niceeval 0.1.0

package/src/o11y/parsers/bub.ts

@@ -0,0 +1,240 @@
+// Bub(republic tape)transcript 解析器,同时充当「通用 JSONL」兜底解析器。
+//
+// 真实 tape 格式(~/.bub/tapes/<md5(ws)__md5(sess)>.jsonl,按 id 升序逐行):
+//   - kind=message      payload={role, content}                 → 对话
+//   - kind=tool_call    payload={calls:[{id,function:{name,arguments}}]}
+//   - kind=tool_result  payload={results:[...]}                  ← 与上一条 tool_call.calls 按位对齐
+//   - kind=event        payload={name, data}                     name=="run" → data.usage 是该次模型调用用量
+//   - kind=anchor       payload={name, state:{summary?}}         name!="session/start" = tape.handoff(压缩检查点)
+//   - kind=system/error
+// 用量路径(已对真实 tape 校验):payload.data.usage.{prompt_tokens,completion_tokens,total_tokens,
+//   prompt_tokens_details.cached_tokens, cost}。compaction = 非 bootstrap 的 anchor(handoff)条数。
+// 也保留 legacy {type:...} 与无 type/kind 的通用兜底。
+
+import type { StreamEvent, Usage, ToolName, JsonValue } from "../../types.ts";
+import type { ParsedTranscript } from "./index.ts";
+
+function normalizeToolName(name: string): ToolName {
+  const toolMap: Record<string, ToolName> = {
+    "fs.read": "file_read", fs_read: "file_read", read_file: "file_read", read: "file_read",
+    "fs.write": "file_write", fs_write: "file_write", write_file: "file_write", write: "file_write", create_file: "file_write",
+    "fs.edit": "file_edit", fs_edit: "file_edit", edit_file: "file_edit", edit: "file_edit", apply_patch: "file_edit",
+    bash: "shell", shell: "shell", exec: "shell", command_execution: "shell",
+    "web.fetch": "web_fetch", web_fetch: "web_fetch", fetch: "web_fetch", curl: "web_fetch",
+    "web.search": "web_search", web_search: "web_search",
+    glob: "glob", grep: "grep", ls: "list_dir", list_dir: "list_dir",
+    task: "agent_task", update_todos: "agent_task",
+  };
+  return toolMap[name] || toolMap[name.toLowerCase()] || "unknown";
+}
+
+function get(obj: unknown, key: string): unknown {
+  return obj && typeof obj === "object" ? (obj as Record<string, unknown>)[key] : undefined;
+}
+
+function num(obj: unknown, ...keys: string[]): number {
+  for (const k of keys) {
+    const v = get(obj, k);
+    if (typeof v === "number" && Number.isFinite(v)) return v;
+  }
+  return 0;
+}
+
+function coerceArgs(value: unknown): JsonValue {
+  if (typeof value === "string") {
+    try {
+      return JSON.parse(value) as JsonValue;
+    } catch {
+      return value;
+    }
+  }
+  return (value ?? {}) as JsonValue;
+}
+
+function extractText(content: unknown): string {
+  if (typeof content === "string") return content;
+  if (Array.isArray(content)) {
+    const parts: string[] = [];
+    for (const b of content) {
+      if (typeof b === "string") parts.push(b);
+      else {
+        const t = get(b, "text") ?? get(b, "content");
+        if (typeof t === "string") parts.push(t);
+      }
+    }
+    return parts.join("\n");
+  }
+  return "";
+}
+
+export function parseBubTranscript(raw: string | undefined): ParsedTranscript {
+  const events: StreamEvent[] = [];
+  let inputTokens = 0;
+  let outputTokens = 0;
+  let cacheReadTokens = 0;
+  let costUSD = 0;
+  let requests = 0;
+  let compactions = 0;
+  let parseSuccess = true;
+
+  if (!raw || !raw.trim()) {
+    return { events, usage: { inputTokens: 0, outputTokens: 0 }, compactions: 0, parseSuccess: true };
+  }
+
+  // tape 的 tool_result 与上一条 tool_call 按位对齐;记下上一批 callId。
+  let lastCallIds: string[] = [];
+  // 无显式 id 的兜底配对。
+  const pendingCallIds: string[] = [];
+  let synth = 0;
+  const nextSynthId = (): string => `bub_${++synth}`;
+
+  // 从 event/run 的 data.usage(或 legacy 顶层 usage)累加用量。
+  const addUsage = (usage: unknown): void => {
+    if (!usage || typeof usage !== "object") return;
+    const input = num(usage, "input_tokens", "prompt_tokens", "inputTokens");
+    const output = num(usage, "output_tokens", "completion_tokens", "outputTokens");
+    let cache = num(usage, "cached_input_tokens", "cache_read_input_tokens", "cacheReadTokens");
+    if (cache === 0) cache = num(get(usage, "prompt_tokens_details"), "cached_tokens");
+    const cost = num(usage, "cost");
+    if (input === 0 && output === 0 && cache === 0 && cost === 0) return;
+    inputTokens += input;
+    outputTokens += output;
+    cacheReadTokens += cache;
+    costUSD += cost;
+    requests += 1;
+  };
+
+  const emitCall = (originalName: string, input: JsonValue, explicitId: unknown): string => {
+    const callId = explicitId != null ? String(explicitId) : nextSynthId();
+    if (explicitId == null) pendingCallIds.push(callId);
+    events.push({ type: "action.called", callId, name: originalName, input, tool: normalizeToolName(originalName) });
+    return callId;
+  };
+
+  const emitResult = (output: unknown, success: boolean, explicitId: unknown): void => {
+    const callId = explicitId != null ? String(explicitId) : pendingCallIds.shift() ?? nextSynthId();
+    events.push({ type: "action.result", callId, output: (output ?? null) as JsonValue, status: success ? "completed" : "failed" });
+  };
+
+  for (const line of raw.split("\n")) {
+    const trimmed = line.trim();
+    if (!trimmed) continue;
+
+    let entry: unknown;
+    try {
+      entry = JSON.parse(trimmed);
+    } catch {
+      parseSuccess = false;
+      continue;
+    }
+
+    try {
+      const type = get(entry, "type");
+      const kind = get(entry, "kind");
+
+      // ── legacy: 顶层 type ──
+      if (typeof type === "string") {
+        addUsage(get(entry, "usage"));
+        if (type === "message") {
+          const role = get(entry, "role") === "user" ? "user" : "assistant";
+          const text = extractText(get(entry, "content") ?? get(entry, "text"));
+          if (text) events.push({ type: "message", role, text });
+        } else if (type === "tool_call") {
+          const name = String(get(entry, "tool_name") ?? get(entry, "name") ?? get(entry, "tool") ?? "unknown");
+          emitCall(name, coerceArgs(get(entry, "args") ?? get(entry, "input")), get(entry, "call_id") ?? get(entry, "id"));
+        } else if (type === "tool_result") {
+          emitResult(get(entry, "result") ?? get(entry, "output") ?? get(entry, "content"), get(entry, "success") !== false && !get(entry, "error"), get(entry, "call_id") ?? get(entry, "tool_call_id") ?? get(entry, "id"));
+        } else if (type === "thinking" || type === "reasoning") {
+          const text = extractText(get(entry, "content") ?? get(entry, "text"));
+          if (text) events.push({ type: "thinking", text });
+        } else if (type === "error") {
+          events.push({ type: "error", message: String(get(entry, "message") ?? get(entry, "error") ?? "error") });
+        }
+        continue;
+      }
+
+      // ── republic tape: kind + payload ──
+      if (typeof kind === "string") {
+        const payload = get(entry, "payload");
+        if (kind === "message") {
+          const role = get(payload, "role") === "user" ? "user" : "assistant";
+          const text = extractText(get(payload, "content") ?? get(payload, "text"));
+          if (text) events.push({ type: "message", role, text });
+        } else if (kind === "tool_call") {
+          const calls = get(payload, "calls");
+          const list = Array.isArray(calls) ? calls : [payload];
+          lastCallIds = [];
+          for (const call of list) {
+            const fn = get(call, "function") ?? call;
+            const name = String(get(fn, "name") ?? get(call, "name") ?? "unknown");
+            const id = emitCall(name, coerceArgs(get(fn, "arguments") ?? get(call, "args") ?? get(call, "input")), get(call, "id") ?? get(call, "call_id"));
+            lastCallIds.push(id);
+          }
+        } else if (kind === "tool_result") {
+          const results = get(payload, "results");
+          const list = Array.isArray(results) ? results : [get(payload, "result") ?? payload];
+          // republic:results 与上一条 tool_call.calls 按位对齐。
+          list.forEach((r, i) => {
+            const isError = typeof r === "object" && r !== null && "error" in (r as Record<string, unknown>);
+            const explicitId = get(r, "tool_call_id") ?? get(r, "id") ?? lastCallIds[i];
+            emitResult(get(r, "output") ?? get(r, "result") ?? get(r, "content") ?? r, !isError, explicitId);
+          });
+        } else if (kind === "event") {
+          const data = get(payload, "data");
+          // 用量:run 事件首选 data.usage;但新版 bub.tape 不一定每轮都发 run/usage,
+          // 所以任何 event 上挂的 usage 也兜底累加(addUsage 自带去重式全零跳过)。
+          addUsage(get(data, "usage"));
+          // 仅在「真失败」时记 error:有 error 字段,或 status 命中失败词。
+          // loop.step 这类控制事件 status=="continue"/"running" 不是错误 —— 别再当 error 刷屏。
+          const status = get(data, "status");
+          const errVal = get(data, "error");
+          const failed =
+            (errVal != null && errVal !== false && errVal !== "") ||
+            (typeof status === "string" && /^(error|failed|aborted|cancell?ed|timeout)$/i.test(status));
+          if (failed) {
+            events.push({ type: "error", message: typeof errVal === "string" && errVal ? errVal : `${String(get(payload, "name"))}: ${String(status)}` });
+          }
+        } else if (kind === "anchor") {
+          // 非 bootstrap 的 anchor = tape.handoff = 一次压缩检查点(供 t.transcript.compactions() 守卫)。
+          const name = String(get(payload, "name") ?? "");
+          if (name && name !== "session/start") {
+            compactions += 1;
+            events.push({ type: "compaction", reason: name });
+          }
+        } else if (kind === "error") {
+          events.push({ type: "error", message: String(get(payload, "message") ?? get(payload, "kind") ?? "error") });
+        }
+        continue;
+      }
+
+      // ── 通用兜底:既无 type 也无 kind ──
+      addUsage(get(entry, "usage"));
+      const role = get(entry, "role");
+      const name = get(entry, "tool") ?? get(entry, "name");
+      if (role === "assistant" || role === "user") {
+        const text = extractText(get(entry, "content") ?? get(entry, "text"));
+        if (text) events.push({ type: "message", role, text });
+      } else if (typeof name === "string") {
+        if (get(entry, "result") !== undefined || get(entry, "output") !== undefined) {
+          emitResult(get(entry, "result") ?? get(entry, "output"), !get(entry, "error"), get(entry, "call_id") ?? get(entry, "id"));
+        } else {
+          emitCall(name, coerceArgs(get(entry, "args") ?? get(entry, "arguments") ?? get(entry, "input")), get(entry, "call_id") ?? get(entry, "id"));
+        }
+      }
+    } catch {
+      parseSuccess = false;
+    }
+  }
+
+  const usage: Usage = { inputTokens, outputTokens };
+  if (cacheReadTokens > 0) usage.cacheReadTokens = cacheReadTokens;
+  if (requests > 0) usage.requests = requests;
+  if (costUSD > 0) usage.costUSD = costUSD;
+
+  return { events, usage, compactions, parseSuccess };
+}
+
+/** 便捷形态:只要 StreamEvent[]。 */
+export function parseBub(raw: string | undefined): StreamEvent[] {
+  return parseBubTranscript(raw).events;
+}

package/src/o11y/parsers/claude-code.ts

@@ -0,0 +1,270 @@
+// Claude Code transcript 解析器。
+// Claude Code 把会话存成 JSONL:~/.claude/projects/{path}/{session}.jsonl。
+//   - 每行 { type:"user"|"assistant", message:{ content:[...], usage:{...} } };
+//   - assistant 行的 content 里混着 text / tool_use / thinking 块;
+//   - 工具结果以 user 行里的 tool_result 块回来(按 tool_use_id 配对)。
+// 目标:归一成 niceeval StreamEvent[]。
+
+import type { StreamEvent, Usage, ToolName, JsonValue } from "../../types.ts";
+import type { ParsedTranscript } from "./index.ts";
+
+// ───────────────────────── 工具名归一 ─────────────────────────
+
+function normalizeToolName(name: string): ToolName {
+  const toolMap: Record<string, ToolName> = {
+    // 文件
+    Read: "file_read",
+    read_file: "file_read",
+    ReadFile: "file_read",
+    Write: "file_write",
+    write_file: "file_write",
+    WriteFile: "file_write",
+    write_to_file: "file_write",
+    create_file: "file_write",
+    Edit: "file_edit",
+    MultiEdit: "file_edit",
+    edit_file: "file_edit",
+    EditFile: "file_edit",
+    str_replace_editor: "file_edit",
+    StrReplace: "file_edit",
+    NotebookEdit: "file_edit",
+
+    // shell
+    Bash: "shell",
+    bash: "shell",
+    BashOutput: "shell",
+    Shell: "shell",
+    shell: "shell",
+    execute_command: "shell",
+    run_command: "shell",
+
+    // web
+    WebFetch: "web_fetch",
+    web_fetch: "web_fetch",
+    fetch_url: "web_fetch",
+    mcp__fetch__fetch: "web_fetch",
+    WebSearch: "web_search",
+    web_search: "web_search",
+
+    // 检索 / 导航
+    Glob: "glob",
+    glob: "glob",
+    list_files: "glob",
+    Grep: "grep",
+    grep: "grep",
+    search_files: "grep",
+    LS: "list_dir",
+    list_dir: "list_dir",
+    ListDir: "list_dir",
+
+    // 子 agent
+    Task: "agent_task",
+    task: "agent_task",
+  };
+  return toolMap[name] || "unknown";
+}
+
+// ───────────────────────── 小工具 ─────────────────────────
+
+function get(obj: unknown, key: string): unknown {
+  return obj && typeof obj === "object" ? (obj as Record<string, unknown>)[key] : undefined;
+}
+
+/** content 既可能是 string,也可能挂在 data.message.content;统一取数组。 */
+function getContentArray(data: unknown): unknown[] | undefined {
+  const direct = get(data, "content");
+  if (Array.isArray(direct)) return direct;
+  const msgContent = get(get(data, "message"), "content");
+  if (Array.isArray(msgContent)) return msgContent;
+  return undefined;
+}
+
+function getStringContent(data: unknown): string | undefined {
+  const direct = get(data, "content");
+  if (typeof direct === "string") return direct;
+  const msgContent = get(get(data, "message"), "content");
+  if (typeof msgContent === "string") return msgContent;
+  return undefined;
+}
+
+/** 抠 text 块(含 message.content[].type==="text")。 */
+function extractText(data: unknown): string {
+  const s = getStringContent(data);
+  if (s) return s;
+  const arr = getContentArray(data);
+  if (arr) {
+    const texts = arr
+      .filter((b) => get(b, "type") === "text")
+      .map((b) => get(b, "text"))
+      .filter((t): t is string => typeof t === "string");
+    if (texts.length > 0) return texts.join("\n");
+  }
+  const t = get(data, "text");
+  return typeof t === "string" ? t : "";
+}
+
+/** 抠 thinking 块。 */
+function extractThinking(data: unknown): string {
+  const arr = getContentArray(data);
+  if (!arr) return "";
+  const blocks = arr
+    .filter((b) => get(b, "type") === "thinking")
+    .map((b) => get(b, "thinking") ?? get(b, "text"))
+    .filter((t): t is string => typeof t === "string");
+  return blocks.join("\n");
+}
+
+// ───────────────────────── usage 聚合 ─────────────────────────
+
+function readUsage(u: unknown): { input: number; output: number; cacheRead: number } | null {
+  if (!u || typeof u !== "object") return null;
+  const o = u as Record<string, unknown>;
+  const num = (...keys: string[]): number => {
+    for (const k of keys) {
+      const v = o[k];
+      if (typeof v === "number" && Number.isFinite(v)) return v;
+    }
+    return 0;
+  };
+  const input = num("input_tokens", "prompt_tokens", "inputTokens");
+  const output = num("output_tokens", "completion_tokens", "outputTokens");
+  const cacheRead = num("cache_read_input_tokens", "cached_input_tokens", "cache_read_tokens");
+  if (input === 0 && output === 0 && cacheRead === 0) return null;
+  return { input, output, cacheRead };
+}
+
+// ───────────────────────── compaction 标记 ─────────────────────────
+
+/** Claude Code 压缩边界:type:"summary" / isCompactSummary / subtype:"compact_boundary"。 */
+function isCompactSummary(data: unknown): boolean {
+  if (get(data, "type") === "summary") return true;
+  if (get(data, "isCompactSummary") === true) return true;
+  if (get(get(data, "message"), "isCompactSummary") === true) return true;
+  if (get(data, "subtype") === "compact_boundary") return true;
+  return false;
+}
+
+// ───────────────────────── 主解析 ─────────────────────────
+
+export function parseClaudeCodeTranscript(raw: string | undefined): ParsedTranscript {
+  const events: StreamEvent[] = [];
+  let inputTokens = 0;
+  let outputTokens = 0;
+  let cacheReadTokens = 0;
+  let requests = 0;
+  let compactions = 0;
+  let parseSuccess = true;
+
+  if (!raw || !raw.trim()) {
+    return { events, usage: { inputTokens: 0, outputTokens: 0 }, compactions: 0, parseSuccess: true };
+  }
+
+  const addUsageFrom = (data: unknown): void => {
+    const u = readUsage(get(get(data, "message"), "usage") ?? get(data, "usage"));
+    if (!u) return;
+    inputTokens += u.input;
+    outputTokens += u.output;
+    cacheReadTokens += u.cacheRead;
+    requests += 1;
+  };
+
+  const lines = raw.split("\n");
+  for (const line of lines) {
+    const trimmed = line.trim();
+    if (!trimmed) continue;
+
+    let data: unknown;
+    try {
+      data = JSON.parse(trimmed);
+    } catch {
+      parseSuccess = false;
+      continue;
+    }
+
+    try {
+      // 压缩边界(在常规分类之前判,优先吃掉)。
+      if (isCompactSummary(data)) {
+        events.push({ type: "compaction" });
+        compactions += 1;
+        continue;
+      }
+
+      const type = get(data, "type");
+      const role = get(data, "role");
+
+      if (type === "user" || role === "user") {
+        addUsageFrom(data);
+        const arr = getContentArray(data);
+        const toolResults = arr?.filter((b) => get(b, "type") === "tool_result") ?? [];
+        if (toolResults.length > 0) {
+          for (const r of toolResults) {
+            const callId = String(get(r, "tool_use_id") ?? get(r, "id") ?? "unknown");
+            const isError = get(r, "is_error") === true || !!get(r, "error");
+            events.push({
+              type: "action.result",
+              callId,
+              output: (get(r, "content") ?? null) as JsonValue,
+              status: isError ? "failed" : "completed",
+            });
+          }
+        } else {
+          const text = extractText(data);
+          if (text) events.push({ type: "message", role: "user", text });
+        }
+      } else if (type === "assistant" || role === "assistant") {
+        addUsageFrom(data);
+
+        const text = extractText(data);
+        if (text) events.push({ type: "message", role: "assistant", text });
+
+        const thinking = extractThinking(data);
+        if (thinking) events.push({ type: "thinking", text: thinking });
+
+        const arr = getContentArray(data);
+        if (arr) {
+          for (const b of arr) {
+            if (get(b, "type") !== "tool_use") continue;
+            const name = String(get(b, "name") ?? "unknown");
+            const callId = String(get(b, "id") ?? "unknown");
+            const input = (get(b, "input") ?? {}) as JsonValue;
+            events.push({
+              type: "action.called",
+              callId,
+              name,
+              input,
+              tool: normalizeToolName(name),
+            });
+          }
+        }
+      } else if (type === "tool_result" || type === "tool_response") {
+        const callId = String(get(data, "tool_use_id") ?? get(data, "id") ?? "unknown");
+        const isError = get(data, "is_error") === true || !!get(data, "error");
+        events.push({
+          type: "action.result",
+          callId,
+          output: (get(data, "content") ?? get(data, "output") ?? get(data, "result") ?? null) as JsonValue,
+          status: isError ? "failed" : "completed",
+        });
+      } else if (type === "system" || role === "system") {
+        // 系统行多为元数据,不进事件流(compact_boundary 已在上面吃掉)。
+      } else if (type === "error" || get(data, "error")) {
+        const err = get(data, "error");
+        const msg = get(err, "message") ?? get(data, "message") ?? "error";
+        events.push({ type: "error", message: String(msg) });
+      }
+    } catch {
+      parseSuccess = false;
+    }
+  }
+
+  const usage: Usage = { inputTokens, outputTokens };
+  if (cacheReadTokens > 0) usage.cacheReadTokens = cacheReadTokens;
+  if (requests > 0) usage.requests = requests;
+
+  return { events, usage, compactions, parseSuccess };
+}
+
+/** 便捷形态:只要 StreamEvent[]。 */
+export function parseClaudeCode(raw: string | undefined): StreamEvent[] {
+  return parseClaudeCodeTranscript(raw).events;
+}