeze,
         "model_capabilities" => {
-          "mimo-v2-omni" => { "vision" => true }.freeze
+          "mimo-v2.5" => { "vision" => true }.freeze
         }.freeze,
-        "default_ocr_model" => "mimo-v2-omni",
+        "default_ocr_model" => "mimo-v2.5",
+        # Xiaomi serves the same V2.5 lineup from two billing endpoints: the
+        # pay-as-you-go API (api.xiaomimimo.com) and the Token Plan subscription
+        # endpoint (token-plan-cn.xiaomimimo.com). Both accept identical model
+        # ids and share one capability profile, so a single preset with
+        # endpoint_variants recognises both. Without this, users on the Token
+        # Plan endpoint fell through to "unknown provider" and the conservative
+        # vision=true default applied to every model — sending images to the
+        # text-only mimo-v2.5-pro, which rejects image input.
+        "endpoint_variants" => [
+          { "label" => "Pay-as-you-go", "label_key" => "settings.models.baseurl.variant.mimo_payg",        "base_url" => "https://api.xiaomimimo.com/v1",          "region" => "cn" }.freeze,
+          { "label" => "Token Plan",    "label_key" => "settings.models.baseurl.variant.mimo_token_plan", "base_url" => "https://token-plan-cn.xiaomimimo.com/v1", "region" => "cn" }.freeze
+        ].freeze,
         "website_url" => "https://platform.xiaomimimo.com/"
       }.freeze,
 
@@ -350,8 +379,8 @@ module Clacky
         "name" => "GLM (Z.ai / Zhipu)",
         "base_url" => "https://open.bigmodel.cn/api/paas/v4",
         "api" => "openai-completions",
-        "default_model" => "glm-5.1",
-        "models" => ["glm-5.1", "glm-5", "glm-5-turbo", "glm-5v-turbo", "glm-4.7"],
+        "default_model" => "glm-5.2",
+        "models" => ["glm-5.2", "glm-5.1", "glm-5", "glm-5-turbo", "glm-5v-turbo", "glm-4.7"],
         # Zhipu / Z.ai expose four functionally-equivalent endpoints:
         # two regional sites (mainland open.bigmodel.cn + international api.z.ai)
         # each with a general-billing and a Coding-Plan subpath. They share the
@@ -508,6 +537,22 @@ module Clacky
 
     MEDIA_KINDS = %w[image video audio stt video_understanding].freeze
 
+    # Per-model maximum output token limits.
+    #
+    # The Agent global default (@max_tokens = 16_384) is tuned for the lowest
+    # common denominator. Strong reasoning models (GLM-5.2, Kimi-K3, MiMo-V2.5)
+    # support 64K–128K output but are artificially throttled to 16K, causing
+    # truncation on long reasoning chains and code generation.
+    #
+    # Entries are matched top-to-bottom; the first match wins. Models not
+    # listed here fall back to the global default.
+    MODEL_MAX_OUTPUT = [
+      { pattern: /glm/i,           limit: 65_536 }, # GLM-5.2: 128K output ceiling; 64K ample for reasoning+answer
+      { pattern: /kimi-k3/i,       limit: 65_536 }, # Kimi K3: max_completion_tokens=131072 (max 1M); 64K ample
+      { pattern: /mimo-v2\.5-pro/i, limit: 65_536 }, # MiMo-V2.5-Pro: max_completion_tokens=131072; 64K ample
+      { pattern: /mimo/i,           limit: 32_768 }  # MiMo-V2.5: max_completion_tokens=32768; full default ceiling
+    ].freeze
+
     class << self
       # Check if a provider preset exists
       # @param provider_id [String] The provider identifier (e.g., "anthropic", "openrouter")
@@ -547,6 +592,18 @@ module Clacky
         preset&.dig("api")
       end
 
+      # Resolve the per-model maximum output token limit.
+      # Returns nil when no MODEL_MAX_OUTPUT entry matches — callers should
+      # then fall back to the global default (@max_tokens).
+      #
+      # @param model_name [String] The model name (e.g. "glm-5.2")
+      # @return [Integer, nil] The max output limit, or nil if undeclared
+      def max_output_for(model_name)
+        return nil if model_name.nil? || model_name.to_s.empty?
+        entry = MODEL_MAX_OUTPUT.find { |e| model_name.to_s.match?(e[:pattern]) }
+        entry&.[](:limit)
+      end
+
       # Resolve the API type for a specific provider+model pair.
       #
       # Resolution order:
@@ -779,6 +836,16 @@ module Clacky
         preset&.dig("fallback_models", model)
       end
 
+      # Get the fallback base URL for a provider (used when primary endpoint is unreachable).
+      # Returns nil if the provider has no secondary gateway configured.
+      # @param provider_id [String] The provider identifier
+      # @return [String, nil] The fallback base URL or nil
+      def fallback_base_url(provider_id)
+        preset = PRESETS[provider_id]
+        url = preset&.dig("fallback_base_url")
+        url&.empty? ? nil : url
+      end
+
       # Find provider ID by base URL.
       # Matches if the given URL starts with the provider's base_url (after normalisation),
       # so both exact matches and sub-path variants (e.g. "/v1") are recognised.