@mindstudio-ai/remy 0.1.273 → 0.1.275

package/dist/headless.js

@@ -497,6 +497,7 @@ var ALLOWED_MODELS_BY_TYPE = {
     "kimi-k3",
     "deepseek-v4-flash-0731",
     "qwen3.8-2.4t-a95b-deepinfra",
+    "qwen3.8-27b-deepinfra",
     "minimax-m3"
   ]
   // vision: undefined — unconstrained
@@ -3300,18 +3301,70 @@ ${partial}` : "[INTERRUPTED] Tool execution was stopped.";
   }
 };
 
-// src/toolResultCap.ts
+// src/historyLimits.ts
 var MAX_TOOL_RESULT_BYTES = 256 * 1024;
-function capToolResult(result) {
+function capToolResult(result, maxBytes = MAX_TOOL_RESULT_BYTES) {
   const total = Buffer.byteLength(result, "utf-8");
-  if (total <= MAX_TOOL_RESULT_BYTES) {
+  if (total <= maxBytes) {
     return result;
   }
-  const head = Buffer.from(result, "utf-8").subarray(0, MAX_TOOL_RESULT_BYTES).toString("utf-8");
+  const head = Buffer.from(result, "utf-8").subarray(0, maxBytes).toString("utf-8");
   return head + `
 
-(tool result truncated at ${(MAX_TOOL_RESULT_BYTES / 1024).toFixed(0)}KB of ${(total / 1024).toFixed(0)}KB \u2014 too large to keep in context. Narrow the call (select fewer fields, paginate, or query a subset) instead of fetching everything.)`;
+(tool result truncated at ${(maxBytes / 1024).toFixed(0)}KB of ${(total / 1024).toFixed(0)}KB \u2014 too large to keep in context. Narrow the call (select fewer fields, paginate, or query a subset) instead of fetching everything.)`;
 }
+var MAX_SUBAGENT_RESULT_BYTES = 32 * 1024;
+var MAX_SUBAGENT_TRANSCRIPT_BYTES = 512 * 1024;
+function capSubAgentTranscript(messages) {
+  for (const msg of messages) {
+    capMessageForHistory(msg, MAX_SUBAGENT_RESULT_BYTES);
+  }
+  const sizes = messages.map(
+    (m) => Buffer.byteLength(JSON.stringify(m), "utf-8") + 1
+  );
+  let total = sizes.reduce((a, b) => a + b, 0);
+  if (total <= MAX_SUBAGENT_TRANSCRIPT_BYTES) {
+    return messages;
+  }
+  let start = 0;
+  while (start < messages.length - 1 && total > MAX_SUBAGENT_TRANSCRIPT_BYTES) {
+    total -= sizes[start];
+    start++;
+  }
+  while (start < messages.length - 1 && messages[start].role === "user" && messages[start].toolCallId) {
+    start++;
+  }
+  return messages.slice(start);
+}
+function capMessageForHistory(msg, maxBytes = MAX_TOOL_RESULT_BYTES) {
+  if (msg.role === "assistant" && Array.isArray(msg.content)) {
+    for (const block of msg.content) {
+      if (block.type !== "tool") {
+        continue;
+      }
+      if (typeof block.result === "string") {
+        block.result = capToolResult(block.result, maxBytes);
+      }
+      if (typeof block.backgroundResult === "string") {
+        block.backgroundResult = capToolResult(
+          block.backgroundResult,
+          maxBytes
+        );
+      }
+      if (Array.isArray(block.subAgentMessages)) {
+        block.subAgentMessages = capSubAgentTranscript(block.subAgentMessages);
+      }
+    }
+  } else if (msg.role === "user" && msg.toolCallId && typeof msg.content === "string") {
+    msg.content = capToolResult(msg.content, maxBytes);
+  }
+}
+var HISTORY_PAGE_MAX_BYTES = 2 * 1024 * 1024;
+var HISTORY_DEFAULT_LIMIT = 500;
+var HISTORY_MAX_LIMIT = 2e3;
+var ROTATE_THRESHOLD_BYTES = 32 * 1024 * 1024;
+var RETAIN_TAIL_BYTES = 16 * 1024 * 1024;
+var ARCHIVE_RETENTION_BYTES = 64 * 1024 * 1024;
 
 // src/statusWatcher.ts
 var INTERNAL_PAYLOAD_MARKERS = [
@@ -3943,7 +3996,7 @@ ${partial}` : "[INTERRUPTED] Agent was interrupted before producing output.",
             block.completedAt = Date.now();
             const innerMsgs = subAgentMessages.get(r.id);
             if (innerMsgs) {
-              block.subAgentMessages = innerMsgs;
+              block.subAgentMessages = capSubAgentTranscript(innerMsgs);
             }
             if (captureArtifacts?.includes(block.name) && !r.isError) {
               try {
@@ -6676,14 +6729,9 @@ import path11 from "path";
 var log10 = createLogger("session");
 var SESSION_FILE = ".remy-session.json";
 var ARCHIVE_DIR = ".logs/sessions";
-var ROTATE_THRESHOLD_BYTES = 32 * 1024 * 1024;
-var RETAIN_TAIL_BYTES = 16 * 1024 * 1024;
-var ARCHIVE_RETENTION_BYTES = 64 * 1024 * 1024;
 var ARCHIVE_NAME_RE = /^(cleared|rotated)-.*\.json$/;
 var archiveSortKey = (name) => name.replace(/^(cleared|rotated)-/, "");
 var ARCHIVE_COUNT_RE = /\.c(\d+)\.json$/;
-var HISTORY_DEFAULT_LIMIT = 500;
-var HISTORY_MAX_LIMIT = 2e3;
 var archiveCountCache = /* @__PURE__ */ new Map();
 var archiveMsgCache = /* @__PURE__ */ new Map();
 var ARCHIVE_MSG_CACHE_MAX = 3;
@@ -6713,22 +6761,11 @@ function loadSession(state) {
   }
   return false;
 }
-function capOversizedResults(msg) {
-  if (msg.role === "assistant" && Array.isArray(msg.content)) {
-    for (const block of msg.content) {
-      if (block.type === "tool" && typeof block.result === "string") {
-        block.result = capToolResult(block.result);
-      }
-    }
-  } else if (msg.role === "user" && msg.toolCallId && typeof msg.content === "string") {
-    msg.content = capToolResult(msg.content);
-  }
-}
 function sanitizeMessages(messages) {
   const result = [];
   for (let i = 0; i < messages.length; i++) {
     const msg = messages[i];
-    capOversizedResults(msg);
+    capMessageForHistory(msg);
     result.push(msg);
     if (msg.role !== "assistant" || !Array.isArray(msg.content)) {
       continue;
@@ -6840,6 +6877,9 @@ function parseArchive(name) {
     const raw = fs21.readFileSync(path11.join(ARCHIVE_DIR, name), "utf-8");
     const data = JSON.parse(raw);
     const messages = Array.isArray(data?.messages) ? data.messages : [];
+    for (const msg of messages) {
+      capMessageForHistory(msg);
+    }
     archiveCountCache.set(name, messages.length);
     archiveMsgCache.set(name, messages);
     while (archiveMsgCache.size > ARCHIVE_MSG_CACHE_MAX) {
@@ -6950,6 +6990,29 @@ function getHistoryPage(state, opts) {
       messages.push(state.messages[i]);
     }
   }
+  if (messages.length > 1) {
+    let bytes = 0;
+    let cut = 0;
+    for (let i = messages.length - 1; i >= 0; i--) {
+      bytes += Buffer.byteLength(JSON.stringify(messages[i]), "utf-8") + 1;
+      if (bytes > HISTORY_PAGE_MAX_BYTES && i < messages.length - 1) {
+        cut = i + 1;
+        break;
+      }
+    }
+    if (cut > 0) {
+      while (cut < messages.length - 1 && messages[cut]?.role === "user" && messages[cut]?.toolCallId) {
+        cut++;
+      }
+      messages.splice(0, cut);
+      startIndex += cut;
+      log10.info("History page trimmed to byte budget", {
+        dropped: cut,
+        kept: messages.length,
+        startIndex
+      });
+    }
+  }
   return { messages, startIndex, endIndex, totalMessageCount: total };
 }
 function rotate(state) {
@@ -7844,6 +7907,8 @@ async function runTurn(params) {
   let lastCallInputTokens = 0;
   let lastCallCacheCreation = 0;
   let lastCallCacheRead = 0;
+  let abnormalStopRecoveries = 0;
+  const MAX_ABNORMAL_STOP_RECOVERIES = 2;
   const statusWatcher = isFirstMessage ? { stop() {
   }, pause() {
   }, resume() {
@@ -8187,6 +8252,18 @@ async function runTurn(params) {
         });
       }
       const toolCalls = getToolCalls(contentBlocks);
+      if (toolCalls.length === 0 && (stopReason === "repetition" || stopReason === "max_tokens") && abnormalStopRecoveries < MAX_ABNORMAL_STOP_RECOVERIES && !signal?.aborted) {
+        abnormalStopRecoveries++;
+        log14.warn("Abnormal stop \u2014 nudging model to continue", {
+          requestId,
+          stopReason,
+          attempt: abnormalStopRecoveries
+        });
+        const nudge = "Your previous response was cut off \u2014 it degenerated into repeated text or hit the output limit, and the repeated portion was removed. Reassess where you are in the task and continue from where you left off. Prefer a tool call over restating what you were about to do.";
+        state.messages.push({ role: "user", content: nudge, hidden: true });
+        onEvent({ type: "user_message", text: nudge, hidden: true });
+        continue;
+      }
       if (stopReason !== "tool_use" || toolCalls.length === 0) {
         statusWatcher.stop();
         saveSession(state);
@@ -8353,7 +8430,7 @@ async function runTurn(params) {
           block.completedAt = Date.now();
           const msgs = subAgentMessages.get(r.id);
           if (msgs) {
-            block.subAgentMessages = msgs;
+            block.subAgentMessages = capSubAgentTranscript(msgs);
           }
         }
       }

package/dist/index.js

@@ -2237,6 +2237,7 @@ var init_surfaces = __esm({
         "kimi-k3",
         "deepseek-v4-flash-0731",
         "qwen3.8-2.4t-a95b-deepinfra",
+        "qwen3.8-27b-deepinfra",
         "minimax-m3"
       ]
       // vision: undefined — unconstrained
@@ -2408,22 +2409,74 @@ var init_cleanMessages = __esm({
   }
 });
 
-// src/toolResultCap.ts
-function capToolResult(result) {
+// src/historyLimits.ts
+function capToolResult(result, maxBytes = MAX_TOOL_RESULT_BYTES) {
   const total = Buffer.byteLength(result, "utf-8");
-  if (total <= MAX_TOOL_RESULT_BYTES) {
+  if (total <= maxBytes) {
     return result;
   }
-  const head = Buffer.from(result, "utf-8").subarray(0, MAX_TOOL_RESULT_BYTES).toString("utf-8");
+  const head = Buffer.from(result, "utf-8").subarray(0, maxBytes).toString("utf-8");
   return head + `
 
-(tool result truncated at ${(MAX_TOOL_RESULT_BYTES / 1024).toFixed(0)}KB of ${(total / 1024).toFixed(0)}KB \u2014 too large to keep in context. Narrow the call (select fewer fields, paginate, or query a subset) instead of fetching everything.)`;
+(tool result truncated at ${(maxBytes / 1024).toFixed(0)}KB of ${(total / 1024).toFixed(0)}KB \u2014 too large to keep in context. Narrow the call (select fewer fields, paginate, or query a subset) instead of fetching everything.)`;
 }
-var MAX_TOOL_RESULT_BYTES;
-var init_toolResultCap = __esm({
-  "src/toolResultCap.ts"() {
+function capSubAgentTranscript(messages) {
+  for (const msg of messages) {
+    capMessageForHistory(msg, MAX_SUBAGENT_RESULT_BYTES);
+  }
+  const sizes = messages.map(
+    (m) => Buffer.byteLength(JSON.stringify(m), "utf-8") + 1
+  );
+  let total = sizes.reduce((a, b) => a + b, 0);
+  if (total <= MAX_SUBAGENT_TRANSCRIPT_BYTES) {
+    return messages;
+  }
+  let start = 0;
+  while (start < messages.length - 1 && total > MAX_SUBAGENT_TRANSCRIPT_BYTES) {
+    total -= sizes[start];
+    start++;
+  }
+  while (start < messages.length - 1 && messages[start].role === "user" && messages[start].toolCallId) {
+    start++;
+  }
+  return messages.slice(start);
+}
+function capMessageForHistory(msg, maxBytes = MAX_TOOL_RESULT_BYTES) {
+  if (msg.role === "assistant" && Array.isArray(msg.content)) {
+    for (const block of msg.content) {
+      if (block.type !== "tool") {
+        continue;
+      }
+      if (typeof block.result === "string") {
+        block.result = capToolResult(block.result, maxBytes);
+      }
+      if (typeof block.backgroundResult === "string") {
+        block.backgroundResult = capToolResult(
+          block.backgroundResult,
+          maxBytes
+        );
+      }
+      if (Array.isArray(block.subAgentMessages)) {
+        block.subAgentMessages = capSubAgentTranscript(block.subAgentMessages);
+      }
+    }
+  } else if (msg.role === "user" && msg.toolCallId && typeof msg.content === "string") {
+    msg.content = capToolResult(msg.content, maxBytes);
+  }
+}
+var MAX_TOOL_RESULT_BYTES, MAX_SUBAGENT_RESULT_BYTES, MAX_SUBAGENT_TRANSCRIPT_BYTES, HISTORY_PAGE_MAX_BYTES, HISTORY_DEFAULT_LIMIT, HISTORY_MAX_LIMIT, ROTATE_THRESHOLD_BYTES, RETAIN_TAIL_BYTES, ARCHIVE_RETENTION_BYTES;
+var init_historyLimits = __esm({
+  "src/historyLimits.ts"() {
     "use strict";
     MAX_TOOL_RESULT_BYTES = 256 * 1024;
+    MAX_SUBAGENT_RESULT_BYTES = 32 * 1024;
+    MAX_SUBAGENT_TRANSCRIPT_BYTES = 512 * 1024;
+    HISTORY_PAGE_MAX_BYTES = 2 * 1024 * 1024;
+    HISTORY_DEFAULT_LIMIT = 500;
+    HISTORY_MAX_LIMIT = 2e3;
+    ROTATE_THRESHOLD_BYTES = 32 * 1024 * 1024;
+    RETAIN_TAIL_BYTES = 16 * 1024 * 1024;
+    ARCHIVE_RETENTION_BYTES = 64 * 1024 * 1024;
   }
 });
 
@@ -2456,22 +2509,11 @@ function loadSession(state) {
   }
   return false;
 }
-function capOversizedResults(msg) {
-  if (msg.role === "assistant" && Array.isArray(msg.content)) {
-    for (const block of msg.content) {
-      if (block.type === "tool" && typeof block.result === "string") {
-        block.result = capToolResult(block.result);
-      }
-    }
-  } else if (msg.role === "user" && msg.toolCallId && typeof msg.content === "string") {
-    msg.content = capToolResult(msg.content);
-  }
-}
 function sanitizeMessages(messages) {
   const result = [];
   for (let i = 0; i < messages.length; i++) {
     const msg = messages[i];
-    capOversizedResults(msg);
+    capMessageForHistory(msg);
     result.push(msg);
     if (msg.role !== "assistant" || !Array.isArray(msg.content)) {
       continue;
@@ -2583,6 +2625,9 @@ function parseArchive(name) {
     const raw = fs10.readFileSync(path4.join(ARCHIVE_DIR, name), "utf-8");
     const data = JSON.parse(raw);
     const messages = Array.isArray(data?.messages) ? data.messages : [];
+    for (const msg of messages) {
+      capMessageForHistory(msg);
+    }
     archiveCountCache.set(name, messages.length);
     archiveMsgCache.set(name, messages);
     while (archiveMsgCache.size > ARCHIVE_MSG_CACHE_MAX) {
@@ -2693,6 +2738,29 @@ function getHistoryPage(state, opts) {
       messages.push(state.messages[i]);
     }
   }
+  if (messages.length > 1) {
+    let bytes = 0;
+    let cut = 0;
+    for (let i = messages.length - 1; i >= 0; i--) {
+      bytes += Buffer.byteLength(JSON.stringify(messages[i]), "utf-8") + 1;
+      if (bytes > HISTORY_PAGE_MAX_BYTES && i < messages.length - 1) {
+        cut = i + 1;
+        break;
+      }
+    }
+    if (cut > 0) {
+      while (cut < messages.length - 1 && messages[cut]?.role === "user" && messages[cut]?.toolCallId) {
+        cut++;
+      }
+      messages.splice(0, cut);
+      startIndex += cut;
+      log3.info("History page trimmed to byte budget", {
+        dropped: cut,
+        kept: messages.length,
+        startIndex
+      });
+    }
+  }
   return { messages, startIndex, endIndex, totalMessageCount: total };
 }
 function rotate(state) {
@@ -2754,7 +2822,7 @@ function clearSession(state) {
     });
   }
 }
-var log3, SESSION_FILE, ARCHIVE_DIR, ROTATE_THRESHOLD_BYTES, RETAIN_TAIL_BYTES, ARCHIVE_RETENTION_BYTES, ARCHIVE_NAME_RE, archiveSortKey, ARCHIVE_COUNT_RE, HISTORY_DEFAULT_LIMIT, HISTORY_MAX_LIMIT, archiveCountCache, archiveMsgCache, ARCHIVE_MSG_CACHE_MAX;
+var log3, SESSION_FILE, ARCHIVE_DIR, ARCHIVE_NAME_RE, archiveSortKey, ARCHIVE_COUNT_RE, archiveCountCache, archiveMsgCache, ARCHIVE_MSG_CACHE_MAX;
 var init_session = __esm({
   "src/session.ts"() {
     "use strict";
@@ -2762,18 +2830,13 @@ var init_session = __esm({
     init_logger();
     init_compaction();
     init_cleanMessages();
-    init_toolResultCap();
+    init_historyLimits();
     log3 = createLogger("session");
     SESSION_FILE = ".remy-session.json";
     ARCHIVE_DIR = ".logs/sessions";
-    ROTATE_THRESHOLD_BYTES = 32 * 1024 * 1024;
-    RETAIN_TAIL_BYTES = 16 * 1024 * 1024;
-    ARCHIVE_RETENTION_BYTES = 64 * 1024 * 1024;
     ARCHIVE_NAME_RE = /^(cleared|rotated)-.*\.json$/;
     archiveSortKey = (name) => name.replace(/^(cleared|rotated)-/, "");
     ARCHIVE_COUNT_RE = /\.c(\d+)\.json$/;
-    HISTORY_DEFAULT_LIMIT = 500;
-    HISTORY_MAX_LIMIT = 2e3;
     archiveCountCache = /* @__PURE__ */ new Map();
     archiveMsgCache = /* @__PURE__ */ new Map();
     ARCHIVE_MSG_CACHE_MAX = 3;
@@ -4919,7 +4982,7 @@ ${partial}` : "[INTERRUPTED] Agent was interrupted before producing output.",
             block.completedAt = Date.now();
             const innerMsgs = subAgentMessages.get(r.id);
             if (innerMsgs) {
-              block.subAgentMessages = innerMsgs;
+              block.subAgentMessages = capSubAgentTranscript(innerMsgs);
             }
             if (captureArtifacts?.includes(block.name) && !r.isError) {
               try {
@@ -5007,7 +5070,7 @@ var init_runner = __esm({
     init_logger();
     init_usageLedger();
     init_toolRegistry();
-    init_toolResultCap();
+    init_historyLimits();
     init_statusWatcher();
     init_cleanMessages();
     log8 = createLogger("sub-agent");
@@ -8438,6 +8501,8 @@ async function runTurn(params) {
   let lastCallInputTokens = 0;
   let lastCallCacheCreation = 0;
   let lastCallCacheRead = 0;
+  let abnormalStopRecoveries = 0;
+  const MAX_ABNORMAL_STOP_RECOVERIES = 2;
   const statusWatcher = isFirstMessage ? { stop() {
   }, pause() {
   }, resume() {
@@ -8781,6 +8846,18 @@ async function runTurn(params) {
         });
       }
       const toolCalls = getToolCalls(contentBlocks);
+      if (toolCalls.length === 0 && (stopReason === "repetition" || stopReason === "max_tokens") && abnormalStopRecoveries < MAX_ABNORMAL_STOP_RECOVERIES && !signal?.aborted) {
+        abnormalStopRecoveries++;
+        log13.warn("Abnormal stop \u2014 nudging model to continue", {
+          requestId,
+          stopReason,
+          attempt: abnormalStopRecoveries
+        });
+        const nudge = "Your previous response was cut off \u2014 it degenerated into repeated text or hit the output limit, and the repeated portion was removed. Reassess where you are in the task and continue from where you left off. Prefer a tool call over restating what you were about to do.";
+        state.messages.push({ role: "user", content: nudge, hidden: true });
+        onEvent({ type: "user_message", text: nudge, hidden: true });
+        continue;
+      }
       if (stopReason !== "tool_use" || toolCalls.length === 0) {
         statusWatcher.stop();
         saveSession(state);
@@ -8947,7 +9024,7 @@ async function runTurn(params) {
           block.completedAt = Date.now();
           const msgs = subAgentMessages.get(r.id);
           if (msgs) {
-            block.subAgentMessages = msgs;
+            block.subAgentMessages = capSubAgentTranscript(msgs);
           }
         }
       }
@@ -9006,7 +9083,7 @@ var init_agent = __esm({
     init_trigger2();
     init_surfaces();
     init_toolRegistry();
-    init_toolResultCap();
+    init_historyLimits();
     log13 = createLogger("agent");
     BRAND_TRIGGERING_TOOLS = /* @__PURE__ */ new Set([
       "writeSpec",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mindstudio-ai/remy",
-  "version": "0.1.273",
+  "version": "0.1.275",
   "description": "Remy coding agent",
   "repository": {
     "type": "git",