@dan-ai-studio/dshopencodego 0.1.67 → 0.1.68

package/README.en.md

@@ -146,10 +146,11 @@ The API key comes from the Harness credential store (reference name `OPENCODE_GO
 - **Two entries, one page**:
   - **A sidebar icon → a full page in the main column**: every request across sessions. The log is a process-wide thing, so the sidebar is where it belongs — a fixed, always-visible spot, and the same pattern the Harness's own plugin manager uses.
   - **A "Request log" View inside the conversation**: this session's requests, in place — answering "why was that turn slow" without leaving the chat.
-- It is a page, not a floating card: the head, the summary (requests / failures / mean latency / mean first token / mean tok/s / mean cache hit / connection reuse) and the filters stay put while the table scrolls. Each row carries time, model, session, input/output/cache, **cache hit**, **tok/s**, first token, latency and status; clicking one expands its HTTP detail (method, URL, redacted headers, request shape, result and usage). Filters: model, result, slow requests, session; groups: session, model, result; **click a column head to sort by it**, click again to flip.
+- It is a page, not a floating card: the head, the summary (requests / failures / mean latency / mean first token / mean tok/s / mean cache hit / connection reuse) and the filters stay put while the table scrolls. Each row carries time, model, session, input/output/cache, **cache hit**, **tok/s**, first token, latency and status; clicking one expands its HTTP detail (method, protocol, URL, body size, redacted headers, request shape, result and usage, reasoning split, cache writes, per-request cost, content span). Filters: model, result, slow requests, session; groups: session, model, result; **click a column head to sort by it**, click again to flip.
 - **tok/s** is the output rate over the decode window (first token to the end), on the same scale as the Harness's own per-turn reading (reasoning tokens included). A request that produced no token has no rate — none is invented. **A burst delivery has no rate either**: when the gateway holds a whole turn and flushes it at the end (common with reasoning models: 8 seconds of thinking invisible on the wire, then a hundred tokens in a few milliseconds), a sub-100ms content span is delivery shape, not generation speed — it shows as — and stays out of the mean.
 - The session column names the **session title** (from the session catalog; falls back to the id, whose full spelling stays one hover away). The session filter matches titles and ids; session groups are headed the same way.
 - **Cache hit** is the share of input tokens served from cache (cache reads over uncached input plus cache reads). No reported input means no share — none is invented either.
+- **The In column is uncached new input**: the gateway reports a prompt as "uncached plus cache hits" and the log records only the new part — a fully prefix-cached turn reads as a tiny input with a 100% hit, which is not a bug. Hover a column head for its definition.
 - Session ids are **always stored whole**: an id is a conversation identity, not a credential, and a truncated one reads two conversations as one.
 - The log is a ring buffer (oldest dropped first) that dies with the process; snapshots are read on demand, never polled, so a page nobody opens costs no traffic.
 - The credential never enters a record: the recorder takes no headers argument at all, only the allowlisted session id and user agent. Failed requests are recorded with their error code; a request cut off before any terminal event is not recorded — no total reported, none invented.

package/README.md

@@ -145,10 +145,11 @@ API Key 通过 Harness 凭证库提供（引用名 `OPENCODE_GO_API_KEY`），
 - **两个入口，同一个页面**：
   - **侧边栏图标 → 主区域整页**：跨会话看全部请求。日志是进程级的数据，放在侧边栏最合理——位置固定、常驻可见，和 DSH 自己的插件管理器同一个范式。
   - **会话里的「请求日志」标签页**：就地看**本会话**的请求，回答"刚才那轮为什么慢"不用离开对话。
-- 页面是一整页而不是浮层：标题、统计（请求数／失败数／平均耗时／平均首字／平均 tok/s／平均缓存命中／连接复用率）、筛选栏固定在上方，只有表格自己滚动。每行含时间、模型、会话、输入／输出／缓存、**缓存命中**、**tok/s**、首字、耗时与状态；点一行展开 HTTP 明细（方法、地址、脱敏后的请求头、请求构成、结果与用量）。筛选有模型／结果／慢请求／会话，分组有会话／模型／结果；**点列头按该列排序**，再点一次反向。
+- 页面是一整页而不是浮层：标题、统计（请求数／失败数／平均耗时／平均首字／平均 tok/s／平均缓存命中／连接复用率）、筛选栏固定在上方，只有表格自己滚动。每行含时间、模型、会话、输入／输出／缓存、**缓存命中**、**tok/s**、首字、耗时与状态；点一行展开 HTTP 明细（方法、协议、地址、请求体大小、脱敏后的请求头、请求构成、结果与用量、推理拆分、缓存写入、单轮花费、内容跨度）。筛选有模型／结果／慢请求／会话，分组有会话／模型／结果；**点列头按该列排序**，再点一次反向。
 - **tok/s** 是**首字到结束**这段解码窗口上的输出速率，口径与 Harness 自己的每轮读数一致（含推理 token）。没有 token 的请求没有速率——不编数。**突发推送同样没有速率**：网关把一整轮攒到最后一次性推过来时（推理模型常见：8 秒思考在 wire 上不可见，然后上百个 token 几毫秒内到达），内容跨度不足 100ms，算出来的"每秒几万 token"不是生成速度——显示为 —，也不计入平均。
 - 会话列显示**会话标题**（标题来自会话目录；拿不到时回退显示 id，悬停可见完整 id）。会话筛选同时匹配标题与 id；按会话分组的组头同样显示标题。
 - **缓存命中**是输入 token 里走缓存的占比（缓存读取／（未缓存输入＋缓存读取））。网关没报输入就没有占比——同样不编数。
+- **输入列是未缓存的新增 token**：网关把整段 prompt 拆成"未缓存＋缓存命中"两部分上报，输入列只记新增的——同一会话前缀缓存全中时输入很小、缓存命中 100%，不是 bug。悬停列头有口径说明。
 - 会话 id **始终完整记录**（它是会话身份，不是凭据；截断后两个会话会读成一个）。
 - 记录的是环形缓冲（超限丢最旧），进程重启即清；只读快照，不轮询，没人打开的页面零流量。
 - 凭据永远不进记录：记录器根本不接收请求头参数，只收白名单里的会话 id 与 UA；失败的请求同样记一条（含错误码），中断到没有终端事件的不记（网关没报总数就不编数）。

package/lib/client.js

@@ -233,10 +233,14 @@ function parseEntry(value) {
   const row = value;
   const shape = row["shape"];
   const headers = row["headers"];
+  const protocol = row["protocol"];
   return {
     seq: count("seq", row),
     atMs: count("atMs", row),
     model: text("model", row),
+    // Older rows predate the protocol recording; they read without one
+    // rather than failing the whole snapshot.
+    ...protocol === "anthropic-messages" || protocol === "openai-completions" || protocol === "openai-responses" ? { protocol } : {},
     session: text("session", row),
     method: text("method", row),
     url: text("url", row),
@@ -244,9 +248,14 @@ function parseEntry(value) {
     ...row["inputTokens"] === void 0 ? {} : { inputTokens: count("inputTokens", row) },
     ...row["outputTokens"] === void 0 ? {} : { outputTokens: count("outputTokens", row) },
     ...row["cacheReadTokens"] === void 0 ? {} : { cacheReadTokens: count("cacheReadTokens", row) },
+    ...row["cacheWriteTokens"] === void 0 ? {} : { cacheWriteTokens: count("cacheWriteTokens", row) },
+    ...row["reasoningTokens"] === void 0 ? {} : { reasoningTokens: count("reasoningTokens", row) },
+    ...row["costUsd"] === void 0 ? {} : { costUsd: rate("costUsd", row) },
+    ...row["requestBytes"] === void 0 ? {} : { requestBytes: count("requestBytes", row) },
     ...row["durationMs"] === void 0 ? {} : { durationMs: count("durationMs", row) },
     ...row["firstTokenMs"] === void 0 ? {} : { firstTokenMs: count("firstTokenMs", row) },
     ...row["decodeMs"] === void 0 ? {} : { decodeMs: count("decodeMs", row) },
+    ...row["streamMs"] === void 0 ? {} : { streamMs: count("streamMs", row) },
     ...row["tps"] === void 0 ? {} : { tps: rate("tps", row) },
     ok: row["ok"] === true,
     ...typeof row["code"] === "string" ? { code: row["code"] } : {},
@@ -536,22 +545,31 @@ var en = {
   requestLogColModel: "Model",
   requestLogColSession: "Session",
   requestLogColIn: "In",
+  requestLogColInHint: "Uncached input tokens: what this turn added beyond the cache",
   requestLogColOut: "Out",
+  requestLogColOutHint: "Generated output tokens",
   requestLogColCache: "Cached",
+  requestLogColCacheHint: "Input tokens served from the cache",
   requestLogColCacheRate: "Cache hit",
   requestLogColLatency: "Latency",
   requestLogColTps: "tok/s",
+  requestLogColStream: "Content span",
   requestLogColFirstToken: "First token",
   requestLogColStatus: "Status",
   requestLogDetailRequest: "Request",
   requestLogDetailTiming: "Timing",
   requestLogDetailResult: "Result",
   requestLogDetailMethod: "Method",
+  requestLogDetailProtocol: "Protocol",
   requestLogDetailUrl: "URL",
+  requestLogDetailRequestBytes: "Request body",
   requestLogDetailSession: "Session",
   requestLogDetailPool: "Pool",
   requestLogDetailError: "Error",
   requestLogDetailTokens: "Tokens",
+  requestLogDetailReasoning: "Reasoning",
+  requestLogDetailCacheWrite: "Cache write",
+  requestLogDetailCost: "Cost",
   requestLogDetailHeaders: "Headers",
   requestLogDetailUserAgent: "User-Agent",
   requestLogDetailShape: "Shape",
@@ -716,22 +734,31 @@ var zh = {
   requestLogColModel: "\u6A21\u578B",
   requestLogColSession: "\u4F1A\u8BDD",
   requestLogColIn: "\u8F93\u5165",
+  requestLogColInHint: "\u672A\u7F13\u5B58\u7684\u8F93\u5165 token\uFF1A\u672C\u8F6E\u65B0\u589E\u7684\uFF0C\u4E0D\u542B\u7F13\u5B58\u547D\u4E2D\u7684\u90E8\u5206",
   requestLogColOut: "\u8F93\u51FA",
+  requestLogColOutHint: "\u751F\u6210\u7684\u8F93\u51FA token",
   requestLogColCache: "\u7F13\u5B58",
+  requestLogColCacheHint: "\u547D\u4E2D\u7F13\u5B58\u7684\u8F93\u5165 token",
   requestLogColCacheRate: "\u7F13\u5B58\u547D\u4E2D",
   requestLogColLatency: "\u8017\u65F6",
   requestLogColTps: "tok/s",
+  requestLogColStream: "\u5185\u5BB9\u8DE8\u5EA6",
   requestLogColFirstToken: "\u9996\u5B57",
   requestLogColStatus: "\u72B6\u6001",
   requestLogDetailRequest: "\u8BF7\u6C42",
   requestLogDetailTiming: "\u8017\u65F6",
   requestLogDetailResult: "\u7ED3\u679C",
   requestLogDetailMethod: "\u65B9\u6CD5",
+  requestLogDetailProtocol: "\u534F\u8BAE",
   requestLogDetailUrl: "\u5730\u5740",
+  requestLogDetailRequestBytes: "\u8BF7\u6C42\u4F53",
   requestLogDetailSession: "\u4F1A\u8BDD",
   requestLogDetailPool: "\u8FDE\u63A5\u6C60",
   requestLogDetailError: "\u9519\u8BEF",
   requestLogDetailTokens: "Token",
+  requestLogDetailReasoning: "\u63A8\u7406",
+  requestLogDetailCacheWrite: "\u7F13\u5B58\u5199\u5165",
+  requestLogDetailCost: "\u82B1\u8D39",
   requestLogDetailHeaders: "\u8BF7\u6C42\u5934",
   requestLogDetailUserAgent: "User-Agent",
   requestLogDetailShape: "\u8BF7\u6C42\u6784\u6210",
@@ -2101,6 +2128,16 @@ function duration(value) {
 function rate2(value) {
   return value === void 0 ? "\u2014" : value.toFixed(1);
 }
+function cost(value) {
+  if (value === void 0) return "\u2014";
+  return `$${value.toFixed(value < 0.01 ? 6 : 4)}`;
+}
+function bytes(value) {
+  if (value === void 0) return "\u2014";
+  if (value < 1024) return `${value} B`;
+  if (value < 1024 * 1024) return `${(value / 1024).toFixed(1)} KB`;
+  return `${(value / (1024 * 1024)).toFixed(2)} MB`;
+}
 function cacheRate(entry) {
   if (entry.inputTokens === void 0) return void 0;
   const cached = entry.cacheReadTokens ?? 0;
@@ -2231,9 +2268,9 @@ function RequestLogPage({ readLog, enabled, t, getLocale, sessionId, useSessions
     return value === void 0 ? [] : [value];
   }));
   const reuse = snapshot !== void 0 && snapshot.totalRecords > 0 ? Math.max(0, Math.min(1, 1 - snapshot.newConnections / snapshot.totalRecords)) : void 0;
-  const head = (key, label) => /* @__PURE__ */ (0, import_jsx_runtime2.jsxs)("th", { className: request_log_default.thSortable, onClick: () => {
+  const head = (key, label, hint) => /* @__PURE__ */ (0, import_jsx_runtime2.jsxs)("th", { className: request_log_default.thSortable, onClick: () => {
     toggleSort(key);
-  }, children: [
+  }, title: hint, children: [
     label,
     sortKey === key ? sortDir === "asc" ? " \u25B2" : " \u25BC" : ""
   ] });
@@ -2282,10 +2319,18 @@ function RequestLogPage({ readLog, enabled, t, getLocale, sessionId, useSessions
               /* @__PURE__ */ (0, import_jsx_runtime2.jsx)("span", { className: request_log_default.detailLabel, children: t("requestLogDetailMethod") }),
               /* @__PURE__ */ (0, import_jsx_runtime2.jsx)("span", { className: `${request_log_default.detailValue} ${request_log_default.mono}`, children: entry.method })
             ] }),
+            entry.protocol !== void 0 && /* @__PURE__ */ (0, import_jsx_runtime2.jsxs)("div", { className: request_log_default.detailItem, children: [
+              /* @__PURE__ */ (0, import_jsx_runtime2.jsx)("span", { className: request_log_default.detailLabel, children: t("requestLogDetailProtocol") }),
+              /* @__PURE__ */ (0, import_jsx_runtime2.jsx)("span", { className: `${request_log_default.detailValue} ${request_log_default.mono}`, children: entry.protocol })
+            ] }),
             /* @__PURE__ */ (0, import_jsx_runtime2.jsxs)("div", { className: request_log_default.detailItem, children: [
               /* @__PURE__ */ (0, import_jsx_runtime2.jsx)("span", { className: request_log_default.detailLabel, children: t("requestLogDetailUrl") }),
               /* @__PURE__ */ (0, import_jsx_runtime2.jsx)("span", { className: `${request_log_default.detailValue} ${request_log_default.mono}`, children: entry.url })
             ] }),
+            entry.requestBytes !== void 0 && /* @__PURE__ */ (0, import_jsx_runtime2.jsxs)("div", { className: request_log_default.detailItem, children: [
+              /* @__PURE__ */ (0, import_jsx_runtime2.jsx)("span", { className: request_log_default.detailLabel, children: t("requestLogDetailRequestBytes") }),
+              /* @__PURE__ */ (0, import_jsx_runtime2.jsx)("span", { className: request_log_default.detailValue, children: bytes(entry.requestBytes) })
+            ] }),
             /* @__PURE__ */ (0, import_jsx_runtime2.jsxs)("div", { className: request_log_default.detailItem, children: [
               /* @__PURE__ */ (0, import_jsx_runtime2.jsx)("span", { className: request_log_default.detailLabel, children: t("requestLogDetailSession") }),
               /* @__PURE__ */ (0, import_jsx_runtime2.jsx)(
@@ -2315,7 +2360,15 @@ function RequestLogPage({ readLog, enabled, t, getLocale, sessionId, useSessions
             ] }),
             /* @__PURE__ */ (0, import_jsx_runtime2.jsxs)("div", { className: request_log_default.detailItem, children: [
               /* @__PURE__ */ (0, import_jsx_runtime2.jsx)("span", { className: request_log_default.detailLabel, children: t("requestLogDetailTokens") }),
-              /* @__PURE__ */ (0, import_jsx_runtime2.jsx)("span", { className: request_log_default.detailValue, children: `${count2(entry.inputTokens)} / ${count2(entry.outputTokens)} / ${count2(entry.cacheReadTokens)}` })
+              /* @__PURE__ */ (0, import_jsx_runtime2.jsxs)("span", { className: request_log_default.detailValue, children: [
+                `${count2(entry.inputTokens)} / ${count2(entry.outputTokens)} / ${count2(entry.cacheReadTokens)}`,
+                entry.cacheWriteTokens !== void 0 && ` \xB7 ${t("requestLogDetailCacheWrite")} ${count2(entry.cacheWriteTokens)}`,
+                entry.reasoningTokens !== void 0 && ` \xB7 ${t("requestLogDetailReasoning")} ${count2(entry.reasoningTokens)}`
+              ] })
+            ] }),
+            entry.costUsd !== void 0 && /* @__PURE__ */ (0, import_jsx_runtime2.jsxs)("div", { className: request_log_default.detailItem, children: [
+              /* @__PURE__ */ (0, import_jsx_runtime2.jsx)("span", { className: request_log_default.detailLabel, children: t("requestLogDetailCost") }),
+              /* @__PURE__ */ (0, import_jsx_runtime2.jsx)("span", { className: request_log_default.detailValue, children: cost(entry.costUsd) })
             ] }),
             /* @__PURE__ */ (0, import_jsx_runtime2.jsxs)("div", { className: request_log_default.detailItem, children: [
               /* @__PURE__ */ (0, import_jsx_runtime2.jsx)("span", { className: request_log_default.detailLabel, children: t("requestLogColCacheRate") }),
@@ -2325,6 +2378,10 @@ function RequestLogPage({ readLog, enabled, t, getLocale, sessionId, useSessions
               /* @__PURE__ */ (0, import_jsx_runtime2.jsx)("span", { className: request_log_default.detailLabel, children: t("requestLogColTps") }),
               /* @__PURE__ */ (0, import_jsx_runtime2.jsx)("span", { className: request_log_default.detailValue, children: rate2(entry.tps) })
             ] }),
+            entry.streamMs !== void 0 && entry.tps === void 0 && /* @__PURE__ */ (0, import_jsx_runtime2.jsxs)("div", { className: request_log_default.detailItem, children: [
+              /* @__PURE__ */ (0, import_jsx_runtime2.jsx)("span", { className: request_log_default.detailLabel, children: t("requestLogColStream") }),
+              /* @__PURE__ */ (0, import_jsx_runtime2.jsx)("span", { className: request_log_default.detailValue, children: duration(entry.streamMs) })
+            ] }),
             !entry.ok && entry.code !== void 0 && /* @__PURE__ */ (0, import_jsx_runtime2.jsxs)("div", { className: request_log_default.detailItem, children: [
               /* @__PURE__ */ (0, import_jsx_runtime2.jsx)("span", { className: request_log_default.detailLabel, children: t("requestLogDetailError") }),
               /* @__PURE__ */ (0, import_jsx_runtime2.jsx)("span", { className: request_log_default.detailValue, children: entry.code })
@@ -2513,9 +2570,9 @@ function RequestLogPage({ readLog, enabled, t, getLocale, sessionId, useSessions
             head("atMs", t("requestLogColTime")),
             head("model", t("requestLogColModel")),
             head("session", t("requestLogColSession")),
-            head("inputTokens", t("requestLogColIn")),
-            head("outputTokens", t("requestLogColOut")),
-            head("cacheReadTokens", t("requestLogColCache")),
+            head("inputTokens", t("requestLogColIn"), t("requestLogColInHint")),
+            head("outputTokens", t("requestLogColOut"), t("requestLogColOutHint")),
+            head("cacheReadTokens", t("requestLogColCache"), t("requestLogColCacheHint")),
             head("cacheRate", t("requestLogColCacheRate")),
             head("tps", t("requestLogColTps")),
             head("firstTokenMs", t("requestLogColFirstToken")),

package/lib/index.js

@@ -76522,6 +76522,10 @@ function spendOf(usage, rates2) {
   const cacheWrite = usage.cacheWriteTokens ?? 0;
   return (usage.inputTokens * rates2.input + usage.outputTokens * rates2.output + cacheRead * (rates2.cacheRead ?? fallback) + cacheWrite * (rates2.cacheWrite ?? fallback)) / 1e6;
 }
+function costOf(usage, rates2) {
+  if (rates2 === void 0) return void 0;
+  return spendOf(usage, rates2);
+}
 function add(totals, usage) {
   const inputTokens = totals.inputTokens + usage.inputTokens;
   const outputTokens = totals.outputTokens + usage.outputTokens;
@@ -76600,6 +76604,12 @@ function requestUrl(input) {
   if (input instanceof URL) return input.toString();
   return input.url;
 }
+function bodyBytes(body) {
+  if (typeof body === "string") return Buffer.byteLength(body);
+  if (body instanceof ArrayBuffer) return body.byteLength;
+  if (ArrayBuffer.isView(body)) return body.byteLength;
+  return void 0;
+}
 async function* observingTerminalUsage(events, observe) {
   for await (const event of events) {
     if (event.type === "done") observe(event.message.usage);
@@ -76773,6 +76783,7 @@ var OpencodeGoAdapter = class extends LlmAdapter {
       const span = this.options.requestLog?.start({
         model: options.model,
         sessionId: options.sessionId === void 0 ? void 0 : String(options.sessionId),
+        protocol: facts.api,
         method: "POST",
         url: config.baseURL,
         messageCount: options.messages.length,
@@ -76788,7 +76799,7 @@ var OpencodeGoAdapter = class extends LlmAdapter {
         const base = fetchFactory();
         return (async (input, init) => {
           const response = await base(input, init);
-          span.request(typeof init?.method === "string" ? init.method : "POST", requestUrl(input));
+          span.request(typeof init?.method === "string" ? init.method : "POST", requestUrl(input), bodyBytes(init?.body));
           span.status(response.status);
           return response;
         });
@@ -76868,14 +76879,18 @@ var OpencodeGoAdapter = class extends LlmAdapter {
             } else if (result.value.type === "finish") {
               const reason = result.value.reason;
               const ok = reason.kind === "stop" || reason.kind === "tool-calls" || reason.kind === "max-tokens";
+              const usage = pendingUsage === void 0 ? void 0 : {
+                inputTokens: pendingUsage.inputTokens,
+                outputTokens: pendingUsage.outputTokens,
+                ...pendingUsage.cacheReadTokens === void 0 ? {} : { cacheReadTokens: pendingUsage.cacheReadTokens },
+                ...pendingUsage.cacheWriteTokens === void 0 ? {} : { cacheWriteTokens: pendingUsage.cacheWriteTokens },
+                ...reportedReasoning === void 0 || reportedReasoning <= 0 ? {} : { reasoningTokens: reportedReasoning }
+              };
               span?.finish(
                 ok,
                 ok ? void 0 : reason.failure?.code,
-                pendingUsage === void 0 ? void 0 : {
-                  inputTokens: pendingUsage.inputTokens,
-                  outputTokens: pendingUsage.outputTokens,
-                  ...pendingUsage.cacheReadTokens === void 0 ? {} : { cacheReadTokens: pendingUsage.cacheReadTokens }
-                }
+                usage,
+                usage === void 0 ? void 0 : costOf(usage, rates2)
               );
             }
             yield result.value;
@@ -77111,10 +77126,14 @@ function parseEntry(value) {
   const row = value;
   const shape = row["shape"];
   const headers = row["headers"];
+  const protocol = row["protocol"];
   return {
     seq: count("seq", row),
     atMs: count("atMs", row),
     model: text("model", row),
+    // Older rows predate the protocol recording; they read without one
+    // rather than failing the whole snapshot.
+    ...protocol === "anthropic-messages" || protocol === "openai-completions" || protocol === "openai-responses" ? { protocol } : {},
     session: text("session", row),
     method: text("method", row),
     url: text("url", row),
@@ -77122,9 +77141,14 @@ function parseEntry(value) {
     ...row["inputTokens"] === void 0 ? {} : { inputTokens: count("inputTokens", row) },
     ...row["outputTokens"] === void 0 ? {} : { outputTokens: count("outputTokens", row) },
     ...row["cacheReadTokens"] === void 0 ? {} : { cacheReadTokens: count("cacheReadTokens", row) },
+    ...row["cacheWriteTokens"] === void 0 ? {} : { cacheWriteTokens: count("cacheWriteTokens", row) },
+    ...row["reasoningTokens"] === void 0 ? {} : { reasoningTokens: count("reasoningTokens", row) },
+    ...row["costUsd"] === void 0 ? {} : { costUsd: rate("costUsd", row) },
+    ...row["requestBytes"] === void 0 ? {} : { requestBytes: count("requestBytes", row) },
     ...row["durationMs"] === void 0 ? {} : { durationMs: count("durationMs", row) },
     ...row["firstTokenMs"] === void 0 ? {} : { firstTokenMs: count("firstTokenMs", row) },
     ...row["decodeMs"] === void 0 ? {} : { decodeMs: count("decodeMs", row) },
+    ...row["streamMs"] === void 0 ? {} : { streamMs: count("streamMs", row) },
     ...row["tps"] === void 0 ? {} : { tps: rate("tps", row) },
     ok: row["ok"] === true,
     ...typeof row["code"] === "string" ? { code: row["code"] } : {},
@@ -77713,6 +77737,7 @@ var RequestLog = class {
     let lastTokenAt;
     let method;
     let url;
+    let requestBytes;
     let status;
     let finished = false;
     return {
@@ -77721,16 +77746,17 @@ var RequestLog = class {
         if (firstTokenAt === void 0) firstTokenAt = now;
         lastTokenAt = now;
       },
-      request: (nextMethod, nextUrl) => {
+      request: (nextMethod, nextUrl, nextRequestBytes) => {
         if (method === void 0) {
           method = nextMethod;
           url = nextUrl;
+          requestBytes = nextRequestBytes;
         }
       },
       status: (value) => {
         status = value;
       },
-      finish: (ok, code, usage) => {
+      finish: (ok, code, usage, costUsd) => {
         if (finished) return;
         finished = true;
         const endedAt = Date.now();
@@ -77741,6 +77767,7 @@ var RequestLog = class {
           seq: ++this.seq,
           atMs: startedAt,
           model: detail.model,
+          ...detail.protocol === void 0 ? {} : { protocol: detail.protocol },
           session,
           method: method ?? detail.method ?? "POST",
           url: url ?? detail.url ?? "",
@@ -77748,11 +77775,16 @@ var RequestLog = class {
           ...usage === void 0 ? {} : {
             inputTokens: usage.inputTokens,
             outputTokens: usage.outputTokens,
-            ...usage.cacheReadTokens === void 0 || usage.cacheReadTokens <= 0 ? {} : { cacheReadTokens: usage.cacheReadTokens }
+            ...usage.cacheReadTokens === void 0 || usage.cacheReadTokens <= 0 ? {} : { cacheReadTokens: usage.cacheReadTokens },
+            ...usage.cacheWriteTokens === void 0 || usage.cacheWriteTokens <= 0 ? {} : { cacheWriteTokens: usage.cacheWriteTokens },
+            ...usage.reasoningTokens === void 0 || usage.reasoningTokens <= 0 ? {} : { reasoningTokens: usage.reasoningTokens }
           },
+          ...costUsd === void 0 ? {} : { costUsd },
+          ...requestBytes === void 0 ? {} : { requestBytes },
           durationMs: Math.max(0, endedAt - startedAt),
           ...firstTokenAt === void 0 ? {} : { firstTokenMs: Math.max(0, firstTokenAt - startedAt) },
           ...decodeMs === void 0 ? {} : { decodeMs },
+          ...streamMs === void 0 ? {} : { streamMs },
           ...tps === void 0 ? {} : { tps },
           ok,
           ...code === void 0 ? {} : { code },
@@ -77893,7 +77925,7 @@ function createUpdateChecker(deps) {
 }
 
 // src/update/version.ts
-var PLUGIN_VERSION = true ? "0.1.67" : "0.0.0-dev";
+var PLUGIN_VERSION = true ? "0.1.68" : "0.0.0-dev";
 
 // src/update/service.ts
 import { TypertRemoteService as TypertRemoteService4 } from "@deepseek-ai/dsh-typert-protocol";

package/lib/types/request-log/store.d.ts

@@ -19,6 +19,7 @@
  *
  * @module @dan-ai-studio/dshopencodego/request-log/store
  */
+import type { WireProtocol } from '../catalog/protocol.ts';
 /** One recorded request. Counts are provider-reported; timings are Host-measured. */
 export interface RequestLogEntry {
     /** Monotonic sequence within this process; gaps mean nothing was recorded. */
@@ -27,6 +28,11 @@ export interface RequestLogEntry {
     readonly atMs: number;
     /** The model id the request was made with. */
     readonly model: string;
+    /**
+     * The wire protocol the request ran on, as the catalog resolved it.
+     * Present from the build that records it; older rows simply lack it.
+     */
+    readonly protocol?: WireProtocol;
     /** The session id the request belonged to, whole. */
     readonly session: string;
     /** HTTP method used. */
@@ -41,6 +47,32 @@ export interface RequestLogEntry {
     readonly outputTokens?: number;
     /** Provider-reported cache-read tokens, when positive. */
     readonly cacheReadTokens?: number;
+    /** Provider-reported cache-write tokens, when positive. */
+    readonly cacheWriteTokens?: number;
+    /**
+     * Provider-reported reasoning tokens, when positive. A subset of
+     * `outputTokens` — never added to a total — recorded so a thinking turn's
+     * shape stays visible.
+     */
+    readonly reasoningTokens?: number;
+    /**
+     * Estimated USD for this request at the rates in force when it ran, by the
+     * same function that prices the since-boot meter. Absent for unpriced
+     * models: no zero standing in for "unknown".
+     */
+    readonly costUsd?: number;
+    /**
+     * Request body size in bytes, as handed to fetch. Explanatory, not
+     * sensitive: a multi-megabyte turn uploads slowly no matter the model.
+     */
+    readonly requestBytes?: number;
+    /**
+     * Wall time from the first to the last content token, in milliseconds —
+     * the window the output actually streamed in. Absent when fewer than two
+     * content tokens arrived; a burst delivery reads here as ~zero while the
+     * whole reply still counts in the totals above.
+     */
+    readonly streamMs?: number;
     /** Wall time from request start to terminal event, in milliseconds. */
     readonly durationMs?: number;
     /** Wall time from request start to the first token delta, in milliseconds. */
@@ -141,8 +173,10 @@ export interface RequestSpan {
      * Record the wire request line, from the fetch wrapper. Called at most
      * once per span; a second call keeps the first, so retries never rewrite
      * the original line.
+     * @param requestBytes - request body size in bytes, when the body was a
+     *   measurable in-memory value rather than a stream.
      */
-    request(method: string, url: string): void;
+    request(method: string, url: string, requestBytes?: number): void;
     /** Mark the response status line's arrival. */
     status(status: number): void;
     /**
@@ -150,17 +184,22 @@ export interface RequestSpan {
      * @param ok - whether the request completed.
      * @param code - failure code for a request that did not complete.
      * @param usage - provider-reported counts, when the terminal usage arrived.
+     * @param costUsd - estimated USD at the call's rates, when priced.
      */
     finish(ok: boolean, code?: string, usage?: {
         readonly inputTokens: number;
         readonly outputTokens: number;
         readonly cacheReadTokens?: number;
-    }): void;
+        readonly cacheWriteTokens?: number;
+        readonly reasoningTokens?: number;
+    }, costUsd?: number): void;
 }
 /** Detail the adapter hands the recorder for one request. */
 export interface RequestStart {
     readonly model: string;
     readonly sessionId: string | undefined;
+    /** The wire protocol, as the catalog resolved it for this model. */
+    readonly protocol?: WireProtocol;
     /** Fallback line, used only when the fetch wrapper never reported the real one. */
     readonly method?: string;
     readonly url?: string;

package/lib/types/usage/meter.d.ts

@@ -123,6 +123,22 @@ export interface GoMeter {
     /** Per-model entries, busiest first. */
     readonly models: readonly MeterModelEntry[];
 }
+/**
+ * What one call's tokens cost at the rates in force for it, without touching
+ * any accumulator.
+ *
+ * The request log prices one row through this while the meter keeps the
+ * since-boot totals: same function, same rates, so the two can never
+ * disagree. Unpriced calls have no estimate rather than a zero one.
+ * @param usage - the call's counts; reasoning tokens ride inside `output`
+ *   and are never billed twice.
+ * @param rates - the rates published for the model at call time.
+ * @returns estimated USD, or undefined when the model published no rates.
+ */
+export declare function costOf(usage: Pick<TokenUsage, 'inputTokens' | 'outputTokens'> & {
+    readonly cacheReadTokens?: number;
+    readonly cacheWriteTokens?: number;
+}, rates: MeterRates | undefined): number | undefined;
 /** Accumulates provider-reported usage for the lifetime of one Host process. */
 export declare class UsageMeter {
     private readonly sinceMs;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@dan-ai-studio/dshopencodego",
-  "version": "0.1.67",
+  "version": "0.1.68",
   "description": "OpenCode Go provider for DeepSeek Harness: live gateway catalog, per-conversation session header, usage",
   "type": "module",
   "license": "MIT",