@rk0429/agentic-relay 0.14.1 → 0.16.0

package/dist/relay.mjs

@@ -446,6 +446,7 @@ function inferFailureReason(stderr, stdout, sdkErrorMetadata) {
   const combined = `${stderr} ${stdout}`.toLowerCase();
   if (combined.includes("timed out") || combined.includes("timeout")) return "timeout";
   if (combined.includes("max turns") || combined.includes("max_turns") || combined.includes("turn limit")) return "max_turns_exhausted";
+  if (combined.includes("429") || combined.includes("capacity_exhausted") || combined.includes("model_capacity_exhausted") || combined.includes("ratelimitexceeded") || combined.includes("resource_exhausted")) return "rate_limit";
   return "adapter_error";
 }
 function buildContextFromEnv() {
@@ -617,7 +618,6 @@ ${input.prompt}`;
   }
   onProgress?.({ stage: "spawning", percent: 10 });
   try {
-    let result;
     const executePromise = (async () => {
       if (input.resumeSessionId) {
         if (!adapter.continueSession) {
@@ -681,7 +681,7 @@ ${input.prompt}`;
         });
       }
     })();
-    result = await executePromise;
+    const result = await executePromise;
     if (result && "_noSession" in result) {
       return {
         sessionId: session.relaySessionId,
@@ -817,7 +817,13 @@ var init_spawn_agent = __esm({
       taskInstructionPath: z2.string().optional().describe(
         "Path to a file containing task instructions. Content is prepended to the prompt. Path is resolved relative to the project root and validated against path traversal."
       ),
-      label: z2.string().optional().describe("Human-readable label for identifying this agent in parallel results and logs.")
+      label: z2.string().optional().describe("Human-readable label for identifying this agent in parallel results and logs."),
+      maxResponseLength: z2.number().int().positive().optional().describe(
+        "Maximum response length in characters. When exceeded, response is truncated or summarized based on responseMode."
+      ),
+      responseMode: z2.enum(["truncate", "summarize"]).optional().describe(
+        "How to handle responses exceeding maxResponseLength. 'truncate' (default) cuts at limit with file save. 'summarize' uses AI to condense."
+      )
     });
   }
 });
@@ -858,11 +864,11 @@ function extractMentionedPaths(stdout) {
   for (const pattern of patterns) {
     let match;
     while ((match = pattern.exec(stdout)) !== null) {
-      let path = match[1];
-      if (path.startsWith("./")) {
-        path = path.slice(2);
+      let path2 = match[1];
+      if (path2.startsWith("./")) {
+        path2 = path2.slice(2);
       }
-      paths.add(path);
+      paths.add(path2);
     }
   }
   return paths;
@@ -897,7 +903,7 @@ async function detectConflicts(before, after, agentResults) {
     }
   }
   const conflicts = Array.from(fileToAgents.entries()).map(
-    ([path, agents]) => ({ path, agents })
+    ([path2, agents]) => ({ path: path2, agents })
   );
   return {
     conflicts,
@@ -1207,8 +1213,128 @@ var init_types = __esm({
   }
 });
 
+// src/mcp-server/response-summarizer.ts
+import Anthropic from "@anthropic-ai/sdk";
+async function summarizeResponse(text, targetLength) {
+  const client = new Anthropic();
+  const message = await client.messages.create({
+    model: "claude-haiku-4-5-20251001",
+    max_tokens: Math.max(1024, Math.ceil(targetLength / 3)),
+    messages: [
+      {
+        role: "user",
+        content: `Summarize the following text in ${targetLength} characters or less. Prioritize:
+1. Key deliverables, artifacts, and file changes
+2. Important decisions and conclusions
+3. Action items and next steps
+
+Preserve technical details (file paths, function names, error messages) when possible.
+
+<text>
+${text}
+</text>`
+      }
+    ]
+  });
+  const content = message.content[0];
+  if (content?.type !== "text") {
+    throw new Error("Unexpected response format from summarization API");
+  }
+  return content.text;
+}
+var init_response_summarizer = __esm({
+  "src/mcp-server/response-summarizer.ts"() {
+    "use strict";
+  }
+});
+
 // src/mcp-server/response-formatter.ts
-function formatSpawnAgentResponse(result) {
+import * as fs from "fs";
+import * as path from "path";
+import * as os from "os";
+function saveFullResponse(content, sessionId) {
+  try {
+    const dir = sessionId ? path.join(RESPONSE_DIR, sessionId) : path.join(RESPONSE_DIR, `anonymous-${Date.now()}`);
+    fs.mkdirSync(dir, { recursive: true });
+    const filePath = path.join(dir, "response.txt");
+    fs.writeFileSync(filePath, content, "utf-8");
+    try {
+      const entries = fs.readdirSync(RESPONSE_DIR).map((name) => {
+        const fullPath = path.join(RESPONSE_DIR, name);
+        try {
+          const stat = fs.statSync(fullPath);
+          return { name, fullPath, mtimeMs: stat.mtimeMs };
+        } catch {
+          return null;
+        }
+      }).filter((e) => e !== null).sort((a, b) => a.mtimeMs - b.mtimeMs);
+      if (entries.length > MAX_SAVED_RESPONSES) {
+        const toRemove = entries.slice(0, entries.length - MAX_SAVED_RESPONSES);
+        for (const entry of toRemove) {
+          fs.rmSync(entry.fullPath, { recursive: true, force: true });
+        }
+      }
+    } catch {
+    }
+    return filePath;
+  } catch {
+    return void 0;
+  }
+}
+async function processStdout(stdout, options) {
+  const maxLength = options?.maxLength ?? STDOUT_DEFAULT_MAX_LENGTH;
+  const mode = options?.mode ?? "truncate";
+  if (stdout.length <= maxLength) {
+    return { text: stdout, truncated: false };
+  }
+  const savedFilePath = saveFullResponse(stdout, options?.sessionId);
+  const savedInfo = savedFilePath ? ` Full response saved to: ${savedFilePath}` : "";
+  if (mode === "summarize") {
+    try {
+      const summary = await summarizeResponse(stdout, maxLength);
+      const footer2 = `
+
+[RESPONSE SUMMARIZED: original ${stdout.length} chars exceeded limit ${maxLength} chars.${savedInfo}]`;
+      return { text: summary + footer2, truncated: true, savedFilePath };
+    } catch {
+    }
+  }
+  const truncated = stdout.slice(0, maxLength);
+  const footer = `
+
+[RESPONSE TRUNCATED: original ${stdout.length} chars exceeded limit ${maxLength} chars.${savedInfo}]`;
+  return { text: truncated + footer, truncated: true, savedFilePath };
+}
+function truncateStderr(stderr) {
+  if (!stderr) return stderr;
+  const blocks = stderr.split("\n\n");
+  if (blocks.length > 1) {
+    const seen = /* @__PURE__ */ new Set();
+    const unique = [];
+    let duplicateCount = 0;
+    for (const block of blocks) {
+      const trimmed = block.trim();
+      if (!trimmed) continue;
+      if (seen.has(trimmed)) {
+        duplicateCount++;
+      } else {
+        seen.add(trimmed);
+        unique.push(block);
+      }
+    }
+    if (duplicateCount > 0) {
+      stderr = unique.join("\n\n") + `
+[${duplicateCount} duplicate block(s) removed]`;
+    }
+  }
+  if (stderr.length > STDERR_MAX_LENGTH) {
+    const totalLength = stderr.length;
+    stderr = stderr.slice(0, STDERR_MAX_LENGTH) + `
+[truncated, ${totalLength} total chars]`;
+  }
+  return stderr;
+}
+async function formatSpawnAgentResponse(result, options) {
   const isError = result.exitCode !== 0;
   let text;
   if (isError) {
@@ -1216,12 +1342,13 @@ function formatSpawnAgentResponse(result) {
     text = `FAILED (exit ${result.exitCode}${reasonPart})`;
     if (result.stderr) {
       text += `
-${result.stderr}`;
+${truncateStderr(result.stderr)}`;
     }
   } else {
+    const processed = await processStdout(result.stdout, options);
     text = `Session: ${result.sessionId}
 
-${result.stdout}`;
+${processed.text}`;
   }
   if (result.metadata) {
     text += `
@@ -1232,7 +1359,7 @@ ${JSON.stringify(result.metadata, null, 2)}
   }
   return { text, isError };
 }
-function formatParallelResponse(result) {
+async function formatParallelResponse(result, options, perAgentOptions) {
   const isError = result.failureCount === result.totalCount;
   const parts = [];
   if (result.hasConflicts && result.conflicts) {
@@ -1252,11 +1379,13 @@ function formatParallelResponse(result) {
     const duration = r.metadata?.durationMs !== void 0 ? `${(r.metadata.durationMs / 1e3).toFixed(1)}s` : "?s";
     if (r.exitCode === 0) {
       parts.push(`--- Agent ${r.index}${labelPart} (${backend}, ${duration}) SUCCESS ---`);
-      parts.push(r.stdout || "(no output)");
+      const agentOptions = perAgentOptions?.get(r.index) ?? options;
+      const processed = await processStdout(r.stdout || "", agentOptions);
+      parts.push(processed.text || "(no output)");
     } else {
       const reasonPart = r.failureReason ? `, reason: ${r.failureReason}` : "";
       parts.push(`--- Agent ${r.index}${labelPart} FAILED (${backend}, ${duration}${reasonPart}) ---`);
-      parts.push(r.stderr || r.error || "(no output)");
+      parts.push(truncateStderr(r.stderr || r.error || "") || "(no output)");
     }
     parts.push("");
   }
@@ -1265,9 +1394,15 @@ ${JSON.stringify(result, null, 2)}
 </metadata>`);
   return { text: parts.join("\n"), isError };
 }
+var STDERR_MAX_LENGTH, STDOUT_DEFAULT_MAX_LENGTH, MAX_SAVED_RESPONSES, RESPONSE_DIR;
 var init_response_formatter = __esm({
   "src/mcp-server/response-formatter.ts"() {
     "use strict";
+    init_response_summarizer();
+    STDERR_MAX_LENGTH = 2e3;
+    STDOUT_DEFAULT_MAX_LENGTH = 2e4;
+    MAX_SAVED_RESPONSES = 100;
+    RESPONSE_DIR = path.join(os.tmpdir(), "agentic-relay-responses");
   }
 });
 
@@ -1314,15 +1449,16 @@ var init_server = __esm({
     };
     MAX_CHILD_HTTP_SESSIONS = 100;
     RelayMCPServer = class {
-      constructor(registry2, sessionManager2, guardConfig, hooksEngine2, contextMonitor2) {
+      constructor(registry2, sessionManager2, guardConfig, hooksEngine2, contextMonitor2, defaultMaxResponseLength) {
         this.registry = registry2;
         this.sessionManager = sessionManager2;
         this.hooksEngine = hooksEngine2;
         this.contextMonitor = contextMonitor2;
+        this.defaultMaxResponseLength = defaultMaxResponseLength;
         this.guard = new RecursionGuard(guardConfig);
         this.backendSelector = new BackendSelector();
         this.server = new McpServer(
-          { name: "agentic-relay", version: "0.14.1" },
+          { name: "agentic-relay", version: "0.16.0" },
           createMcpServerOptions()
         );
         this.registerTools(this.server);
@@ -1357,7 +1493,12 @@ var init_server = __esm({
               this.backendSelector,
               this._childHttpUrl
             );
-            const { text, isError } = formatSpawnAgentResponse(result);
+            const controlOptions = {
+              maxLength: params.maxResponseLength ?? this.defaultMaxResponseLength ?? STDOUT_DEFAULT_MAX_LENGTH,
+              mode: params.responseMode ?? "truncate",
+              sessionId: result.sessionId
+            };
+            const { text, isError } = await formatSpawnAgentResponse(result, controlOptions);
             const callToolResult = {
               content: [{ type: "text", text }],
               isError
@@ -1404,7 +1545,11 @@ var init_server = __esm({
               this.backendSelector,
               this._childHttpUrl
             );
-            const { text, isError } = formatParallelResponse(result);
+            const controlOptions = {
+              maxLength: this.defaultMaxResponseLength ?? STDOUT_DEFAULT_MAX_LENGTH,
+              mode: "truncate"
+            };
+            const { text, isError } = await formatParallelResponse(result, controlOptions);
             const callToolResult = {
               content: [{ type: "text", text }],
               isError
@@ -1520,7 +1665,11 @@ var init_server = __esm({
                 this.backendSelector,
                 this._childHttpUrl
               );
-              const { text, isError } = formatParallelResponse(result);
+              const controlOptions = {
+                maxLength: this.defaultMaxResponseLength ?? STDOUT_DEFAULT_MAX_LENGTH,
+                mode: "truncate"
+              };
+              const { text, isError } = await formatParallelResponse(result, controlOptions);
               return {
                 content: [{ type: "text", text }],
                 isError
@@ -1684,7 +1833,7 @@ var init_server = __esm({
             sessionIdGenerator: () => randomUUID()
           });
           const server = new McpServer(
-            { name: "agentic-relay", version: "0.14.1" },
+            { name: "agentic-relay", version: "0.16.0" },
             createMcpServerOptions()
           );
           this.registerTools(server);
@@ -1740,7 +1889,7 @@ var init_server = __esm({
 
 // src/bin/relay.ts
 import { defineCommand as defineCommand10, runMain } from "citty";
-import { join as join9 } from "path";
+import { join as join10 } from "path";
 import { homedir as homedir6 } from "os";
 
 // src/infrastructure/process-manager.ts
@@ -2856,6 +3005,12 @@ init_logger();
 import { readFile as readFile3, writeFile as writeFile3, mkdir as mkdir3 } from "fs/promises";
 import { homedir as homedir3 } from "os";
 import { join as join3, dirname as dirname3 } from "path";
+var GEMINI_FALLBACK_CHAIN = [
+  "gemini-3.1-pro-preview",
+  "gemini-2.5-pro",
+  "gemini-2.5-flash"
+];
+var RATE_LIMIT_PATTERN = /429|MODEL_CAPACITY_EXHAUSTED|rateLimitExceeded|RESOURCE_EXHAUSTED/i;
 var GeminiAdapter = class extends BaseAdapter {
   id = "gemini";
   command = "gemini";
@@ -2935,6 +3090,21 @@ ${prompt}`;
     if (!flags.prompt) {
       throw new Error("execute requires a prompt (-p flag)");
     }
+    if (flags.model) {
+      return this.executeOnce(flags);
+    }
+    let lastResult;
+    for (const model of GEMINI_FALLBACK_CHAIN) {
+      const result = await this.executeOnce({ ...flags, model });
+      if (result.exitCode === 0 || !RATE_LIMIT_PATTERN.test(result.stderr)) {
+        return result;
+      }
+      logger.warn(`Gemini model ${model} rate limited, trying next fallback...`);
+      lastResult = result;
+    }
+    return lastResult;
+  }
+  async executeOnce(flags) {
     const effectivePrompt = this.buildEffectivePrompt(flags);
     const args = [];
     if (flags.model) {
@@ -3263,7 +3433,8 @@ var relayConfigSchema = z.object({
   mcpServerMode: z.object({
     maxDepth: z.number().int().positive(),
     maxCallsPerSession: z.number().int().positive(),
-    timeoutSec: z.number().positive()
+    timeoutSec: z.number().positive(),
+    maxResponseLength: z.number().int().positive().optional()
   }).optional(),
   telemetry: z.object({
     enabled: z.boolean()
@@ -3292,8 +3463,8 @@ function deepMerge(target, source) {
 function isPlainObject(value) {
   return typeof value === "object" && value !== null && !Array.isArray(value) && Object.getPrototypeOf(value) === Object.prototype;
 }
-function getByPath(obj, path) {
-  const parts = path.split(".");
+function getByPath(obj, path2) {
+  const parts = path2.split(".");
   let current = obj;
   for (const part of parts) {
     if (!isPlainObject(current)) return void 0;
@@ -3301,8 +3472,8 @@ function getByPath(obj, path) {
   }
   return current;
 }
-function setByPath(obj, path, value) {
-  const parts = path.split(".");
+function setByPath(obj, path2, value) {
+  const parts = path2.split(".");
   let current = obj;
   for (let i = 0; i < parts.length - 1; i++) {
     const part = parts[i];
@@ -4494,6 +4665,7 @@ function createMCPCommand(configManager2, registry2, sessionManager2, hooksEngin
             return;
           }
           let guardConfig;
+          let defaultMaxResponseLength;
           try {
             const config = await configManager2.getConfig();
             if (config.mcpServerMode) {
@@ -4502,6 +4674,7 @@ function createMCPCommand(configManager2, registry2, sessionManager2, hooksEngin
                 maxCallsPerSession: config.mcpServerMode.maxCallsPerSession ?? 20,
                 timeoutSec: config.mcpServerMode.timeoutSec ?? 86400
               };
+              defaultMaxResponseLength = config.mcpServerMode.maxResponseLength;
             }
           } catch {
           }
@@ -4512,7 +4685,8 @@ function createMCPCommand(configManager2, registry2, sessionManager2, hooksEngin
             sessionManager2,
             guardConfig,
             hooksEngine2,
-            contextMonitor2
+            contextMonitor2,
+            defaultMaxResponseLength
           );
           await server.start({ transport, port });
         }
@@ -4674,7 +4848,7 @@ function createVersionCommand(registry2) {
       description: "Show relay and backend versions"
     },
     async run() {
-      const relayVersion = "0.14.1";
+      const relayVersion = "0.16.0";
       console.log(`agentic-relay v${relayVersion}`);
       console.log("");
       console.log("Backends:");
@@ -4699,7 +4873,7 @@ function createVersionCommand(registry2) {
 // src/commands/doctor.ts
 import { defineCommand as defineCommand8 } from "citty";
 import { access, constants, readdir as readdir2 } from "fs/promises";
-import { join as join7 } from "path";
+import { join as join8 } from "path";
 import { homedir as homedir5 } from "os";
 import { execFile as execFile2 } from "child_process";
 import { promisify as promisify2 } from "util";
@@ -4760,8 +4934,8 @@ async function checkConfig(configManager2) {
   }
 }
 async function checkSessionsDir() {
-  const relayHome2 = process.env["RELAY_HOME"] ?? join7(homedir5(), ".relay");
-  const sessionsDir = join7(relayHome2, "sessions");
+  const relayHome2 = process.env["RELAY_HOME"] ?? join8(homedir5(), ".relay");
+  const sessionsDir = join8(relayHome2, "sessions");
   try {
     await access(sessionsDir, constants.W_OK);
     return {
@@ -4874,8 +5048,8 @@ async function checkBackendAuthEnv() {
   return results;
 }
 async function checkSessionsDiskUsage() {
-  const relayHome2 = process.env["RELAY_HOME"] ?? join7(homedir5(), ".relay");
-  const sessionsDir = join7(relayHome2, "sessions");
+  const relayHome2 = process.env["RELAY_HOME"] ?? join8(homedir5(), ".relay");
+  const sessionsDir = join8(relayHome2, "sessions");
   try {
     const entries = await readdir2(sessionsDir);
     const fileCount = entries.length;
@@ -4949,7 +5123,7 @@ function createDoctorCommand(registry2, configManager2) {
 init_logger();
 import { defineCommand as defineCommand9 } from "citty";
 import { mkdir as mkdir6, writeFile as writeFile6, access as access2, readFile as readFile6 } from "fs/promises";
-import { join as join8 } from "path";
+import { join as join9 } from "path";
 var DEFAULT_CONFIG2 = {
   defaultBackend: "claude",
   backends: {},
@@ -4963,8 +5137,8 @@ function createInitCommand() {
     },
     async run() {
       const projectDir = process.cwd();
-      const relayDir = join8(projectDir, ".relay");
-      const configPath = join8(relayDir, "config.json");
+      const relayDir = join9(projectDir, ".relay");
+      const configPath = join9(relayDir, "config.json");
       try {
         await access2(relayDir);
         logger.info(
@@ -4980,7 +5154,7 @@ function createInitCommand() {
         "utf-8"
       );
       logger.success(`Created ${configPath}`);
-      const gitignorePath = join8(projectDir, ".gitignore");
+      const gitignorePath = join9(projectDir, ".gitignore");
       try {
         const gitignoreContent = await readFile6(gitignorePath, "utf-8");
         if (!gitignoreContent.includes(".relay/config.local.json")) {
@@ -5006,8 +5180,8 @@ registry.registerLazy("claude", () => new ClaudeAdapter(processManager));
 registry.registerLazy("codex", () => new CodexAdapter(processManager));
 registry.registerLazy("gemini", () => new GeminiAdapter(processManager));
 var sessionManager = new SessionManager();
-var relayHome = process.env["RELAY_HOME"] ?? join9(homedir6(), ".relay");
-var projectRelayDir = join9(process.cwd(), ".relay");
+var relayHome = process.env["RELAY_HOME"] ?? join10(homedir6(), ".relay");
+var projectRelayDir = join10(process.cwd(), ".relay");
 var configManager = new ConfigManager(relayHome, projectRelayDir);
 var authManager = new AuthManager(registry);
 var eventBus = new EventBus();
@@ -5024,7 +5198,7 @@ void configManager.getConfig().then((config) => {
 var main = defineCommand10({
   meta: {
     name: "relay",
-    version: "0.14.1",
+    version: "0.16.0",
     description: "Unified CLI proxy for Claude Code, Codex CLI, and Gemini CLI"
   },
   subCommands: {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@rk0429/agentic-relay",
-  "version": "0.14.1",
+  "version": "0.16.0",
   "description": "Unified CLI proxy for Claude Code, Codex CLI, and Gemini CLI with MCP-based multi-layer sub-agent orchestration",
   "type": "module",
   "license": "Apache-2.0",
@@ -50,6 +50,7 @@
   },
   "dependencies": {
     "@anthropic-ai/claude-agent-sdk": "^0.2.59",
+    "@anthropic-ai/sdk": "^0.78.0",
     "@modelcontextprotocol/sdk": "^1.27.1",
     "@openai/codex-sdk": "^0.105.0",
     "citty": "^0.1.6",