whale-code 6.5.6 → 6.5.8

package/dist/node/task-runner.js

@@ -0,0 +1,276 @@
+/**
+ * Task Runner — persistent agent execution on the node.
+ *
+ * When the server assigns a task via the gateway (assign_task command),
+ * the node runs its own agent loop using the Anthropic SDK directly.
+ * It has full filesystem access and no timeout constraints.
+ *
+ * Progress events stream back to the server via the gateway WebSocket,
+ * which relays them to session observers.
+ */
+
+import Anthropic from "@anthropic-ai/sdk";
+import { LoopDetector, estimateCostUsd } from "../shared/agent-core.js";
+import { validateExecCommand, auditExecAttempt } from "../shared/exec-validator.js";
+export class TaskRunner {
+  activeTasks = new Map();
+  constructor(gateway, config) {
+    this.gateway = gateway;
+    this.config = config;
+  }
+  async onTaskAssigned(task) {
+    const abort = new AbortController();
+    this.activeTasks.set(task.session_id, abort);
+    const apiKey = this.config.anthropic_api_key || process.env.ANTHROPIC_API_KEY;
+    if (!apiKey) {
+      this.gateway.sendJson({
+        type: "task_result",
+        session_id: task.session_id,
+        success: false,
+        error: "No Anthropic API key configured on this node"
+      });
+      this.activeTasks.delete(task.session_id);
+      return;
+    }
+    const anthropic = new Anthropic({
+      apiKey
+    });
+    const model = "claude-sonnet-4-6";
+    const maxTurns = task.max_turns || 25;
+    const systemPrompt = task.system_prompt || `You are a helpful assistant executing a task on a local machine.\n\nTask: ${task.description}`;
+    const messages = [...(task.history || []), {
+      role: "user",
+      content: task.description
+    }];
+
+    // Basic tool definitions for local execution
+    const tools = [{
+      name: "run_command",
+      description: "Execute a shell command on the local machine. Returns stdout, stderr, and exit code.",
+      input_schema: {
+        type: "object",
+        properties: {
+          command: {
+            type: "string",
+            description: "The shell command to execute"
+          },
+          timeout_ms: {
+            type: "number",
+            description: "Timeout in milliseconds (default: 30000)"
+          }
+        },
+        required: ["command"]
+      }
+    }];
+    let totalTokens = {
+      input: 0,
+      output: 0
+    };
+    let turnCount = 0;
+    const loopDetector = new LoopDetector();
+    try {
+      while (turnCount < maxTurns) {
+        if (abort.signal.aborted) break;
+        turnCount++;
+        this.gateway.sendJson({
+          type: "task_progress",
+          session_id: task.session_id,
+          event: {
+            type: "turn_start",
+            turn: turnCount
+          }
+        });
+        const response = await anthropic.messages.create({
+          model,
+          max_tokens: 8192,
+          system: systemPrompt,
+          messages,
+          tools
+        });
+        totalTokens.input += response.usage.input_tokens;
+        totalTokens.output += response.usage.output_tokens;
+
+        // Collect text and tool use blocks
+        const textBlocks = response.content.filter(b => b.type === "text");
+        const toolBlocks = response.content.filter(b => b.type === "tool_use");
+
+        // Stream text progress
+        for (const block of textBlocks) {
+          if (block.type === "text") {
+            this.gateway.sendJson({
+              type: "task_progress",
+              session_id: task.session_id,
+              event: {
+                type: "text",
+                text: block.text
+              }
+            });
+          }
+        }
+
+        // Add assistant response to history
+        messages.push({
+          role: "assistant",
+          content: response.content
+        });
+
+        // If no tool use, we're done
+        if (response.stop_reason === "end_turn" || toolBlocks.length === 0) {
+          const finalText = textBlocks.map(b => b.type === "text" ? b.text : "").join("");
+          this.gateway.sendJson({
+            type: "task_result",
+            session_id: task.session_id,
+            success: true,
+            response: finalText,
+            tokens: totalTokens,
+            cost_usd: estimateCostUsd(totalTokens.input, totalTokens.output, model)
+          });
+          break;
+        }
+
+        // Execute tools with loop detection
+        const toolResults = [];
+        for (const block of toolBlocks) {
+          if (block.type !== "tool_use") continue;
+
+          // Loop detection: check if this tool call is repetitive
+          const loopCheck = loopDetector.recordCall(block.name, block.input);
+          if (loopCheck.blocked) {
+            toolResults.push({
+              type: "tool_result",
+              tool_use_id: block.id,
+              content: `Blocked by loop detector: ${loopCheck.reason}. Try a different approach.`,
+              is_error: true
+            });
+            continue;
+          }
+          this.gateway.sendJson({
+            type: "task_progress",
+            session_id: task.session_id,
+            event: {
+              type: "tool_start",
+              name: block.name,
+              input: block.input
+            }
+          });
+          let result;
+          try {
+            if (block.name === "run_command") {
+              const input = block.input;
+              result = await this.executeCommand(input.command, input.timeout_ms || 30000);
+            } else {
+              result = `Unknown tool: ${block.name}`;
+            }
+          } catch (err) {
+            result = `Error: ${err.message}`;
+          }
+          this.gateway.sendJson({
+            type: "task_progress",
+            session_id: task.session_id,
+            event: {
+              type: "tool_result",
+              name: block.name,
+              result: result.slice(0, 500)
+            }
+          });
+          toolResults.push({
+            type: "tool_result",
+            tool_use_id: block.id,
+            content: result
+          });
+        }
+        messages.push({
+          role: "user",
+          content: toolResults
+        });
+
+        // End-of-turn bail check
+        const bailCheck = loopDetector.endTurn();
+        if (bailCheck.shouldBail) {
+          this.gateway.sendJson({
+            type: "task_result",
+            session_id: task.session_id,
+            success: false,
+            error: bailCheck.message || "Consecutive failures detected. Stopping.",
+            tokens: totalTokens,
+            cost_usd: estimateCostUsd(totalTokens.input, totalTokens.output, model)
+          });
+          break;
+        }
+
+        // Cost guard
+        const costSoFar = estimateCostUsd(totalTokens.input, totalTokens.output, model);
+        if (task.max_cost_usd && costSoFar >= task.max_cost_usd) {
+          this.gateway.sendJson({
+            type: "task_result",
+            session_id: task.session_id,
+            success: false,
+            error: `Cost limit reached ($${costSoFar.toFixed(4)} >= $${task.max_cost_usd})`,
+            tokens: totalTokens,
+            cost_usd: costSoFar
+          });
+          break;
+        }
+      }
+    } catch (err) {
+      this.gateway.sendJson({
+        type: "task_result",
+        session_id: task.session_id,
+        success: false,
+        error: err.message,
+        tokens: totalTokens,
+        cost_usd: estimateCostUsd(totalTokens.input, totalTokens.output, model)
+      });
+    } finally {
+      this.activeTasks.delete(task.session_id);
+    }
+  }
+  cancel(sessionId) {
+    this.activeTasks.get(sessionId)?.abort();
+  }
+  getActiveTasks() {
+    return Array.from(this.activeTasks.keys());
+  }
+  async executeCommand(command, timeoutMs) {
+    // Validate against exec allowlist
+    const validation = validateExecCommand(command);
+    auditExecAttempt(validation, {
+      nodeId: this.config.node_id
+    });
+    if (!validation.allowed) {
+      return `Command blocked: ${validation.reason}\nexit code: 126`;
+    }
+    const {
+      execFile
+    } = await import("node:child_process");
+    const {
+      promisify
+    } = await import("node:util");
+    const execFileAsync = promisify(execFile);
+
+    // Split command into binary + args for execFile (no shell)
+    const args = command.trim().split(/\s+/);
+    const binary = args.shift();
+    try {
+      const {
+        stdout,
+        stderr
+      } = await execFileAsync(binary, args, {
+        timeout: timeoutMs,
+        cwd: process.env.HOME,
+        maxBuffer: 1024 * 1024
+      });
+      const parts = [];
+      if (stdout) parts.push(stdout);
+      if (stderr) parts.push(`stderr: ${stderr}`);
+      return parts.join("\n") || "(no output)";
+    } catch (err) {
+      const parts = [];
+      if (err.stdout) parts.push(err.stdout);
+      if (err.stderr) parts.push(`stderr: ${err.stderr}`);
+      if (!parts.length) parts.push(err.message);
+      return parts.join("\n") + `\nexit code: ${err.code || 1}`;
+    }
+  }
+}
+//# sourceMappingURL=task-runner.js.map

package/dist/node/task-runner.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"task-runner.js","names":["Anthropic","LoopDetector","estimateCostUsd","validateExecCommand","auditExecAttempt","TaskRunner","activeTasks","Map","constructor","gateway","config","onTaskAssigned","task","abort","AbortController","set","session_id","apiKey","anthropic_api_key","process","env","ANTHROPIC_API_KEY","sendJson","type","success","error","delete","anthropic","model","maxTurns","max_turns","systemPrompt","system_prompt","description","messages","history","role","content","tools","name","input_schema","properties","command","timeout_ms","required","totalTokens","input","output","turnCount","loopDetector","signal","aborted","event","turn","response","create","max_tokens","system","usage","input_tokens","output_tokens","textBlocks","filter","b","toolBlocks","block","text","push","stop_reason","length","finalText","map","join","tokens","cost_usd","toolResults","loopCheck","recordCall","blocked","tool_use_id","id","reason","is_error","result","executeCommand","err","message","slice","bailCheck","endTurn","shouldBail","costSoFar","max_cost_usd","toFixed","cancel","sessionId","get","getActiveTasks","Array","from","keys","timeoutMs","validation","nodeId","node_id","allowed","execFile","promisify","execFileAsync","args","trim","split","binary","shift","stdout","stderr","timeout","cwd","HOME","maxBuffer","parts","code"],"sources":["../../src/node/task-runner.ts"],"sourcesContent":["/**\n * Task Runner — persistent agent execution on the node.\n *\n * When the server assigns a task via the gateway (assign_task command),\n * the node runs its own agent loop using the Anthropic SDK directly.\n * It has full filesystem access and no timeout constraints.\n *\n * Progress events stream back to the server via the gateway WebSocket,\n * which relays them to session observers.\n */\n\nimport Anthropic from \"@anthropic-ai/sdk\";\nimport type { GatewayClient } from \"./gateway-client.js\";\nimport type { NodeConfig } from \"./config.js\";\nimport { LoopDetector, estimateCostUsd } from \"../shared/agent-core.js\";\nimport { validateExecCommand, auditExecAttempt } from \"../shared/exec-validator.js\";\n\nexport interface TaskAssignment {\n  session_id: string;\n  agent_id: string;\n  description: string;\n  tools?: string[];\n  system_prompt?: string;\n  history?: Anthropic.MessageParam[];\n  max_turns?: number;\n  max_cost_usd?: number;\n}\n\nexport class TaskRunner {\n  private activeTasks = new Map<string, AbortController>();\n\n  constructor(\n    private gateway: GatewayClient,\n    private config: NodeConfig,\n  ) {}\n\n  async onTaskAssigned(task: TaskAssignment): Promise<void> {\n    const abort = new AbortController();\n    this.activeTasks.set(task.session_id, abort);\n\n    const apiKey = this.config.anthropic_api_key || process.env.ANTHROPIC_API_KEY;\n    if (!apiKey) {\n      this.gateway.sendJson({\n        type: \"task_result\",\n        session_id: task.session_id,\n        success: false,\n        error: \"No Anthropic API key configured on this node\",\n      });\n      this.activeTasks.delete(task.session_id);\n      return;\n    }\n\n    const anthropic = new Anthropic({ apiKey });\n    const model = \"claude-sonnet-4-6\";\n    const maxTurns = task.max_turns || 25;\n\n    const systemPrompt = task.system_prompt || `You are a helpful assistant executing a task on a local machine.\\n\\nTask: ${task.description}`;\n\n    const messages: Anthropic.MessageParam[] = [\n      ...(task.history || []),\n      { role: \"user\", content: task.description },\n    ];\n\n    // Basic tool definitions for local execution\n    const tools: Anthropic.Tool[] = [\n      {\n        name: \"run_command\",\n        description: \"Execute a shell command on the local machine. Returns stdout, stderr, and exit code.\",\n        input_schema: {\n          type: \"object\" as const,\n          properties: {\n            command: { type: \"string\", description: \"The shell command to execute\" },\n            timeout_ms: { type: \"number\", description: \"Timeout in milliseconds (default: 30000)\" },\n          },\n          required: [\"command\"],\n        },\n      },\n    ];\n\n    let totalTokens = { input: 0, output: 0 };\n    let turnCount = 0;\n    const loopDetector = new LoopDetector();\n\n    try {\n      while (turnCount < maxTurns) {\n        if (abort.signal.aborted) break;\n\n        turnCount++;\n        this.gateway.sendJson({\n          type: \"task_progress\",\n          session_id: task.session_id,\n          event: { type: \"turn_start\", turn: turnCount },\n        });\n\n        const response = await anthropic.messages.create({\n          model,\n          max_tokens: 8192,\n          system: systemPrompt,\n          messages,\n          tools,\n        });\n\n        totalTokens.input += response.usage.input_tokens;\n        totalTokens.output += response.usage.output_tokens;\n\n        // Collect text and tool use blocks\n        const textBlocks = response.content.filter(b => b.type === \"text\");\n        const toolBlocks = response.content.filter(b => b.type === \"tool_use\");\n\n        // Stream text progress\n        for (const block of textBlocks) {\n          if (block.type === \"text\") {\n            this.gateway.sendJson({\n              type: \"task_progress\",\n              session_id: task.session_id,\n              event: { type: \"text\", text: block.text },\n            });\n          }\n        }\n\n        // Add assistant response to history\n        messages.push({ role: \"assistant\", content: response.content });\n\n        // If no tool use, we're done\n        if (response.stop_reason === \"end_turn\" || toolBlocks.length === 0) {\n          const finalText = textBlocks.map(b => b.type === \"text\" ? b.text : \"\").join(\"\");\n          this.gateway.sendJson({\n            type: \"task_result\",\n            session_id: task.session_id,\n            success: true,\n            response: finalText,\n            tokens: totalTokens,\n            cost_usd: estimateCostUsd(totalTokens.input, totalTokens.output, model),\n          });\n          break;\n        }\n\n        // Execute tools with loop detection\n        const toolResults: Anthropic.MessageParam[\"content\"] = [];\n        for (const block of toolBlocks) {\n          if (block.type !== \"tool_use\") continue;\n\n          // Loop detection: check if this tool call is repetitive\n          const loopCheck = loopDetector.recordCall(block.name, block.input as Record<string, unknown>);\n          if (loopCheck.blocked) {\n            (toolResults as any[]).push({\n              type: \"tool_result\",\n              tool_use_id: block.id,\n              content: `Blocked by loop detector: ${loopCheck.reason}. Try a different approach.`,\n              is_error: true,\n            });\n            continue;\n          }\n\n          this.gateway.sendJson({\n            type: \"task_progress\",\n            session_id: task.session_id,\n            event: { type: \"tool_start\", name: block.name, input: block.input },\n          });\n\n          let result: string;\n          try {\n            if (block.name === \"run_command\") {\n              const input = block.input as { command: string; timeout_ms?: number };\n              result = await this.executeCommand(input.command, input.timeout_ms || 30000);\n            } else {\n              result = `Unknown tool: ${block.name}`;\n            }\n          } catch (err) {\n            result = `Error: ${(err as Error).message}`;\n          }\n\n          this.gateway.sendJson({\n            type: \"task_progress\",\n            session_id: task.session_id,\n            event: { type: \"tool_result\", name: block.name, result: result.slice(0, 500) },\n          });\n\n          (toolResults as any[]).push({\n            type: \"tool_result\",\n            tool_use_id: block.id,\n            content: result,\n          });\n        }\n\n        messages.push({ role: \"user\", content: toolResults });\n\n        // End-of-turn bail check\n        const bailCheck = loopDetector.endTurn();\n        if (bailCheck.shouldBail) {\n          this.gateway.sendJson({\n            type: \"task_result\",\n            session_id: task.session_id,\n            success: false,\n            error: bailCheck.message || \"Consecutive failures detected. Stopping.\",\n            tokens: totalTokens,\n            cost_usd: estimateCostUsd(totalTokens.input, totalTokens.output, model),\n          });\n          break;\n        }\n\n        // Cost guard\n        const costSoFar = estimateCostUsd(totalTokens.input, totalTokens.output, model);\n        if (task.max_cost_usd && costSoFar >= task.max_cost_usd) {\n          this.gateway.sendJson({\n            type: \"task_result\",\n            session_id: task.session_id,\n            success: false,\n            error: `Cost limit reached ($${costSoFar.toFixed(4)} >= $${task.max_cost_usd})`,\n            tokens: totalTokens,\n            cost_usd: costSoFar,\n          });\n          break;\n        }\n      }\n    } catch (err) {\n      this.gateway.sendJson({\n        type: \"task_result\",\n        session_id: task.session_id,\n        success: false,\n        error: (err as Error).message,\n        tokens: totalTokens,\n        cost_usd: estimateCostUsd(totalTokens.input, totalTokens.output, model),\n      });\n    } finally {\n      this.activeTasks.delete(task.session_id);\n    }\n  }\n\n  cancel(sessionId: string): void {\n    this.activeTasks.get(sessionId)?.abort();\n  }\n\n  getActiveTasks(): string[] {\n    return Array.from(this.activeTasks.keys());\n  }\n\n  private async executeCommand(command: string, timeoutMs: number): Promise<string> {\n    // Validate against exec allowlist\n    const validation = validateExecCommand(command);\n    auditExecAttempt(validation, { nodeId: this.config.node_id });\n    if (!validation.allowed) {\n      return `Command blocked: ${validation.reason}\\nexit code: 126`;\n    }\n\n    const { execFile } = await import(\"node:child_process\");\n    const { promisify } = await import(\"node:util\");\n    const execFileAsync = promisify(execFile);\n\n    // Split command into binary + args for execFile (no shell)\n    const args = command.trim().split(/\\s+/);\n    const binary = args.shift()!;\n\n    try {\n      const { stdout, stderr } = await execFileAsync(binary, args, {\n        timeout: timeoutMs,\n        cwd: process.env.HOME,\n        maxBuffer: 1024 * 1024,\n      });\n      const parts: string[] = [];\n      if (stdout) parts.push(stdout);\n      if (stderr) parts.push(`stderr: ${stderr}`);\n      return parts.join(\"\\n\") || \"(no output)\";\n    } catch (err: any) {\n      const parts: string[] = [];\n      if (err.stdout) parts.push(err.stdout);\n      if (err.stderr) parts.push(`stderr: ${err.stderr}`);\n      if (!parts.length) parts.push(err.message);\n      return parts.join(\"\\n\") + `\\nexit code: ${err.code || 1}`;\n    }\n  }\n}\n"],"mappings":"AAAA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;;AAEA,OAAOA,SAAS,MAAM,mBAAmB;AAGzC,SAASC,YAAY,EAAEC,eAAe,QAAQ,yBAAyB;AACvE,SAASC,mBAAmB,EAAEC,gBAAgB,QAAQ,6BAA6B;AAanF,OAAO,MAAMC,UAAU,CAAC;EACdC,WAAW,GAAG,IAAIC,GAAG,CAA0B,CAAC;EAExDC,WAAWA,CACDC,OAAsB,EACtBC,MAAkB,EAC1B;IAAA,KAFQD,OAAsB,GAAtBA,OAAsB;IAAA,KACtBC,MAAkB,GAAlBA,MAAkB;EACzB;EAEH,MAAMC,cAAcA,CAACC,IAAoB,EAAiB;IACxD,MAAMC,KAAK,GAAG,IAAIC,eAAe,CAAC,CAAC;IACnC,IAAI,CAACR,WAAW,CAACS,GAAG,CAACH,IAAI,CAACI,UAAU,EAAEH,KAAK,CAAC;IAE5C,MAAMI,MAAM,GAAG,IAAI,CAACP,MAAM,CAACQ,iBAAiB,IAAIC,OAAO,CAACC,GAAG,CAACC,iBAAiB;IAC7E,IAAI,CAACJ,MAAM,EAAE;MACX,IAAI,CAACR,OAAO,CAACa,QAAQ,CAAC;QACpBC,IAAI,EAAE,aAAa;QACnBP,UAAU,EAAEJ,IAAI,CAACI,UAAU;QAC3BQ,OAAO,EAAE,KAAK;QACdC,KAAK,EAAE;MACT,CAAC,CAAC;MACF,IAAI,CAACnB,WAAW,CAACoB,MAAM,CAACd,IAAI,CAACI,UAAU,CAAC;MACxC;IACF;IAEA,MAAMW,SAAS,GAAG,IAAI3B,SAAS,CAAC;MAAEiB;IAAO,CAAC,CAAC;IAC3C,MAAMW,KAAK,GAAG,mBAAmB;IACjC,MAAMC,QAAQ,GAAGjB,IAAI,CAACkB,SAAS,IAAI,EAAE;IAErC,MAAMC,YAAY,GAAGnB,IAAI,CAACoB,aAAa,IAAI,6EAA6EpB,IAAI,CAACqB,WAAW,EAAE;IAE1I,MAAMC,QAAkC,GAAG,CACzC,IAAItB,IAAI,CAACuB,OAAO,IAAI,EAAE,CAAC,EACvB;MAAEC,IAAI,EAAE,MAAM;MAAEC,OAAO,EAAEzB,IAAI,CAACqB;IAAY,CAAC,CAC5C;;IAED;IACA,MAAMK,KAAuB,GAAG,CAC9B;MACEC,IAAI,EAAE,aAAa;MACnBN,WAAW,EAAE,sFAAsF;MACnGO,YAAY,EAAE;QACZjB,IAAI,EAAE,QAAiB;QACvBkB,UAAU,EAAE;UACVC,OAAO,EAAE;YAAEnB,IAAI,EAAE,QAAQ;YAAEU,WAAW,EAAE;UAA+B,CAAC;UACxEU,UAAU,EAAE;YAAEpB,IAAI,EAAE,QAAQ;YAAEU,WAAW,EAAE;UAA2C;QACxF,CAAC;QACDW,QAAQ,EAAE,CAAC,SAAS;MACtB;IACF,CAAC,CACF;IAED,IAAIC,WAAW,GAAG;MAAEC,KAAK,EAAE,CAAC;MAAEC,MAAM,EAAE;IAAE,CAAC;IACzC,IAAIC,SAAS,GAAG,CAAC;IACjB,MAAMC,YAAY,GAAG,IAAIhD,YAAY,CAAC,CAAC;IAEvC,IAAI;MACF,OAAO+C,SAAS,GAAGnB,QAAQ,EAAE;QAC3B,IAAIhB,KAAK,CAACqC,MAAM,CAACC,OAAO,EAAE;QAE1BH,SAAS,EAAE;QACX,IAAI,CAACvC,OAAO,CAACa,QAAQ,CAAC;UACpBC,IAAI,EAAE,eAAe;UACrBP,UAAU,EAAEJ,IAAI,CAACI,UAAU;UAC3BoC,KAAK,EAAE;YAAE7B,IAAI,EAAE,YAAY;YAAE8B,IAAI,EAAEL;UAAU;QAC/C,CAAC,CAAC;QAEF,MAAMM,QAAQ,GAAG,MAAM3B,SAAS,CAACO,QAAQ,CAACqB,MAAM,CAAC;UAC/C3B,KAAK;UACL4B,UAAU,EAAE,IAAI;UAChBC,MAAM,EAAE1B,YAAY;UACpBG,QAAQ;UACRI;QACF,CAAC,CAAC;QAEFO,WAAW,CAACC,KAAK,IAAIQ,QAAQ,CAACI,KAAK,CAACC,YAAY;QAChDd,WAAW,CAACE,MAAM,IAAIO,QAAQ,CAACI,KAAK,CAACE,aAAa;;QAElD;QACA,MAAMC,UAAU,GAAGP,QAAQ,CAACjB,OAAO,CAACyB,MAAM,CAACC,CAAC,IAAIA,CAAC,CAACxC,IAAI,KAAK,MAAM,CAAC;QAClE,MAAMyC,UAAU,GAAGV,QAAQ,CAACjB,OAAO,CAACyB,MAAM,CAACC,CAAC,IAAIA,CAAC,CAACxC,IAAI,KAAK,UAAU,CAAC;;QAEtE;QACA,KAAK,MAAM0C,KAAK,IAAIJ,UAAU,EAAE;UAC9B,IAAII,KAAK,CAAC1C,IAAI,KAAK,MAAM,EAAE;YACzB,IAAI,CAACd,OAAO,CAACa,QAAQ,CAAC;cACpBC,IAAI,EAAE,eAAe;cACrBP,UAAU,EAAEJ,IAAI,CAACI,UAAU;cAC3BoC,KAAK,EAAE;gBAAE7B,IAAI,EAAE,MAAM;gBAAE2C,IAAI,EAAED,KAAK,CAACC;cAAK;YAC1C,CAAC,CAAC;UACJ;QACF;;QAEA;QACAhC,QAAQ,CAACiC,IAAI,CAAC;UAAE/B,IAAI,EAAE,WAAW;UAAEC,OAAO,EAAEiB,QAAQ,CAACjB;QAAQ,CAAC,CAAC;;QAE/D;QACA,IAAIiB,QAAQ,CAACc,WAAW,KAAK,UAAU,IAAIJ,UAAU,CAACK,MAAM,KAAK,CAAC,EAAE;UAClE,MAAMC,SAAS,GAAGT,UAAU,CAACU,GAAG,CAACR,CAAC,IAAIA,CAAC,CAACxC,IAAI,KAAK,MAAM,GAAGwC,CAAC,CAACG,IAAI,GAAG,EAAE,CAAC,CAACM,IAAI,CAAC,EAAE,CAAC;UAC/E,IAAI,CAAC/D,OAAO,CAACa,QAAQ,CAAC;YACpBC,IAAI,EAAE,aAAa;YACnBP,UAAU,EAAEJ,IAAI,CAACI,UAAU;YAC3BQ,OAAO,EAAE,IAAI;YACb8B,QAAQ,EAAEgB,SAAS;YACnBG,MAAM,EAAE5B,WAAW;YACnB6B,QAAQ,EAAExE,eAAe,CAAC2C,WAAW,CAACC,KAAK,EAAED,WAAW,CAACE,MAAM,EAAEnB,KAAK;UACxE,CAAC,CAAC;UACF;QACF;;QAEA;QACA,MAAM+C,WAA8C,GAAG,EAAE;QACzD,KAAK,MAAMV,KAAK,IAAID,UAAU,EAAE;UAC9B,IAAIC,KAAK,CAAC1C,IAAI,KAAK,UAAU,EAAE;;UAE/B;UACA,MAAMqD,SAAS,GAAG3B,YAAY,CAAC4B,UAAU,CAACZ,KAAK,CAAC1B,IAAI,EAAE0B,KAAK,CAACnB,KAAgC,CAAC;UAC7F,IAAI8B,SAAS,CAACE,OAAO,EAAE;YACpBH,WAAW,CAAWR,IAAI,CAAC;cAC1B5C,IAAI,EAAE,aAAa;cACnBwD,WAAW,EAAEd,KAAK,CAACe,EAAE;cACrB3C,OAAO,EAAE,6BAA6BuC,SAAS,CAACK,MAAM,6BAA6B;cACnFC,QAAQ,EAAE;YACZ,CAAC,CAAC;YACF;UACF;UAEA,IAAI,CAACzE,OAAO,CAACa,QAAQ,CAAC;YACpBC,IAAI,EAAE,eAAe;YACrBP,UAAU,EAAEJ,IAAI,CAACI,UAAU;YAC3BoC,KAAK,EAAE;cAAE7B,IAAI,EAAE,YAAY;cAAEgB,IAAI,EAAE0B,KAAK,CAAC1B,IAAI;cAAEO,KAAK,EAAEmB,KAAK,CAACnB;YAAM;UACpE,CAAC,CAAC;UAEF,IAAIqC,MAAc;UAClB,IAAI;YACF,IAAIlB,KAAK,CAAC1B,IAAI,KAAK,aAAa,EAAE;cAChC,MAAMO,KAAK,GAAGmB,KAAK,CAACnB,KAAiD;cACrEqC,MAAM,GAAG,MAAM,IAAI,CAACC,cAAc,CAACtC,KAAK,CAACJ,OAAO,EAAEI,KAAK,CAACH,UAAU,IAAI,KAAK,CAAC;YAC9E,CAAC,MAAM;cACLwC,MAAM,GAAG,iBAAiBlB,KAAK,CAAC1B,IAAI,EAAE;YACxC;UACF,CAAC,CAAC,OAAO8C,GAAG,EAAE;YACZF,MAAM,GAAG,UAAWE,GAAG,CAAWC,OAAO,EAAE;UAC7C;UAEA,IAAI,CAAC7E,OAAO,CAACa,QAAQ,CAAC;YACpBC,IAAI,EAAE,eAAe;YACrBP,UAAU,EAAEJ,IAAI,CAACI,UAAU;YAC3BoC,KAAK,EAAE;cAAE7B,IAAI,EAAE,aAAa;cAAEgB,IAAI,EAAE0B,KAAK,CAAC1B,IAAI;cAAE4C,MAAM,EAAEA,MAAM,CAACI,KAAK,CAAC,CAAC,EAAE,GAAG;YAAE;UAC/E,CAAC,CAAC;UAEDZ,WAAW,CAAWR,IAAI,CAAC;YAC1B5C,IAAI,EAAE,aAAa;YACnBwD,WAAW,EAAEd,KAAK,CAACe,EAAE;YACrB3C,OAAO,EAAE8C;UACX,CAAC,CAAC;QACJ;QAEAjD,QAAQ,CAACiC,IAAI,CAAC;UAAE/B,IAAI,EAAE,MAAM;UAAEC,OAAO,EAAEsC;QAAY,CAAC,CAAC;;QAErD;QACA,MAAMa,SAAS,GAAGvC,YAAY,CAACwC,OAAO,CAAC,CAAC;QACxC,IAAID,SAAS,CAACE,UAAU,EAAE;UACxB,IAAI,CAACjF,OAAO,CAACa,QAAQ,CAAC;YACpBC,IAAI,EAAE,aAAa;YACnBP,UAAU,EAAEJ,IAAI,CAACI,UAAU;YAC3BQ,OAAO,EAAE,KAAK;YACdC,KAAK,EAAE+D,SAAS,CAACF,OAAO,IAAI,0CAA0C;YACtEb,MAAM,EAAE5B,WAAW;YACnB6B,QAAQ,EAAExE,eAAe,CAAC2C,WAAW,CAACC,KAAK,EAAED,WAAW,CAACE,MAAM,EAAEnB,KAAK;UACxE,CAAC,CAAC;UACF;QACF;;QAEA;QACA,MAAM+D,SAAS,GAAGzF,eAAe,CAAC2C,WAAW,CAACC,KAAK,EAAED,WAAW,CAACE,MAAM,EAAEnB,KAAK,CAAC;QAC/E,IAAIhB,IAAI,CAACgF,YAAY,IAAID,SAAS,IAAI/E,IAAI,CAACgF,YAAY,EAAE;UACvD,IAAI,CAACnF,OAAO,CAACa,QAAQ,CAAC;YACpBC,IAAI,EAAE,aAAa;YACnBP,UAAU,EAAEJ,IAAI,CAACI,UAAU;YAC3BQ,OAAO,EAAE,KAAK;YACdC,KAAK,EAAE,wBAAwBkE,SAAS,CAACE,OAAO,CAAC,CAAC,CAAC,QAAQjF,IAAI,CAACgF,YAAY,GAAG;YAC/EnB,MAAM,EAAE5B,WAAW;YACnB6B,QAAQ,EAAEiB;UACZ,CAAC,CAAC;UACF;QACF;MACF;IACF,CAAC,CAAC,OAAON,GAAG,EAAE;MACZ,IAAI,CAAC5E,OAAO,CAACa,QAAQ,CAAC;QACpBC,IAAI,EAAE,aAAa;QACnBP,UAAU,EAAEJ,IAAI,CAACI,UAAU;QAC3BQ,OAAO,EAAE,KAAK;QACdC,KAAK,EAAG4D,GAAG,CAAWC,OAAO;QAC7Bb,MAAM,EAAE5B,WAAW;QACnB6B,QAAQ,EAAExE,eAAe,CAAC2C,WAAW,CAACC,KAAK,EAAED,WAAW,CAACE,MAAM,EAAEnB,KAAK;MACxE,CAAC,CAAC;IACJ,CAAC,SAAS;MACR,IAAI,CAACtB,WAAW,CAACoB,MAAM,CAACd,IAAI,CAACI,UAAU,CAAC;IAC1C;EACF;EAEA8E,MAAMA,CAACC,SAAiB,EAAQ;IAC9B,IAAI,CAACzF,WAAW,CAAC0F,GAAG,CAACD,SAAS,CAAC,EAAElF,KAAK,CAAC,CAAC;EAC1C;EAEAoF,cAAcA,CAAA,EAAa;IACzB,OAAOC,KAAK,CAACC,IAAI,CAAC,IAAI,CAAC7F,WAAW,CAAC8F,IAAI,CAAC,CAAC,CAAC;EAC5C;EAEA,MAAchB,cAAcA,CAAC1C,OAAe,EAAE2D,SAAiB,EAAmB;IAChF;IACA,MAAMC,UAAU,GAAGnG,mBAAmB,CAACuC,OAAO,CAAC;IAC/CtC,gBAAgB,CAACkG,UAAU,EAAE;MAAEC,MAAM,EAAE,IAAI,CAAC7F,MAAM,CAAC8F;IAAQ,CAAC,CAAC;IAC7D,IAAI,CAACF,UAAU,CAACG,OAAO,EAAE;MACvB,OAAO,oBAAoBH,UAAU,CAACrB,MAAM,kBAAkB;IAChE;IAEA,MAAM;MAAEyB;IAAS,CAAC,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC;IACvD,MAAM;MAAEC;IAAU,CAAC,GAAG,MAAM,MAAM,CAAC,WAAW,CAAC;IAC/C,MAAMC,aAAa,GAAGD,SAAS,CAACD,QAAQ,CAAC;;IAEzC;IACA,MAAMG,IAAI,GAAGnE,OAAO,CAACoE,IAAI,CAAC,CAAC,CAACC,KAAK,CAAC,KAAK,CAAC;IACxC,MAAMC,MAAM,GAAGH,IAAI,CAACI,KAAK,CAAC,CAAE;IAE5B,IAAI;MACF,MAAM;QAAEC,MAAM;QAAEC;MAAO,CAAC,GAAG,MAAMP,aAAa,CAACI,MAAM,EAAEH,IAAI,EAAE;QAC3DO,OAAO,EAAEf,SAAS;QAClBgB,GAAG,EAAElG,OAAO,CAACC,GAAG,CAACkG,IAAI;QACrBC,SAAS,EAAE,IAAI,GAAG;MACpB,CAAC,CAAC;MACF,MAAMC,KAAe,GAAG,EAAE;MAC1B,IAAIN,MAAM,EAAEM,KAAK,CAACrD,IAAI,CAAC+C,MAAM,CAAC;MAC9B,IAAIC,MAAM,EAAEK,KAAK,CAACrD,IAAI,CAAC,WAAWgD,MAAM,EAAE,CAAC;MAC3C,OAAOK,KAAK,CAAChD,IAAI,CAAC,IAAI,CAAC,IAAI,aAAa;IAC1C,CAAC,CAAC,OAAOa,GAAQ,EAAE;MACjB,MAAMmC,KAAe,GAAG,EAAE;MAC1B,IAAInC,GAAG,CAAC6B,MAAM,EAAEM,KAAK,CAACrD,IAAI,CAACkB,GAAG,CAAC6B,MAAM,CAAC;MACtC,IAAI7B,GAAG,CAAC8B,MAAM,EAAEK,KAAK,CAACrD,IAAI,CAAC,WAAWkB,GAAG,CAAC8B,MAAM,EAAE,CAAC;MACnD,IAAI,CAACK,KAAK,CAACnD,MAAM,EAAEmD,KAAK,CAACrD,IAAI,CAACkB,GAAG,CAACC,OAAO,CAAC;MAC1C,OAAOkC,KAAK,CAAChD,IAAI,CAAC,IAAI,CAAC,GAAG,gBAAgBa,GAAG,CAACoC,IAAI,IAAI,CAAC,EAAE;IAC3D;EACF;AACF","ignoreList":[]}

package/dist/server/__tests__/gateway-fast-fail.test.js

@@ -0,0 +1,160 @@
+/**
+ * Local Agent Gateway — Fast-Fail on Agent Disconnect
+ *
+ * Tests the fix where pending requests are immediately failed when an agent
+ * disconnects, instead of waiting 65s for individual request timeouts.
+ */
+
+import { describe, it, expect, vi, beforeEach, afterEach } from "vitest";
+
+// ============================================================================
+// Mocks — must be before imports
+// ============================================================================
+
+vi.mock("ws", () => ({
+  WebSocketServer: class {
+    on = vi.fn();
+    handleUpgrade = vi.fn();
+    emit = vi.fn();
+    close = vi.fn();
+  },
+  WebSocket: {
+    OPEN: 1,
+    CLOSED: 3
+  }
+}));
+vi.mock("@supabase/supabase-js", () => ({
+  createClient: vi.fn(() => ({
+    from: vi.fn().mockReturnValue({
+      select: vi.fn().mockReturnValue({
+        eq: vi.fn().mockReturnValue({
+          eq: vi.fn().mockReturnValue({
+            single: vi.fn().mockResolvedValue({
+              data: {
+                id: "key-1",
+                store_id: "store-1",
+                owner_user_id: "user-1"
+              },
+              error: null
+            })
+          })
+        })
+      }),
+      insert: vi.fn().mockReturnValue({
+        then: vi.fn(cb => cb({
+          error: null
+        }))
+      }),
+      update: vi.fn().mockReturnValue({
+        eq: vi.fn().mockReturnValue({
+          then: vi.fn(cb => cb({
+            error: null
+          }))
+        })
+      })
+    })
+  }))
+}));
+import { hasLocalAgent, getAgentInfo, getGatewayStats, shutdownGateway, executeOnLocalAgent, executeToolOnLocalAgent, executeClusterCommand, discoverLocalTools } from "../local-agent-gateway.js";
+
+// ============================================================================
+// Setup
+// ============================================================================
+
+beforeEach(() => {
+  vi.clearAllMocks();
+});
+afterEach(() => {
+  shutdownGateway();
+});
+
+// ============================================================================
+// No-agent error paths (fast-fail before promise creation)
+// ============================================================================
+
+describe("execute functions: no agent connected", () => {
+  it("executeOnLocalAgent returns error immediately", async () => {
+    const result = await executeOnLocalAgent("nonexistent", "echo hello");
+    expect(result.success).toBe(false);
+    expect(result.error).toContain("No local agent");
+  });
+  it("executeToolOnLocalAgent returns error immediately", async () => {
+    const result = await executeToolOnLocalAgent("nonexistent", "tool", {});
+    expect(result.success).toBe(false);
+    expect(result.error).toContain("No local agent");
+  });
+  it("executeClusterCommand returns error immediately", async () => {
+    const result = await executeClusterCommand("nonexistent", {
+      action: "status"
+    });
+    expect(result.success).toBe(false);
+    expect(result.error).toContain("No local agent");
+  });
+  it("discoverLocalTools returns error immediately", async () => {
+    const result = await discoverLocalTools("nonexistent");
+    expect(result.success).toBe(false);
+    expect(result.error).toContain("No local agent");
+  });
+});
+
+// ============================================================================
+// Gateway stats show 0 pending after shutdown
+// ============================================================================
+
+describe("shutdownGateway cleans up pending requests", () => {
+  it("pending_requests is 0 after shutdown", () => {
+    shutdownGateway();
+    const stats = getGatewayStats();
+    expect(stats.pending_requests).toBe(0);
+    expect(stats.total_agents).toBe(0);
+  });
+  it("can be called multiple times safely", () => {
+    expect(() => shutdownGateway()).not.toThrow();
+    expect(() => shutdownGateway()).not.toThrow();
+    expect(() => shutdownGateway()).not.toThrow();
+  });
+});
+
+// ============================================================================
+// Verify send() return type change (boolean) is in effect
+// ============================================================================
+
+describe("send() boolean return and fast-fail path", () => {
+  it("executeOnLocalAgent handles closed connection via send() returning false", async () => {
+    // With no connected agents, the function returns early before even calling send()
+    // This verifies the "no agent" fast-fail path works
+    const t0 = Date.now();
+    const result = await executeOnLocalAgent("store-1", "test");
+    const elapsed = Date.now() - t0;
+    expect(result.success).toBe(false);
+    // Should be nearly instant (< 100ms), not waiting 65s
+    expect(elapsed).toBeLessThan(1000);
+  });
+  it("executeClusterCommand returns immediately for unknown store", async () => {
+    const t0 = Date.now();
+    const result = await executeClusterCommand("store-1", {
+      action: "launch",
+      task: "test"
+    });
+    const elapsed = Date.now() - t0;
+    expect(result.success).toBe(false);
+    expect(elapsed).toBeLessThan(1000);
+  });
+});
+
+// ============================================================================
+// Structural verification
+// ============================================================================
+
+describe("exported functions include all expected APIs", () => {
+  it("all gateway functions exist", () => {
+    expect(typeof hasLocalAgent).toBe("function");
+    expect(typeof getAgentInfo).toBe("function");
+    expect(typeof getGatewayStats).toBe("function");
+    expect(typeof shutdownGateway).toBe("function");
+    expect(typeof executeOnLocalAgent).toBe("function");
+    expect(typeof executeToolOnLocalAgent).toBe("function");
+    expect(typeof executeClusterCommand).toBe("function");
+    expect(typeof discoverLocalTools).toBe("function");
+  });
+});

package/dist/server/__tests__/local-agent-gateway.test.js

@@ -0,0 +1,186 @@
+import { describe, it, expect, vi, beforeEach, afterEach } from "vitest";
+
+// ============================================================================
+// Mocks — use inline classes/objects to avoid hoisting issues
+// ============================================================================
+
+vi.mock("ws", () => {
+  const OPEN = 1;
+  const CLOSED = 3;
+  return {
+    WebSocketServer: class {
+      on = vi.fn();
+      handleUpgrade = vi.fn();
+      emit = vi.fn();
+      close = vi.fn();
+    },
+    WebSocket: {
+      OPEN,
+      CLOSED
+    }
+  };
+});
+vi.mock("@supabase/supabase-js", () => ({
+  createClient: vi.fn(() => ({
+    from: vi.fn().mockReturnValue({
+      select: vi.fn().mockReturnValue({
+        eq: vi.fn().mockReturnValue({
+          eq: vi.fn().mockReturnValue({
+            single: vi.fn().mockResolvedValue({
+              data: {
+                id: "key-1",
+                store_id: "store-1",
+                owner_user_id: "user-1"
+              },
+              error: null
+            })
+          })
+        })
+      }),
+      insert: vi.fn().mockReturnValue({
+        then: vi.fn(cb => cb({
+          error: null
+        }))
+      }),
+      update: vi.fn().mockReturnValue({
+        eq: vi.fn().mockReturnValue({
+          then: vi.fn(cb => cb({
+            error: null
+          }))
+        })
+      })
+    })
+  }))
+}));
+import { hasLocalAgent, getAgentInfo, getGatewayStats, shutdownGateway } from "../local-agent-gateway.js";
+
+// ============================================================================
+// Setup
+// ============================================================================
+
+beforeEach(() => {
+  vi.clearAllMocks();
+});
+afterEach(() => {
+  shutdownGateway();
+});
+
+// ============================================================================
+// hasLocalAgent
+// ============================================================================
+
+describe("hasLocalAgent", () => {
+  it("returns false when no agents are connected", () => {
+    expect(hasLocalAgent("store-1")).toBe(false);
+  });
+  it("is case insensitive for store IDs", () => {
+    expect(hasLocalAgent("STORE-1")).toBe(false);
+    expect(hasLocalAgent("Store-1")).toBe(false);
+  });
+  it("returns false for empty string", () => {
+    expect(hasLocalAgent("")).toBe(false);
+  });
+});
+
+// ============================================================================
+// getAgentInfo
+// ============================================================================
+
+describe("getAgentInfo", () => {
+  it("returns empty array when no agents for store", () => {
+    expect(getAgentInfo("store-1")).toEqual([]);
+  });
+  it("is case insensitive for store IDs", () => {
+    expect(getAgentInfo("STORE-1")).toEqual([]);
+  });
+  it("returns array type", () => {
+    const info = getAgentInfo("any-store");
+    expect(Array.isArray(info)).toBe(true);
+  });
+});
+
+// ============================================================================
+// getGatewayStats
+// ============================================================================
+
+describe("getGatewayStats", () => {
+  it("returns zero stats when no agents connected", () => {
+    const stats = getGatewayStats();
+    expect(stats.total_agents).toBe(0);
+    expect(stats.total_stores).toBe(0);
+    expect(stats.pending_requests).toBe(0);
+    expect(stats.agents_by_store).toEqual({});
+  });
+  it("stats structure has all required fields", () => {
+    const stats = getGatewayStats();
+    expect(stats).toHaveProperty("total_agents");
+    expect(stats).toHaveProperty("total_stores");
+    expect(stats).toHaveProperty("pending_requests");
+    expect(stats).toHaveProperty("agents_by_store");
+    expect(typeof stats.total_agents).toBe("number");
+    expect(typeof stats.total_stores).toBe("number");
+    expect(typeof stats.pending_requests).toBe("number");
+    expect(typeof stats.agents_by_store).toBe("object");
+  });
+});
+
+// ============================================================================
+// shutdownGateway
+// ============================================================================
+
+describe("shutdownGateway", () => {
+  it("can be called multiple times without error", () => {
+    expect(() => shutdownGateway()).not.toThrow();
+    expect(() => shutdownGateway()).not.toThrow();
+  });
+  it("returns zero stats after shutdown", () => {
+    shutdownGateway();
+    const stats = getGatewayStats();
+    expect(stats.total_agents).toBe(0);
+    expect(stats.total_stores).toBe(0);
+    expect(stats.pending_requests).toBe(0);
+  });
+  it("clears agent pool completely", () => {
+    shutdownGateway();
+    expect(hasLocalAgent("store-1")).toBe(false);
+    expect(getAgentInfo("store-1")).toEqual([]);
+  });
+});
+
+// ============================================================================
+// Type exports / function existence
+// ============================================================================
+
+describe("exported functions", () => {
+  it("hasLocalAgent is a function", () => {
+    expect(typeof hasLocalAgent).toBe("function");
+  });
+  it("getAgentInfo is a function", () => {
+    expect(typeof getAgentInfo).toBe("function");
+  });
+  it("getGatewayStats is a function", () => {
+    expect(typeof getGatewayStats).toBe("function");
+  });
+  it("shutdownGateway is a function", () => {
+    expect(typeof shutdownGateway).toBe("function");
+  });
+});
+
+// ============================================================================
+// Edge cases
+// ============================================================================
+
+describe("edge cases", () => {
+  it("handles special characters in store IDs", () => {
+    expect(hasLocalAgent("store-with-dashes")).toBe(false);
+    expect(hasLocalAgent("store_with_underscores")).toBe(false);
+    expect(hasLocalAgent("UPPERCASE")).toBe(false);
+  });
+  it("getAgentInfo returns empty for unknown stores", () => {
+    expect(getAgentInfo("unknown-store-id")).toEqual([]);
+  });
+  it("getGatewayStats agents_by_store is empty object initially", () => {
+    const stats = getGatewayStats();
+    expect(Object.keys(stats.agents_by_store)).toHaveLength(0);
+  });
+});