claudish 9.1.0 → 9.2.1

package/dist/index.js

@@ -715,7 +715,7 @@ var init_onepassword_config = __esm(() => {
 });
 
 // src/version.ts
-var VERSION = "9.1.0";
+var VERSION = "9.2.1";
 
 // src/logger.ts
 import { appendFile, existsSync as existsSync2, mkdirSync, readdirSync, unlinkSync, writeFileSync as writeFileSync2 } from "fs";
@@ -33599,6 +33599,26 @@ data: ${JSON.stringify(data)}
         }
         openToolBlocks.clear();
       };
+      const endTurnWithText = (text) => {
+        const idx = curIdx++;
+        send("content_block_start", {
+          type: "content_block_start",
+          index: idx,
+          content_block: { type: "text", text: "" }
+        });
+        send("content_block_delta", {
+          type: "content_block_delta",
+          index: idx,
+          delta: { type: "text_delta", text }
+        });
+        send("content_block_stop", { type: "content_block_stop", index: idx });
+        send("message_delta", {
+          type: "message_delta",
+          delta: { stop_reason: "end_turn", stop_sequence: null },
+          usage: { input_tokens: inputTokens, output_tokens: outputTokens }
+        });
+        send("message_stop", { type: "message_stop" });
+      };
       send("message_start", {
         type: "message_start",
         message: {
@@ -33809,24 +33829,7 @@ data: ${JSON.stringify(data)}
 
 [Context limit reached] This model's backend enforces a smaller context window${cap} than its API spec. ` + "Run /clear to start fresh (/compact will also fail \u2014 it re-sends the full conversation), " + `or route the model via \`oai@${opts.modelName}\` to use the full-size window.`;
                 }
-                const errorIdx = curIdx++;
-                send("content_block_start", {
-                  type: "content_block_start",
-                  index: errorIdx,
-                  content_block: { type: "text", text: "" }
-                });
-                send("content_block_delta", {
-                  type: "content_block_delta",
-                  index: errorIdx,
-                  delta: { type: "text_delta", text: errorText }
-                });
-                send("content_block_stop", { type: "content_block_stop", index: errorIdx });
-                send("message_delta", {
-                  type: "message_delta",
-                  delta: { stop_reason: "end_turn", stop_sequence: null },
-                  usage: { input_tokens: inputTokens, output_tokens: outputTokens }
-                });
-                send("message_stop", { type: "message_stop" });
+                endTurnWithText(errorText);
                 isClosed = true;
                 if (pingInterval) {
                   clearInterval(pingInterval);
@@ -33877,27 +33880,19 @@ data: ${JSON.stringify(data)}
           try {
             closeReasoning();
             closeText();
+            const toolCallCutOff = openToolBlocks.size > 0;
             closeTools();
-            const errorIdx = curIdx++;
-            send("content_block_start", {
-              type: "content_block_start",
-              index: errorIdx,
-              content_block: { type: "text", text: "" }
-            });
-            send("content_block_delta", {
-              type: "content_block_delta",
-              index: errorIdx,
-              delta: { type: "text_delta", text: `
+            if (toolCallCutOff) {
+              log("[ResponsesSSE] tool arguments cut off mid-stream \u2192 error event");
+              send("error", {
+                type: "error",
+                error: { type: "api_error", message: `Stream error: ${error}` }
+              });
+            } else {
+              endTurnWithText(`
 
-[Stream error: ${error}]` }
-            });
-            send("content_block_stop", { type: "content_block_stop", index: errorIdx });
-            send("message_delta", {
-              type: "message_delta",
-              delta: { stop_reason: "end_turn", stop_sequence: null },
-              usage: { input_tokens: inputTokens, output_tokens: outputTokens }
-            });
-            send("message_stop", { type: "message_stop" });
+[Stream error: ${error}]`);
+            }
           } catch {}
           isClosed = true;
           if (opts.onTokenUpdate)
@@ -40410,6 +40405,18 @@ function teamSlotActivity(teamSessionId) {
   }
   return out;
 }
+function teamSlotLiveBytes(teamSessionId) {
+  const run = liveTeamRuns.get(teamSessionId);
+  if (!run)
+    return null;
+  const out = {};
+  for (const slotId of run.processes.keys()) {
+    const bytes = run.liveBytesFor(slotId);
+    if (bytes !== null)
+      out[slotId] = bytes;
+  }
+  return out;
+}
 async function cancelTeamRun(teamSessionId, slotId) {
   const run = liveTeamRuns.get(teamSessionId);
   if (!run)
@@ -40862,6 +40869,7 @@ async function startModels(sessionPath, opts = {}) {
     processes,
     idleMsFor: (slotId) => runtimes.get(slotId)?.getIdleMs() ?? null,
     activityFor: (slotId) => runtimes.get(slotId)?.getActivity() ?? null,
+    liveBytesFor: (slotId) => runtimes.get(slotId)?.getByteCount() ?? null,
     cancelledSlots
   });
   const done = (async () => {
@@ -47107,6 +47115,28 @@ function buildChildClaudeFlags(agent, claudeFlags) {
   const flags = [...named ? ["--agent", named] : [], ...cleaned];
   return flags.length > 0 ? flags : undefined;
 }
+function teamStatusNote(opts) {
+  if (!opts.anyRunning)
+    return;
+  const trap = "outputSize is the size of the FINAL answer and is written once, when a slot " + "finishes. Every RUNNING slot therefore reports 0 however much work it has done: " + "it is NOT a progress signal, and 0 there does not mean the slot produced nothing.";
+  if (!opts.live) {
+    return `${trap} This run is not live in this process \u2014 it was started elsewhere, or this ` + "server restarted \u2014 so there are no liveness fields for it and the states above are " + "whatever was last written to status.json.";
+  }
+  return `${trap} Judge a running slot on three fields read TOGETHER: ` + "live_output_bytes_by_slot (answer bytes so far, the number outputSize will become), " + "idle_seconds_by_slot (seconds of silence) and activity_by_slot (what it is doing). " + "Silence in tool_executing is a build or test suite running, and is not a failure " + 'signal. Nothing cancels on your behalf \u2014 use mode:"cancel" if you decide to.';
+}
+function buildTeamStatusPayload(opts) {
+  const { status, sessionPath, idle, activity, liveBytes } = opts;
+  const anyRunning = Object.values(status.models).some((m) => m.state === "RUNNING");
+  const note = teamStatusNote({ anyRunning, live: idle !== null });
+  return {
+    ...status,
+    live_output_bytes_by_slot: liveBytes,
+    idle_seconds_by_slot: idle,
+    activity_by_slot: activity,
+    ...note ? { note } : {},
+    ...anyRunning ? {} : { summary: formatTeamResult(status, sessionPath) }
+  };
+}
 function formatTeamResult(status, sessionPath) {
   const entries = Object.entries(status.models).sort(([a], [b]) => a.localeCompare(b));
   const failed = entries.filter(([, m]) => m.state === "FAILED" || m.state === "TIMEOUT" || m.state === "EMPTY");
@@ -47592,14 +47622,14 @@ Use with: run_prompt(model="${suggested}", prompt="your prompt")`;
   });
   tools.push({
     name: "team",
-    description: "Run AI models on a task with anonymized outputs and optional blind judging. " + "Modes: 'run' (START the models and return a slot map immediately \u2014 it does NOT " + "wait), 'status' (per-slot state, plus how long each slot has been silent), " + "'cancel' (stop one slot or the whole run), 'judge' (blind-vote on existing " + "outputs), 'run-and-judge' (the blocking pipeline). " + "NO SLOT IS EVER KILLED ON A TIMER. A team slot is a full Claude Code session and " + "may work for a long time; a slot inside a build or test suite emits nothing for " + "minutes and is working, not stuck. Poll 'status', judge the silence against the " + "task you set, and use 'cancel' if you decide a slot is wedged.",
+    description: "Run AI models on a task with anonymized outputs and optional blind judging. " + "Modes: 'run' (START the models and return a slot map immediately \u2014 it does NOT " + "wait), 'status' (per-slot state, plus live answer bytes, seconds of silence and " + "current activity per slot), " + "'cancel' (stop one slot or the whole run), 'judge' (blind-vote on existing " + "outputs), 'run-and-judge' (the blocking pipeline). " + "NO SLOT IS EVER KILLED ON A TIMER. A team slot is a full Claude Code session and " + "may work for a long time; a slot inside a build or test suite emits nothing for " + "minutes and is working, not stuck. Poll 'status', judge the silence against the " + "task you set, and use 'cancel' if you decide a slot is wedged. " + "JUDGING A RUNNING SLOT: read live_output_bytes_by_slot, idle_seconds_by_slot and " + "activity_by_slot. Do NOT use 'outputSize' \u2014 it is the size of the final answer, " + "written only when a slot finishes, so it is 0 for every running slot no matter how " + "much that slot has produced.",
     inputSchema: {
       type: "object",
       properties: {
         mode: {
           type: "string",
           enum: ["run", "judge", "run-and-judge", "status", "cancel"],
-          description: "Operation mode. 'run' STARTS the models and returns immediately with a " + "slot map \u2014 it does not wait. Poll 'status' for progress, then 'judge' once " + "the slots have finished. 'run-and-judge' is the blocking pipeline and holds " + "the call open for the whole run. 'cancel' stops one slot or the whole run."
+          description: "Operation mode. 'run' STARTS the models and returns immediately with a " + "slot map \u2014 it does not wait. Poll 'status' for progress (read it from " + "live_output_bytes_by_slot and activity_by_slot, never from outputSize, " + "which stays 0 until a slot finishes), then 'judge' once the slots have " + "finished. 'run-and-judge' is the blocking pipeline and holds the call open " + "for the whole run. 'cancel' stops one slot or the whole run."
         },
         slot: {
           type: "string",
@@ -47750,21 +47780,17 @@ Use with: run_prompt(model="${suggested}", prompt="your prompt")`;
           case "status": {
             const status = getStatus(resolved);
             const teamSessionId = resolved.split("/").filter(Boolean).pop() ?? "team";
-            const idle = teamSlotIdleSeconds(teamSessionId);
-            const settled = !Object.values(status.models).some((m) => m.state === "RUNNING");
             return {
               content: [
                 {
                   type: "text",
-                  text: JSON.stringify({
-                    ...status,
-                    idle_seconds_by_slot: idle,
-                    activity_by_slot: teamSlotActivity(teamSessionId),
-                    ...idle ? {
-                      note: "idle_seconds_by_slot is how long each slot has been silent; " + "read it against activity_by_slot. Silence in tool_executing " + "is a build or test suite running, and is not a failure " + 'signal. Nothing cancels on your behalf \u2014 use mode:"cancel" ' + "if you decide to."
-                    } : {},
-                    ...settled ? { summary: formatTeamResult(status, resolved) } : {}
-                  }, null, 2)
+                  text: JSON.stringify(buildTeamStatusPayload({
+                    status,
+                    sessionPath: resolved,
+                    idle: teamSlotIdleSeconds(teamSessionId),
+                    activity: teamSlotActivity(teamSessionId),
+                    liveBytes: teamSlotLiveBytes(teamSessionId)
+                  }), null, 2)
                 }
               ]
             };
@@ -69386,7 +69412,8 @@ Team Status`);
     }
     const rawArgs = process.argv.slice(2);
     const explicitNoAutoApprove = rawArgs.includes("--no-auto-approve");
-    if (cliConfig.autoApprove && !explicitNoAutoApprove && !cliConfig.stdin && cliConfig.interactive) {
+    const explicitAutoApprove = rawArgs.includes("-y") || rawArgs.includes("--auto-approve");
+    if (cliConfig.autoApprove && !explicitNoAutoApprove && !explicitAutoApprove && !cliConfig.stdin && cliConfig.interactive) {
       await Promise.resolve().then(() => init_profile_config());
       try {
         const cfg = loadConfig();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claudish",
-  "version": "9.1.0",
+  "version": "9.2.1",
   "description": "Run Claude Code with any model - OpenRouter, Ollama, LM Studio & local models",
   "type": "module",
   "main": "./dist/index.js",
@@ -60,10 +60,10 @@
     "ai"
   ],
   "optionalDependencies": {
-    "@claudish/magmux-darwin-arm64": "9.1.0",
-    "@claudish/magmux-darwin-x64": "9.1.0",
-    "@claudish/magmux-linux-arm64": "9.1.0",
-    "@claudish/magmux-linux-x64": "9.1.0"
+    "@claudish/magmux-darwin-arm64": "9.2.1",
+    "@claudish/magmux-darwin-x64": "9.2.1",
+    "@claudish/magmux-linux-arm64": "9.2.1",
+    "@claudish/magmux-linux-x64": "9.2.1"
   },
   "author": "Jack Rudenko <i@madappgang.com>",
   "license": "MIT",