@fieldwangai/agentflow 0.1.25

package/bin/lib/agents-path.mjs

@@ -0,0 +1,61 @@
+import fs from "fs";
+import path from "path";
+import { PACKAGE_AGENTS_DIR, getUserAgentsDirAbs } from "./paths.mjs";
+
+/** 解析 agent 身份 prompt 路径：优先使用包内 agents/<subagent>.md，否则工作区 */
+export function getAgentPath(workspaceRoot, subagent) {
+  const packagePath = path.join(PACKAGE_AGENTS_DIR, `${subagent}.md`);
+  if (fs.existsSync(packagePath)) return path.resolve(packagePath);
+  return path.join(getUserAgentsDirAbs(), `${subagent}.md`);
+}
+
+/**
+ * 读取 agent 文件内容并替换路径占位符为真实路径。
+ * 返回替换后的整段文本；若文件不存在则返回空字符串。
+ */
+export function loadAgentPromptWithReplacements(workspaceRoot, subagent, replacements) {
+  const agentPath = getAgentPath(workspaceRoot, subagent);
+  if (!fs.existsSync(agentPath)) return "";
+  let content = fs.readFileSync(agentPath, "utf8");
+  for (const [key, value] of Object.entries(replacements)) {
+    if (value != null && typeof value === "string") {
+      const placeholder = "${" + key + "}";
+      content = content.split(placeholder).join(value);
+    }
+  }
+  return content;
+}
+
+/** 去掉 Markdown 顶部的 YAML frontmatter（--- ... ---），返回正文 */
+export function stripYamlFrontmatter(content) {
+  if (!content || typeof content !== "string") return "";
+  const first = content.indexOf("---");
+  if (first !== 0) return content.trim();
+  const afterFirst = content.indexOf("---", 3);
+  if (afterFirst === -1) return content.trim();
+  return content.slice(afterFirst + 3).trim();
+}
+
+/** 从 agent .md 文件读取 frontmatter 的 name、description（简易解析） */
+export function readAgentFrontmatter(filePath) {
+  if (!fs.existsSync(filePath)) return { name: null, description: null };
+  try {
+    const raw = fs.readFileSync(filePath, "utf8");
+    const start = raw.indexOf("---");
+    if (start === -1) return { name: null, description: null };
+    const afterFirst = raw.indexOf("---", start + 3);
+    if (afterFirst === -1) return { name: null, description: null };
+    const block = raw.slice(start + 3, afterFirst).trim();
+    let name = null;
+    let description = null;
+    for (const line of block.split("\n")) {
+      const m = line.match(/^\s*name:\s*(.+)$/);
+      if (m) name = m[1].trim().replace(/^["']|["']$/g, "");
+      const d = line.match(/^\s*description:\s*(.+)$/);
+      if (d) description = d[1].trim().replace(/^["']|["']$/g, "");
+    }
+    return { name, description };
+  } catch (_) {
+    return { name: null, description: null };
+  }
+}

package/bin/lib/api-runner.mjs

@@ -0,0 +1,361 @@
+/**
+ * 直调 AI API 运行器，绕过 Cursor/OpenCode CLI 冷启动开销。
+ *
+ * 支持的模型格式（在 flow.yaml 或模型配置中指定）：
+ *   api:openai/<model>        — OpenAI 或兼容端点（Together/Groq/DeepSeek/Azure 等）
+ *   api:anthropic/<model>     — Anthropic Claude
+ *
+ * 相关环境变量：
+ *   OPENAI_API_KEY            — OpenAI 或兼容端点的 API key
+ *   OPENAI_BASE_URL           — 兼容端点基础 URL（默认 https://api.openai.com/v1）
+ *   ANTHROPIC_API_KEY         — Anthropic API key
+ *   AGENTFLOW_API_MAX_ROUNDS  — 工具调用最大轮数（默认 30）
+ *   AGENTFLOW_API_MAX_TOKENS  — 单次响应最大 token 数（默认 8192）
+ */
+
+import fs from "fs";
+import path from "path";
+import { spawnSync } from "child_process";
+
+import { loadAgentPromptWithReplacements, stripYamlFrontmatter } from "./agents-path.mjs";
+import { appendRunLogLine } from "./run-events.mjs";
+
+const DEFAULT_OPENAI_BASE = "https://api.openai.com/v1";
+const MAX_TOOL_ROUNDS = parseInt(process.env.AGENTFLOW_API_MAX_ROUNDS ?? "30", 10) || 30;
+const MAX_TOKENS = parseInt(process.env.AGENTFLOW_API_MAX_TOKENS ?? "8192", 10) || 8192;
+
+// ─── 工具定义 ────────────────────────────────────────────────────────────────
+
+const TOOL_DEFS = [
+  {
+    name: "read_file",
+    description: "Read the contents of a file. Returns the file content as text.",
+    parameters: {
+      type: "object",
+      properties: {
+        path: { type: "string", description: "File path (absolute, or relative to workspace root)" },
+      },
+      required: ["path"],
+    },
+  },
+  {
+    name: "write_file",
+    description: "Write content to a file, creating parent directories as needed. Overwrites existing content.",
+    parameters: {
+      type: "object",
+      properties: {
+        path: { type: "string", description: "File path (absolute, or relative to workspace root)" },
+        content: { type: "string", description: "Full content to write" },
+      },
+      required: ["path", "content"],
+    },
+  },
+  {
+    name: "run_command",
+    description: "Run a shell command in the workspace root directory. Returns stdout, stderr and exit code.",
+    parameters: {
+      type: "object",
+      properties: {
+        command: { type: "string", description: "Shell command to execute" },
+        timeout_ms: { type: "number", description: "Timeout in milliseconds (default: 30000)" },
+      },
+      required: ["command"],
+    },
+  },
+  {
+    name: "list_dir",
+    description: "List files and subdirectories at a path.",
+    parameters: {
+      type: "object",
+      properties: {
+        path: { type: "string", description: "Directory path (absolute, or relative to workspace root)" },
+      },
+      required: ["path"],
+    },
+  },
+];
+
+/** OpenAI format: tools array */
+function openAiTools() {
+  return TOOL_DEFS.map((t) => ({
+    type: "function",
+    function: { name: t.name, description: t.description, parameters: t.parameters },
+  }));
+}
+
+/** Anthropic format: tools array */
+function anthropicTools() {
+  return TOOL_DEFS.map((t) => ({
+    name: t.name,
+    description: t.description,
+    input_schema: t.parameters,
+  }));
+}
+
+// ─── 工具执行 ────────────────────────────────────────────────────────────────
+
+function toAbs(workspaceRoot, p) {
+  return path.isAbsolute(p) ? p : path.resolve(workspaceRoot, p);
+}
+
+function executeTool(workspaceRoot, toolName, input) {
+  try {
+    switch (toolName) {
+      case "read_file": {
+        const abs = toAbs(workspaceRoot, input.path);
+        if (!fs.existsSync(abs)) return { error: `File not found: ${input.path}` };
+        const content = fs.readFileSync(abs, "utf-8");
+        return { content };
+      }
+
+      case "write_file": {
+        const abs = toAbs(workspaceRoot, input.path);
+        fs.mkdirSync(path.dirname(abs), { recursive: true });
+        fs.writeFileSync(abs, input.content, "utf-8");
+        return { ok: true, bytes_written: Buffer.byteLength(input.content, "utf-8") };
+      }
+
+      case "run_command": {
+        const timeoutMs = typeof input.timeout_ms === "number" ? input.timeout_ms : 30000;
+        const r = spawnSync(input.command, [], {
+          cwd: workspaceRoot,
+          shell: true,
+          encoding: "utf-8",
+          timeout: timeoutMs,
+        });
+        return {
+          exit_code: r.status ?? -1,
+          stdout: (r.stdout ?? "").slice(0, 20000),
+          stderr: (r.stderr ?? "").slice(0, 5000),
+          timed_out: r.signal === "SIGTERM",
+        };
+      }
+
+      case "list_dir": {
+        const abs = toAbs(workspaceRoot, input.path);
+        if (!fs.existsSync(abs)) return { error: `Path not found: ${input.path}` };
+        const entries = fs.readdirSync(abs, { withFileTypes: true });
+        return {
+          items: entries.map((e) => ({ name: e.name, type: e.isDirectory() ? "dir" : "file" })),
+        };
+      }
+
+      default:
+        return { error: `Unknown tool: ${toolName}` };
+    }
+  } catch (err) {
+    return { error: err.message };
+  }
+}
+
+// ─── API 调用封装 ─────────────────────────────────────────────────────────────
+
+async function fetchOpenAi(apiKey, baseUrl, model, messages) {
+  const url = `${baseUrl.replace(/\/$/, "")}/chat/completions`;
+  const resp = await fetch(url, {
+    method: "POST",
+    headers: {
+      "Content-Type": "application/json",
+      Authorization: `Bearer ${apiKey}`,
+    },
+    body: JSON.stringify({
+      model,
+      messages,
+      tools: openAiTools(),
+      tool_choice: "auto",
+      max_tokens: MAX_TOKENS,
+    }),
+  });
+  if (!resp.ok) {
+    const txt = await resp.text().catch(() => "");
+    throw new Error(`OpenAI API ${resp.status}: ${txt.slice(0, 600)}`);
+  }
+  return resp.json();
+}
+
+async function fetchAnthropic(apiKey, model, messages, systemPrompt) {
+  const resp = await fetch("https://api.anthropic.com/v1/messages", {
+    method: "POST",
+    headers: {
+      "Content-Type": "application/json",
+      "x-api-key": apiKey,
+      "anthropic-version": "2023-06-01",
+    },
+    body: JSON.stringify({
+      model,
+      system: systemPrompt,
+      messages,
+      tools: anthropicTools(),
+      max_tokens: MAX_TOKENS,
+    }),
+  });
+  if (!resp.ok) {
+    const txt = await resp.text().catch(() => "");
+    throw new Error(`Anthropic API ${resp.status}: ${txt.slice(0, 600)}`);
+  }
+  return resp.json();
+}
+
+// ─── 工具调用循环 ─────────────────────────────────────────────────────────────
+
+async function runOpenAiLoop(apiKey, baseUrl, model, systemPrompt, userContent, workspaceRoot, log, options) {
+  const messages = [
+    ...(systemPrompt ? [{ role: "system", content: systemPrompt }] : []),
+    { role: "user", content: userContent },
+  ];
+
+  for (let round = 0; round < MAX_TOOL_ROUNDS; round++) {
+    log(`[api/openai] round ${round + 1}`);
+    const resp = await fetchOpenAi(apiKey, baseUrl, model, messages);
+    const choice = resp.choices?.[0];
+    if (!choice) throw new Error("OpenAI: no choices in response");
+
+    const msg = choice.message;
+    messages.push(msg);
+
+    // 向外广播助手文本（用于 spinner 显示）
+    if (options.onToolCall) {
+      const txt = typeof msg.content === "string" ? msg.content : "";
+      if (txt.trim()) options.onToolCall("assistant", txt.slice(0, 200));
+    }
+
+    if (choice.finish_reason === "stop" || choice.finish_reason === "end_turn" || !msg.tool_calls?.length) {
+      log(`[api/openai] finished (${choice.finish_reason ?? "no-tool-calls"})`);
+      break;
+    }
+
+    const toolResults = [];
+    for (const tc of msg.tool_calls) {
+      const toolName = tc.function.name;
+      let toolInput = {};
+      try { toolInput = JSON.parse(tc.function.arguments); } catch (_) { /**/ }
+
+      log(`[api/openai] tool: ${toolName} ${JSON.stringify(toolInput).slice(0, 120)}`);
+      if (options.onToolCall) options.onToolCall("tool_call", toolName);
+
+      const result = executeTool(workspaceRoot, toolName, toolInput);
+      log(`[api/openai] tool result: ${JSON.stringify(result).slice(0, 200)}`);
+
+      toolResults.push({ role: "tool", tool_call_id: tc.id, content: JSON.stringify(result) });
+    }
+    messages.push(...toolResults);
+  }
+}
+
+async function runAnthropicLoop(apiKey, model, systemPrompt, userContent, workspaceRoot, log, options) {
+  const messages = [{ role: "user", content: userContent }];
+
+  for (let round = 0; round < MAX_TOOL_ROUNDS; round++) {
+    log(`[api/anthropic] round ${round + 1}`);
+    const resp = await fetchAnthropic(apiKey, model, messages, systemPrompt);
+
+    // 助手消息
+    messages.push({ role: "assistant", content: resp.content });
+
+    if (options.onToolCall) {
+      const textBlock = resp.content?.find((b) => b.type === "text");
+      if (textBlock?.text) options.onToolCall("assistant", textBlock.text.slice(0, 200));
+    }
+
+    if (resp.stop_reason === "end_turn" || resp.stop_reason === "stop_sequence") {
+      log(`[api/anthropic] finished (${resp.stop_reason})`);
+      break;
+    }
+
+    const toolUses = (resp.content ?? []).filter((b) => b.type === "tool_use");
+    if (!toolUses.length) {
+      log("[api/anthropic] finished (no tool_use blocks)");
+      break;
+    }
+
+    const toolResults = [];
+    for (const tb of toolUses) {
+      log(`[api/anthropic] tool: ${tb.name} ${JSON.stringify(tb.input).slice(0, 120)}`);
+      if (options.onToolCall) options.onToolCall("tool_call", tb.name);
+
+      const result = executeTool(workspaceRoot, tb.name, tb.input ?? {});
+      log(`[api/anthropic] tool result: ${JSON.stringify(result).slice(0, 200)}`);
+
+      toolResults.push({ type: "tool_result", tool_use_id: tb.id, content: JSON.stringify(result) });
+    }
+    messages.push({ role: "user", content: toolResults });
+  }
+}
+
+// ─── 公共解析函数 ─────────────────────────────────────────────────────────────
+
+/**
+ * 从模型字符串中解析 provider 和 model。
+ * 例：
+ *   "api:openai/gpt-4o"          → { provider: "openai", model: "gpt-4o" }
+ *   "api:anthropic/claude-opus-4-5" → { provider: "anthropic", model: "claude-opus-4-5" }
+ *   "api:gpt-4o"                 → { provider: "openai", model: "gpt-4o" }  (默认 openai)
+ */
+export function parseApiModel(str) {
+  const s = String(str ?? "").replace(/^api:/, "");
+  const slash = s.indexOf("/");
+  if (slash < 0) return { provider: "openai", model: s };
+  const provider = s.slice(0, slash).toLowerCase();
+  const model = s.slice(slash + 1);
+  return { provider, model };
+}
+
+// ─── 主入口 ───────────────────────────────────────────────────────────────────
+
+/**
+ * 直调 AI API 执行节点，绕过 Cursor/OpenCode CLI。
+ *
+ * @param {string} workspaceRoot
+ * @param {{ promptPath: string, nodeContext?: string, taskBody?: string, subagent?: string, instanceId?: string }} nodeInfo
+ * @param {object} options
+ *   model      — "api:openai/gpt-4o" | "api:anthropic/claude-opus-4-5" 等
+ *   flowName   — 用于日志
+ *   uuid       — 用于日志
+ *   onToolCall — (subtype: string, name: string) => void  供 spinner 展示
+ */
+export async function runApiAgentForNode(workspaceRoot, { promptPath, nodeContext, taskBody, subagent, instanceId }, options = {}) {
+  const absRoot = path.resolve(workspaceRoot);
+  const flowName = options.flowName ?? null;
+  const uuid = options.uuid ?? null;
+
+  const log = (msg) => {
+    if (flowName && uuid) appendRunLogLine(absRoot, flowName, uuid, "api-runner", msg);
+  };
+
+  // ── 解析 provider / model ──────────────────────────────────────────────────
+  const modelRaw = String(options.model ?? "api:openai/gpt-4o");
+  const { provider, model } = parseApiModel(modelRaw);
+  log(`start provider=${provider} model=${model} instanceId=${instanceId ?? "-"}`);
+
+  // ── 读取 Agent 角色提示，注入 nodeContext/taskBody ────────────────────────
+  const replacements = {
+    workspaceRoot: absRoot,
+    nodeContext: nodeContext ?? "",
+    taskBody: taskBody ?? "",
+    flowName: flowName ?? "",
+    uuid: uuid ?? "",
+    instanceId: instanceId ?? "",
+  };
+  const agentRaw = loadAgentPromptWithReplacements(workspaceRoot, subagent ?? "agentflow-node-executor", replacements);
+  const renderedBody = stripYamlFrontmatter(agentRaw);
+
+  // ── system/user 拆分：## 节点上下文 之前为角色指令，之后为任务内容 ─────────
+  const splitMarker = "## 节点上下文";
+  const splitIdx = renderedBody.indexOf(splitMarker);
+  const systemPrompt = splitIdx > 0 ? renderedBody.slice(0, splitIdx).trim() : renderedBody;
+  const userContent = `## 节点上下文\n\n${nodeContext ?? ""}\n\n## 执行任务\n\n${taskBody ?? ""}`;
+
+  // ── 确认 API Key ───────────────────────────────────────────────────────────
+  if (provider === "anthropic") {
+    const key = process.env.ANTHROPIC_API_KEY;
+    if (!key) throw new Error("[api-runner] ANTHROPIC_API_KEY is required for api:anthropic/* models");
+    await runAnthropicLoop(key, model, systemPrompt, userContent, absRoot, log, options);
+  } else {
+    const key = process.env.OPENAI_API_KEY;
+    if (!key) throw new Error("[api-runner] OPENAI_API_KEY is required for api:openai/* models");
+    const baseUrl = (process.env.OPENAI_BASE_URL ?? DEFAULT_OPENAI_BASE).trim();
+    await runOpenAiLoop(key, baseUrl, model, systemPrompt, userContent, absRoot, log, options);
+  }
+
+  log(`done instanceId=${instanceId ?? "-"}`);
+}