@huai_xin/dsh-workbuddy 0.1.5 → 0.1.6

package/README.md

@@ -350,8 +350,22 @@ node test/live/stream-cut-probe.mjs cn --selftest     # 离线验证"断流"分
 node test/live/stream-cut-probe.mjs intl 10 --long    # 真实压测断流率（免费模型零消耗）
 node test/live/find-message.mjs "关键词" <会话前缀>     # 按文本定位某一步，并打出前后事件
 node test/live/dump-stream.mjs <会话前缀> <seq,…>      # 打印那一步存下来的原始流帧
+node test/live/sampling-probe.mjs intl 4              # 同一提示在各 temperature 下的"提前 stop"率
 ```
 
+### 采样参数（默认不发，可按需覆盖）
+
+插件默认**不发送** `temperature` / `top_p` / `top_k`（harness 没暴露温度设置，`options.temperature` 实际为空），因此走上游默认采样。`stop` 本质是模型采到 EOS，偏热的默认值更容易在"现在我来改……："这种地方收尾，所以留了两个覆盖开关：
+
+```powershell
+$env:WORKBUDDY_TEMPERATURE = '0.3'
+$env:WORKBUDDY_TOP_P = '0.95'
+```
+
+优先级：**调用方显式传入的 `temperature`** > 环境变量 > 不发 ✓（默认行为对所有人不变 ✓）。改完需重启 DSH ✓。
+
+> 实测（2026-10-22，海外版免费档 `deepseek-v4.1-flash`）：一个明确要求调用工具的短提示，在"不发 / 0.2 / 0.5 / 1.0"四档各跑 4 次，**16/16 都正确调用了工具** —— 单纯温度解释不了真实会话里的提前收尾（那次输入有 213k 缓存 token，`effort=max` 却只有 0 个思考 token，更像长上下文下的例行续写）。这个开关是留给真实流量验证用的，默认不改输出风格。
+
 ## 许可
 
 MIT，见 [`LICENSE`](LICENSE)。再次说明：本项目为第三方非官方插件，与腾讯 / WorkBuddy 无关联。

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@huai_xin/dsh-workbuddy",
-  "version": "0.1.5",
+  "version": "0.1.6",
   "description": "WorkBuddy model provider for DeepSeek Harness (international + China deployments)",
   "type": "module",
   "main": "./src/index.js",

package/src/adapter.js

@@ -651,7 +651,20 @@ export class WorkBuddyAdapter {
       function: { name: tool.name, description: tool.description, parameters: tool.parameters }
     }))
     if (tools !== undefined && tools.length > 0) body.tools = tools
-    if (options.temperature !== undefined) body.temperature = options.temperature
+    // Sampling is whatever the caller asked for, else the deployment default —
+    // except for an explicit override. `stop` is the model sampling its
+    // end-of-sequence token, so on a hot default it can end a turn right after
+    // "现在我来改……：" instead of calling the tool, and nothing in the harness
+    // exposes a temperature to tune that. These two variables let an operator try a
+    // cooler profile without changing the default behaviour:
+    //   WORKBUDDY_TEMPERATURE=0.3  WORKBUDDY_TOP_P=0.95
+    // Measured 2026-10-22: a short tool-using prompt called the tool 16/16 times at
+    // the default and at 0.2/0.5/1.0, so sampling alone does not explain the early
+    // stops; the override exists to test the long-context case on real traffic.
+    const temperature = options.temperature ?? envNumber('WORKBUDDY_TEMPERATURE')
+    if (temperature !== undefined) body.temperature = temperature
+    const topP = envNumber('WORKBUDDY_TOP_P')
+    if (topP !== undefined) body.top_p = topP
     if (options.stop !== undefined && options.stop.length > 0) body.stop = options.stop
     const maxTokens = positiveInt(options.maxTokens) ?? entry?.maxTokens
     if (maxTokens !== undefined) body.max_tokens = maxTokens
@@ -1027,6 +1040,14 @@ function positiveInt(value) {
   return typeof value === 'number' && Number.isSafeInteger(value) && value > 0 ? value : undefined
 }
 
+/** A numeric override from the environment, or `undefined` when unset/unusable. */
+function envNumber(name) {
+  const raw = process.env[name]
+  if (typeof raw !== 'string' || raw.trim().length === 0) return undefined
+  const value = Number(raw)
+  return Number.isFinite(value) ? value : undefined
+}
+
 /**
  * Describe a stream that ended without its terminator.
  *