quota-keepalive 0.1.7 → 0.1.8

package/README.en.md

@@ -44,8 +44,10 @@ hook. This plugin only coordinates those capabilities and presents the results.
 - Respect provider-level and account-level switches; unsupported or signed-out
   accounts are reported explicitly instead of being silently skipped.
 - Send real-chat keepalive through each provider's signed adapter.
-- Use short probes by default. Long-context probes are opt-in and respect
-  `inputTokens`.
+- Use a real activity conversation by default. For WorkBuddy CN this follows
+  dsh's measured probe shape: about 25k input tokens and one output token, which
+  consumes credits. Use the short probe only when a low-cost connectivity check
+  is enough.
 - Prefer models that recently completed a real keepalive successfully, then fall
   back by success rate, recency, and known failures. Provider model order is used
   when there is no history.
@@ -132,7 +134,9 @@ enable automatic keepalive by default.
 - **Refresh quota**: refresh all enabled providers; each provider row also has a
   provider-only refresh button.
 - **Save schedule**: set the shared interval and action type.
-- **Short probe / Long context**: choose the request shape used by keepalive.
+- **Real activity / Short probe**: real activity fills the configured input and
+  can leave an activity record, but consumes credits. A short probe is cheaper
+  and faster, but does not guarantee an activity record.
 
 Saving a schedule starts the next run after the configured interval. The next run
 time is persisted, and a task already due when Magpie starts is run immediately.
@@ -166,10 +170,12 @@ Keepalive does not blindly choose the cheapest or a free model:
 6. Fall back to the next candidate on a missing model, credit rejection, upstream
    outage, or empty reply.
 
-For WorkBuddy CN, the keepalive path follows dsh probe semantics: an HTTP 2xx means
-the upstream accepted and recorded a real chat, even when a one-token response only
-contains `[DONE]`, usage data, or an empty visible body. JSON business errors, HTML
-login pages, and timeouts are failures.
+For WorkBuddy CN, the real activity path follows dsh probe semantics and sends
+about 25k input tokens with a one-token output cap. Activity mode requires either
+visible model content or usage/credit/prompt-token evidence in the response. A bare
+`[DONE]`, an empty 200, JSON business errors, HTML login pages, and timeouts are
+failures. The short-probe mode remains available for low-cost checks, but it does
+not claim that an activity record was created.
 
 ## Configuration
 
@@ -185,7 +191,7 @@ magpie plugin options quota-keepalive '{
     "codearts": "deepseek-v4.1-flash"
   },
   "modelCandidates": 4,
-  "probeMode": "short",
+  "probeMode": "activity",
   "probeMessage": "Magpie daily connectivity check. Reply with OK.",
   "inputTokens": 25000,
   "maxAttempts": 3,
@@ -200,15 +206,21 @@ magpie plugin options quota-keepalive '{
 - `model`: optional preferred model; it is not a hard allowlist.
 - `modelPreferences`: preferred model per provider.
 - `modelCandidates`: maximum fallback candidates per account, default `4`.
-- `probeMode`: `short` by default; `context` sends the configured long input.
+- `probeMode`: `activity` by default and sends the configured long activity input;
+  `short` sends only a short message; `context` is the backwards-compatible alias
+  for `activity`.
 - `probeMessage`: short probe text, up to 500 characters.
-- `inputTokens`: long-context input size, only used with `probeMode=context`.
+- `inputTokens`: long-input size for `activity`/`context`, default 25000.
 - `maxAttempts`: maximum keepalive attempts per account per Beijing day.
 - `timeoutMs`: overall timeout for one model request, default 90 seconds.
 - `logMaxBytes`: rotate one previous log after this size, default 2 MiB.
 - `statePath`, `logPath`, `snapshotPath`, `commandPath`, `schedulePath`: optional
   absolute paths.
 
+Upgrade note: an old installation that explicitly saved `probeMode: "short"` keeps
+that choice. Select **Real activity** and save the schedule (or update plugin
+options) when you want WorkBuddy activity records to be written.
+
 ## Logs
 
 The default log is:

package/README.md

@@ -34,7 +34,8 @@ quota-keepalive 不接管任何供应商的登录，也不会把页面伪装成
 - 一键签到：通过各供应商登录插件公开的 `auth.checkin` 接口执行，尊重供应商、
   账号两级启停；页面会明确显示未适配或未登录的账号，不会静默跳过。
 - 真实对话保活：通过各供应商自己的登录插件取得签名请求，发送一条真实消息。
-  默认使用“短探活”，只发送一条短消息；需要长上下文时才显式切换“长上下文”模式。
+  默认使用“真实活跃对话”，WorkBuddy CN 会按 dsh 实测语义填充约 25k 输入 token
+  并消耗额度；只想做低成本连通性检查时才切换到“短探活”。
 - 模型选择：只从该账号实际列出的模型中选择。默认优先近期真实保活成功过的模型，
   再按成功率、成功时间和失败次数排序；没有历史记录时沿用供应商自己的模型顺序，
   不再按免费或价格抢先选择。第一个模型不可用时自动换下一档，全部不可用才记失败。
@@ -149,8 +150,8 @@ npm publish --access public --otp=<6 位动态码>
   不会因为快速连续点击触发“每秒只能写入一次”的报错。
 - `刷新额度`按钮：立即刷新全部启用供应商额度；供应商卡片内的刷新按钮只刷新该供应商。
 - `保存计划`：设置统一执行间隔和任务类型。计划默认关闭，防止无意间消耗积分。
-- `短探活/长上下文`：短探活风险、耗时和额度消耗更低；长上下文会按 `inputTokens`
-  填充请求，适合明确需要维持长上下文窗口的场景。
+- `真实活跃对话/短探活`：真实活跃对话按 `inputTokens` 填充请求，能增加上活跃记录
+  的概率并可能消耗积分；短探活风险、耗时和额度消耗更低，但不保证供应商记录为一次对话。
 - 定时计划保存或修改后，从下一个间隔开始执行；下次执行时间会持久化，Magpie 重启或
   电脑休眠后会补跑已经到期的任务。需要立即执行时使用页面顶部的手动按钮。
 
@@ -170,9 +171,10 @@ npm publish --access public --otp=<6 位动态码>
 3. 账号没有记录时，优先该供应商其他账号成功过的模型；仍无记录则跟随供应商模型列表顺序。
 4. 已知失败的模型排在未验证模型之后；单个账号最多尝试配置数量的候选模型。
 5. 模型不存在、额度拒绝、上游不可用或空回复时切换下一个候选。
-6. WorkBuddy CN 按 dsh 的 probe 语义，HTTP 2xx 表示上游已经接受并记录一次真实对话；
-   `max_tokens` 为 1 时可能只有 `[DONE]`、usage 或空正文，仍算成功。JSON 业务错误、
-   HTML 登录页和超时不算成功。
+6. WorkBuddy CN 的真实活跃对话按 dsh 的 probe 语义发送约 25k 输入和 1 个输出 token。
+   活动模式要求响应里出现模型正文，或 usage/credit/prompt token 证据；只有 `[DONE]`、
+   空 200、JSON 业务错误、HTML 登录页和超时都不算成功。短探活仍保留低成本兼容模式，
+   但不会声称已经留下真实活跃记录。
 
 账号卡片会优先显示最近一次真实保活实际使用的模型（`实际模型`）；没有执行历史时，
 依次显示用户指定的 `所选模型` 或自动排序出的 `自动首选`。供应商设置行也会汇总
@@ -196,7 +198,7 @@ magpie plugin options quota-keepalive '{
     "codearts": "deepseek-v4.1-flash"
   },
   "modelCandidates": 4,
-  "probeMode": "short",
+  "probeMode": "activity",
   "probeMessage": "Magpie daily connectivity check. Reply with OK.",
   "inputTokens": 25000,
   "maxAttempts": 3,
@@ -214,9 +216,10 @@ magpie plugin options quota-keepalive '{
   WorkBuddy CN 默认优先 `deepseek-v4.1-flash`，这是 dsh 实测较稳定的真实对话模型；
   其他供应商按近 5 天可用性排序，没有成功记录时按供应商模型顺序选择，不按免费或倍率抢先选择。
 - `modelCandidates`：每个账号每次最多尝试的模型数，默认 4，范围 1–12。
-- `probeMode`：`short`（默认）只发一条短消息；`context` 才使用长上下文填充。
+- `probeMode`：`activity`（默认）发送真实活跃对话，约 25k 输入；`short` 只发短消息；
+  `context` 是旧版配置的兼容别名，行为与 `activity` 相同。
 - `probeMessage`：短探活消息，最多 500 字符；留空时使用默认连通性检查文本。
-- `inputTokens`：仅 `probeMode=context` 时生效，默认 25000，范围 1000–25000。
+- `inputTokens`：`activity`/`context` 模式的长输入规模，默认 25000，范围 1000–25000。
 - `maxAttempts`：每个账号每个北京时间日最多保活尝试轮数，默认 3。
 - `timeoutMs`：单个模型请求的整体截止时间，默认 90 秒，范围 10–180 秒。WorkBuddy
   CN 的 25k 输入请求可能先返回响应头、再继续流式读取；90 秒比 25 秒更接近 dsh 的
@@ -225,6 +228,10 @@ magpie plugin options quota-keepalive '{
 - `logMaxBytes`：单个日志上限，默认 2 MiB，超出后轮转一份 `.1`。
 - `statePath`、`logPath`、`snapshotPath`、`commandPath`、`schedulePath`：可选绝对路径。
 
+升级提示：旧版本如果已经显式保存过 `probeMode: "short"`，新版本会尊重这项用户设置，
+不会静默改成 `activity`。需要点亮 WorkBuddy 活跃记录时，请在页面选择“真实活跃对话”
+并保存计划，或者执行 `magpie plugin options quota-keepalive` 更新配置。
+
 ## 执行周期与失败
 
 Magpie 自己的每日签到调度仍会在启动后约 2 分钟进行首轮，之后约每 30 分钟巡检；
@@ -236,9 +243,9 @@ Magpie 自己的每日签到调度仍会在启动后约 2 分钟进行首轮，
 
 保活失败会按 `maxAttempts` 重试；达到上限后当天不再自动发送。网络失败、登录过期、
 模型不可用和空回复都会进入日志。遇到明确的 HTTP 402/429 额度耗尽或免费额度用尽时，
-当天停止自动重试，避免持续发送无效请求。手动“发送保活”或账号级“执行保活”可以重试
-当天失败或额度耗尽的账号；签到或保活当天已经成功时，即使手动触发也会跳过，不会重复
-发送。供应商整体停用时，账号级手动按钮也会禁用。
+当天停止自动重试，避免持续发送无效请求。当天已经成功后，自动任务和手动触发都会
+跳过，不会重复发送；手动“发送保活”只能在当天尚未成功时重试失败账号。供应商整体
+停用时，账号级手动按钮也会禁用。
 
 ## 日志
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "quota-keepalive",
-  "version": "0.1.7",
+  "version": "0.1.8",
   "description": "Unofficial multi-provider quota inventory, check-in, keepalive scheduling, history, and quota trends for Magpie",
   "keywords": [
     "magpie",

package/src/config.js

@@ -19,7 +19,7 @@ export const DEFAULT_KEEPALIVE = Object.freeze({
   model: "",
   modelCandidates: 4,
   inputTokens: 25_000,
-  probeMode: "short",
+  probeMode: "activity",
   probeMessage: DEFAULT_PROBE_MESSAGE,
   maxAttempts: 3,
   scheduleEnabled: false,
@@ -75,7 +75,9 @@ export function resolveKeepaliveOptions(raw) {
     scheduleMinutes: integerIn(source.scheduleMinutes, DEFAULT_KEEPALIVE.scheduleMinutes, 5, 10_080),
     configHome: absolutePath(source.configHome, undefined),
     inputTokens: integerIn(source.inputTokens, DEFAULT_KEEPALIVE.inputTokens, 1_000, 25_000),
-    probeMode: source.probeMode === "context" ? "context" : DEFAULT_KEEPALIVE.probeMode,
+    probeMode: ["short", "context", "activity"].includes(source.probeMode)
+      ? source.probeMode
+      : DEFAULT_KEEPALIVE.probeMode,
     probeMessage: probeMessage || DEFAULT_PROBE_MESSAGE,
     maxAttempts: integerIn(source.maxAttempts, DEFAULT_KEEPALIVE.maxAttempts, 1, 10),
     timeoutMs: integerIn(source.timeoutMs, DEFAULT_KEEPALIVE.timeoutMs, 10_000, 180_000),

package/src/keepalive.js

@@ -4,6 +4,9 @@ const ORIGIN = "https://copilot.tencent.com"
 const CHAT_PATH = "/v2/chat/completions"
 const FILLER = "The quick brown fox jumps over the lazy dog. "
 const CHARS_PER_TOKEN = 4.5
+const DEFAULT_ACTIVITY_INPUT_TOKENS = 25_000
+const ACTIVITY_PROMPT = "Reply with a single word."
+const ACTIVITY_SYSTEM_PROMPT = "You are a connectivity check. Reply with a single word."
 export const DEFAULT_PROBE_MESSAGE = "Magpie daily connectivity check. Reply with OK."
 
 function probeMessage(value) {
@@ -18,18 +21,26 @@ export function buildKeepaliveBody({
   probeMessage: message = "",
 }) {
   const prompt = `${probeMessage(message)} Today is ${day}.`
+  const activity = probeMode === "activity"
+  const context = activity || probeMode === "context"
   let content = prompt
-  if (probeMode === "context") {
-    const targetChars = Math.ceil(inputTokens * CHARS_PER_TOKEN)
+  let system = "You are a concise connectivity check. Reply with a single word."
+  if (context) {
+    const tokens = Number(inputTokens) || (activity ? DEFAULT_ACTIVITY_INPUT_TOKENS : 0)
+    const targetChars = Math.ceil(tokens * CHARS_PER_TOKEN)
     const filler = FILLER.repeat(Math.ceil(targetChars / FILLER.length))
-    content = `${filler}${prompt}`
+    // WorkBuddy's CN gateway measures a real request by its input size. The
+    // activity shape mirrors dsh's probe body so the request is recorded as
+    // a conversation instead of a short connectivity check.
+    content = activity ? `${filler}${ACTIVITY_PROMPT}` : `${filler}${prompt}`
+    if (activity) system = ACTIVITY_SYSTEM_PROMPT
   }
   return {
     model,
     messages: [
       {
         role: "system",
-        content: "You are a concise connectivity check. Reply with a single word.",
+        content: system,
       },
       {
         role: "user",
@@ -53,6 +64,64 @@ function parsedJson(value) {
   }
 }
 
+function numberValue(value) {
+  if (typeof value === "number") return Number.isFinite(value) ? value : undefined
+  if (typeof value !== "string" || !value.trim()) return undefined
+  const n = Number(value.trim())
+  return Number.isFinite(n) ? n : undefined
+}
+
+function hasKey(value, name) {
+  return object(value) && Object.prototype.hasOwnProperty.call(value, name)
+}
+
+function numberField(value, names) {
+  for (const name of names) {
+    if (!hasKey(value, name)) continue
+    const n = numberValue(value[name])
+    if (n !== undefined) return n
+  }
+  return undefined
+}
+
+function usageEvidence(payloads) {
+  const result = {
+    sawUsage: false,
+    credit: undefined,
+    promptTokens: undefined,
+    completionTokens: undefined,
+    totalTokens: undefined,
+  }
+  for (const payload of Array.isArray(payloads) ? payloads : []) {
+    const candidates = [
+      payload,
+      payload?.usage,
+      payload?.data?.usage,
+      payload?.data,
+    ].filter(object)
+    for (const candidate of candidates) {
+      const credit = numberField(candidate, ["credit", "credits"])
+      const promptTokens = numberField(candidate, ["prompt_tokens", "promptTokens", "input_tokens", "inputTokens"])
+      const completionTokens = numberField(candidate, [
+        "completion_tokens",
+        "completionTokens",
+        "output_tokens",
+        "outputTokens",
+      ])
+      const totalTokens = numberField(candidate, ["total_tokens", "totalTokens"])
+      if ([credit, promptTokens, completionTokens, totalTokens].every((value) => value === undefined)) {
+        continue
+      }
+      result.sawUsage = true
+      result.credit ??= credit
+      result.promptTokens ??= promptTokens
+      result.completionTokens ??= completionTokens
+      result.totalTokens ??= totalTokens
+    }
+  }
+  return result
+}
+
 function businessFailure(body) {
   if (!object(body)) return ""
   if (body.success === false) {
@@ -71,6 +140,26 @@ function businessFailure(body) {
   return ""
 }
 
+function contentValue(value) {
+  if (typeof value === "string") return value
+  if (Array.isArray(value)) return value.map((part) => contentValue(part?.text)).join("")
+  return ""
+}
+
+function payloadReply(payload) {
+  const choices = Array.isArray(payload?.choices) ? payload.choices : []
+  for (const choice of choices) {
+    if (!object(choice)) continue
+    const value = choice.delta?.content ?? choice.message?.content ?? choice.text
+    const content = contentValue(value)
+    if (content) return content
+  }
+  const blocks = Array.isArray(payload?.content) ? payload.content : []
+  const blockText = blocks.map((block) => contentValue(block?.text)).join("")
+  if (blockText) return blockText
+  return contentValue(payload?.output_text)
+}
+
 export function responseMessage(raw) {
   const body = parsedJson(raw)
   if (object(body)) {
@@ -107,9 +196,10 @@ export function assistantReply(raw) {
         choice.delta?.content ??
         choice.message?.content ??
         choice.text
-      if (typeof value === "string" && value) {
+      const piece = contentValue(value)
+      if (piece) {
         sawReply = true
-        content += value
+        content += piece
       }
     }
   }
@@ -121,7 +211,7 @@ export function assistantReply(raw) {
   }
 }
 
-export function workbuddyProbeResult(response, raw) {
+export function workbuddyProbeResult(response, raw, { probeMode = "short" } = {}) {
   const source = text(raw)
   const body = source.trim()
   const contentType = text(response?.headers?.get?.("content-type")).toLowerCase()
@@ -153,13 +243,37 @@ export function workbuddyProbeResult(response, raw) {
     }
   }
 
-  // WorkBuddy can accept a real conversation without emitting assistant text:
-  // max_tokens is one and the signed gateway may finish with only [DONE].
-  // dsh treats HTTP 2xx as a successful probe for that reason.
   const reply = assistantReply(source)
+  let content = reply.content
+  let hasReply = reply.ok
+  for (const payload of payloads) {
+    if (content) break
+    const direct = payloadReply(payload)
+    if (direct) {
+      content = direct
+      hasReply = true
+      break
+    }
+  }
+  const evidence = usageEvidence(payloads)
+  const activity = probeMode === "activity" || probeMode === "context"
+  const activityVerified = activity && (hasReply || evidence.sawUsage)
+  if (activity && !activityVerified) {
+    return {
+      ok: false,
+      content,
+      activityVerified: false,
+      error: "WorkBuddy's keepalive chat returned 200 without a model reply or usage evidence; no active conversation was confirmed",
+    }
+  }
   return {
     ok: true,
-    content: reply.content,
+    content: sanitize(content, 120),
+    activityVerified,
+    ...(evidence.credit === undefined ? {} : { credit: evidence.credit }),
+    ...(evidence.promptTokens === undefined ? {} : { promptTokens: evidence.promptTokens }),
+    ...(evidence.completionTokens === undefined ? {} : { completionTokens: evidence.completionTokens }),
+    ...(evidence.totalTokens === undefined ? {} : { totalTokens: evidence.totalTokens }),
   }
 }
 
@@ -168,7 +282,7 @@ export async function sendKeepalive({
   origin = ORIGIN,
   model,
   inputTokens,
-  probeMode,
+  probeMode = "activity",
   probeMessage,
   timeoutMs,
   day,
@@ -207,12 +321,17 @@ export async function sendKeepalive({
     error.retryToday = !terminal
     throw error
   }
-  const probe = workbuddyProbeResult(response, raw)
+  const probe = workbuddyProbeResult(response, raw, { probeMode })
   if (!probe.ok) {
     throw new Error(probe.error)
   }
   return {
     model,
     reply: probe.content,
+    activityVerified: probe.activityVerified,
+    ...(probe.credit === undefined ? {} : { credit: probe.credit }),
+    ...(probe.promptTokens === undefined ? {} : { promptTokens: probe.promptTokens }),
+    ...(probe.completionTokens === undefined ? {} : { completionTokens: probe.completionTokens }),
+    ...(probe.totalTokens === undefined ? {} : { totalTokens: probe.totalTokens }),
   }
 }

package/src/log.js

@@ -230,6 +230,7 @@ export function formatLogRecord(record) {
     ["probeMode", record.probeMode],
     ["inputTokens", record.inputTokens],
     ["credit", record.credit],
+    ["activityVerified", record.activityVerified],
     ["streak", record.streak],
     ["durationMs", record.durationMs],
     ["retry", record.retry],

package/src/provider-chat.js

@@ -56,7 +56,7 @@ export function anthropicBody({
   const message = sanitize(probeMessage, 500).trim() || DEFAULT_PROBE_MESSAGE
   const prompt = `${message} Today is ${day}.`
   let content = prompt
-  if (probeMode === "context") {
+  if (probeMode === "context" || probeMode === "activity") {
     const targetChars = Math.ceil(inputTokens * CHARS_PER_TOKEN)
     const filler = FILLER.repeat(Math.ceil(targetChars / FILLER.length))
     content = `${filler}${prompt}`

package/src/snapshot.js

@@ -53,7 +53,7 @@ export function publicRecord(record) {
     if (value) out[name] = value
   }
   if (out.account && !ACCOUNT.test(out.account)) delete out.account
-  for (const name of ["attempts", "inputTokens", "credit", "streak", "durationMs", "retry", "persisted", "keepalive"]) {
+  for (const name of ["attempts", "inputTokens", "credit", "streak", "durationMs", "retry", "persisted", "keepalive", "activityVerified"]) {
     const value = record?.[name]
     if (value !== undefined && value !== null && (typeof value === "number" || typeof value === "boolean")) {
       out[name] = value

package/src/workbuddy.js

@@ -267,7 +267,8 @@ export function authIdentity(auth, key) {
 }
 
 export function resultFields(result, config, model = "") {
-  const inputTokens = config?.probeMode === "context" ? Number(config.inputTokens) || 0 : 0
+  const longProbe = config?.probeMode === "context" || config?.probeMode === "activity"
+  const configuredInputTokens = longProbe ? Number(config.inputTokens) || 0 : 0
   switch (result?.kind) {
     case "skipped":
       if (result.reason === "done-today") {
@@ -313,6 +314,8 @@ export function resultFields(result, config, model = "") {
       }
     case "sent":
       const terminal = result.cause?.retryToday === false
+      const reply = result.result || {}
+      const inputTokens = Number(reply.promptTokens) || configuredInputTokens
       return {
         status: result.ok ? "success" : "failed",
         outcome: result.ok
@@ -323,8 +326,11 @@ export function resultFields(result, config, model = "") {
               ? "retry"
               : "exhausted",
         attempts: result.attempts,
-        model: result.result?.model || model,
+        model: reply.model || model,
+        probeMode: config?.probeMode,
         inputTokens,
+        ...(reply.credit === undefined ? {} : { credit: reply.credit }),
+        ...(reply.activityVerified === undefined ? {} : { activityVerified: reply.activityVerified }),
         retry: !result.ok && !terminal && result.attempts < config.maxAttempts,
         persisted: result.persisted !== false,
         error: result.error,
@@ -528,17 +534,21 @@ export function createProviderKeepalive(options = {}) {
           run,
           model: candidates[0].id,
           models: candidates.map((candidate) => candidate.id).join(","),
-          inputTokens: config.probeMode === "context" ? config.inputTokens : 0,
+          inputTokens: ["context", "activity"].includes(config.probeMode) ? config.inputTokens : 0,
           probeMode: config.probeMode,
           keepalive: true,
         })
 
+        const activityMode = config.probeMode === "short" ? "short" : "activity"
         const result = await runDailyOnce({
           statePath: config.statePath,
-          account: accountTag(`${provider}|${identity}`),
+          // A legacy short probe and a real activity conversation are not the
+          // same task. Keeping them separate lets a user switch to activity
+          // after a short probe has already been recorded today.
+          account: accountTag(`${provider}|${identity}|keepalive:${activityMode}`),
           day,
           model: candidates[0].id,
-          inputTokens: config.probeMode === "context" ? config.inputTokens : 0,
+          inputTokens: ["context", "activity"].includes(config.probeMode) ? config.inputTokens : 0,
           probeMode: config.probeMode,
           maxAttempts: config.maxAttempts,
           lockWaitMs: config.lockWaitMs,

package/ui/app.js

@@ -3,7 +3,7 @@
   const ACTIONS = {
     detect: { title: "检测账号", detail: "正在读取 Magpie 当前已登录的供应商与账号。" },
     checkin: { title: "全部签到", detail: "正在让已有签到能力的供应商执行签到。" },
-    keepalive: { title: "发送保活", detail: "正在为全部已适配供应商账号发送真实对话。" },
+    keepalive: { title: "发送保活", detail: "正在为全部已适配供应商账号发送真实活跃对话；会消耗额度。" },
     refresh: { title: "刷新额度", detail: "正在刷新全部启用供应商的实时额度。" },
   }
   const EVENT_NAMES = {
@@ -995,9 +995,12 @@
       ["供应商", record?.provider],
       ["账号", record?.account],
       ["模型", record?.model],
-      ["保活方式", record?.probeMode === "context" ? "长上下文" : record?.probeMode === "short" ? "短探活" : ""],
+      ["保活方式", record?.probeMode === "activity" || record?.probeMode === "context"
+        ? "真实活跃对话"
+        : record?.probeMode === "short" ? "短探活" : ""],
       ["输入", number(record?.inputTokens) ? `${formatNumber(record.inputTokens)} tokens` : ""],
       ["本次积分", number(record?.credit) ? `+${formatNumber(record.credit)}` : ""],
+      ["活跃确认", record?.activityVerified === true ? "已验证" : ""],
       ["连续签到", number(record?.streak) ? `${formatNumber(record.streak)} 天` : ""],
       ["耗时", durationText(record?.durationMs)],
       ["日期", record?.day],
@@ -1192,7 +1195,9 @@
       keepalive: "真实保活",
       all: "签到 + 真实保活",
     }[options.scheduleAction] || "签到"
-    const probe = options.probeMode === "context" ? "长上下文保活" : "短探活"
+    const probe = options.probeMode === "activity" || options.probeMode === "context"
+      ? "真实活跃对话（会消耗积分）"
+      : "短探活（不保证活跃记录）"
     return `每 ${number(options.scheduleMinutes) || 60} 分钟执行一次${action}；${probe}。`
   }
 
@@ -1211,7 +1216,7 @@
       $("schedule-action").value = ["checkin", "keepalive", "all"].includes(options.scheduleAction)
         ? options.scheduleAction
         : "checkin"
-      $("probe-mode").value = options.probeMode === "context" ? "context" : "short"
+      $("probe-mode").value = options.probeMode === "short" ? "short" : "activity"
       $("schedule-status").textContent = scheduleDescription(options)
       if (showToast) toast("定时计划已保存")
     } catch (error) {

package/ui/index.html

@@ -112,10 +112,10 @@
           </svg>
         </label>
         <label class="select-field probe-field"
-               title="短探活只发送一条短消息；长上下文会填充到配置的输入规模">
+               title="真实活跃对话会发送约 25k 输入并消耗积分；短探活只做低成本连通性检查">
           <select id="probe-mode" aria-label="保活方式">
+            <option value="activity">真实活跃对话</option>
             <option value="short">短探活</option>
-            <option value="context">长上下文</option>
           </select>
           <svg viewBox="0 0 16 16" fill="none" stroke="currentColor" stroke-width="1.5"
                stroke-linecap="round" stroke-linejoin="round" aria-hidden="true">
@@ -167,7 +167,7 @@
     <section class="confirm-card" role="dialog" aria-modal="true" aria-labelledby="confirm-title">
       <p class="eyebrow">REAL REQUEST</p>
       <h2 id="confirm-title">发送真实保活消息？</h2>
-      <p id="confirm-detail">这会为全部已适配供应商的账号发送真实对话，优先尝试近期验证可用且能成功回复的模型；仍可能消耗积分。</p>
+      <p id="confirm-detail">这会为全部已适配供应商的账号发送真实活跃对话，优先尝试近期验证可用且能成功回复的模型；请求可能消耗积分。</p>
       <div class="confirm-actions">
         <button class="button" id="confirm-cancel" type="button">取消</button>
         <button class="button caution" id="confirm-send" type="button">确认发送</button>