zelari-code 1.27.0 → 1.27.2

package/dist/cli/main.bundled.js

@@ -35269,10 +35269,17 @@ var init_missionSlice = __esm({
 var planner_exports = {};
 __export(planner_exports, {
   KRAKEN_PLANNER_SYSTEM_PROMPT: () => KRAKEN_PLANNER_SYSTEM_PROMPT,
+  PlannerTransportError: () => PlannerTransportError,
   buildGraphFromPlan: () => buildGraphFromPlan,
   extractJsonObject: () => extractJsonObject,
   planTaskGraph: () => planTaskGraph
 });
+function resolvePlannerTimeoutMs(env = process.env) {
+  const raw = env.ZELARI_KRAKEN_PLANNER_TIMEOUT_MS;
+  if (raw === void 0 || raw === "") return DEFAULT_LLM_TIMEOUT_MS;
+  const n = Number.parseInt(raw, 10);
+  return Number.isFinite(n) && n >= 0 ? n : DEFAULT_LLM_TIMEOUT_MS;
+}
 function resolvePlannerMaxTokens(env = process.env) {
   const raw = env.ZELARI_KRAKEN_PLANNER_MAX_TOKENS;
   if (raw === void 0 || raw === "") return DEFAULT_LLM_MAX_TOKENS;
@@ -35321,6 +35328,23 @@ function extractBalancedJsonObject2(s) {
   }
   return null;
 }
+function escapeJsonControlChar(c) {
+  if (c >= " ") return c;
+  switch (c) {
+    case "\n":
+      return "\\n";
+    case "\r":
+      return "\\r";
+    case "	":
+      return "\\t";
+    case "\b":
+      return "\\b";
+    case "\f":
+      return "\\f";
+    default:
+      return `\\u${c.charCodeAt(0).toString(16).padStart(4, "0")}`;
+  }
+}
 function repairLooseJson(s) {
   let out = "";
   let i = 0;
@@ -35332,15 +35356,16 @@ function repairLooseJson(s) {
       i += 1;
       while (i < n) {
         const c = s[i];
-        out += c;
         i += 1;
         if (c === "\\") {
+          out += c;
           if (i < n) {
             out += s[i];
             i += 1;
           }
           continue;
         }
+        out += escapeJsonControlChar(c);
         if (c === '"') break;
       }
       continue;
@@ -35359,7 +35384,7 @@ function repairLooseJson(s) {
           i += 1;
           break;
         }
-        value += c;
+        value += escapeJsonControlChar(c);
         i += 1;
       }
       out += `"${value.replace(/"/g, '\\"')}"`;
@@ -35431,7 +35456,7 @@ async function resolveLlm(opts) {
   if (!baseUrl) {
     throw new Error(`No base URL for provider '${active}'. Set a custom endpoint in Settings.`);
   }
-  const model = opts.model?.trim() || getModelForProvider(active) || process.env.ZELARI_MODEL || "";
+  const model = opts.model?.trim() || process.env.ZELARI_KRAKEN_PLANNER_MODEL?.trim() || getModelForProvider(active) || process.env.ZELARI_MODEL || "";
   if (!model) {
     throw new Error(`No model selected for provider '${active}'`);
   }
@@ -35441,31 +35466,50 @@ async function createDefaultLlmClient(opts) {
   const llm = await resolveLlm(opts);
   return {
     async complete({ system, user }) {
+      const timeoutMs = resolvePlannerTimeoutMs();
       const controller = new AbortController();
-      const t = setTimeout(() => controller.abort(), LLM_TIMEOUT_MS);
+      let timedOut = false;
+      const t = timeoutMs > 0 ? setTimeout(() => {
+        timedOut = true;
+        controller.abort();
+      }, timeoutMs) : void 0;
       try {
         const url2 = `${llm.baseUrl.replace(/\/$/, "")}/chat/completions`;
-        const res = await fetch(url2, {
-          method: "POST",
-          signal: controller.signal,
-          headers: {
-            "content-type": "application/json",
-            authorization: `Bearer ${llm.apiKey}`
-          },
-          body: JSON.stringify({
-            model: llm.model,
-            temperature: 0.2,
-            max_tokens: resolvePlannerMaxTokens(),
-            stream: false,
-            messages: [
-              { role: "system", content: system },
-              { role: "user", content: user }
-            ]
-          })
-        });
+        let res;
+        try {
+          res = await fetch(url2, {
+            method: "POST",
+            signal: controller.signal,
+            headers: {
+              "content-type": "application/json",
+              authorization: `Bearer ${llm.apiKey}`
+            },
+            body: JSON.stringify({
+              model: llm.model,
+              temperature: 0.2,
+              max_tokens: resolvePlannerMaxTokens(),
+              stream: false,
+              messages: [
+                { role: "system", content: system },
+                { role: "user", content: user }
+              ]
+            })
+          });
+        } catch (err) {
+          if (timedOut) {
+            throw new PlannerTransportError(
+              `Planner request timed out after ${Math.round(timeoutMs / 1e3)}s (no response). Raise ZELARI_KRAKEN_PLANNER_TIMEOUT_MS, or set ZELARI_KRAKEN_PLANNER_MODEL to a faster non-reasoning model.`
+            );
+          }
+          throw new PlannerTransportError(
+            `Planner request failed: ${err instanceof Error ? err.message : String(err)}`
+          );
+        }
         if (!res.ok) {
           const errBody = await res.text().catch(() => "");
-          throw new Error(`LLM HTTP ${res.status}${errBody ? `: ${errBody.slice(0, 200)}` : ""}`);
+          throw new PlannerTransportError(
+            `LLM HTTP ${res.status}${errBody ? `: ${errBody.slice(0, 200)}` : ""}`
+          );
         }
         const json2 = await res.json();
         const choice = json2.choices?.[0];
@@ -35478,7 +35522,7 @@ async function createDefaultLlmClient(opts) {
           `Empty model response${reason ? ` (finish_reason=${reason})` : ""}` + (reason === "length" ? " \u2014 the model likely ran out of tokens before producing JSON; raise ZELARI_KRAKEN_PLANNER_MAX_TOKENS or simplify the prompt." : "")
         );
       } finally {
-        clearTimeout(t);
+        if (t !== void 0) clearTimeout(t);
       }
     }
   };
@@ -35568,6 +35612,7 @@ async function planTaskGraph(opts) {
       }
       return graph;
     } catch (err) {
+      if (err instanceof PlannerTransportError) throw err;
       lastError = err instanceof Error ? err.message : String(err);
       userMessage = `${userBase}
 
@@ -35578,7 +35623,7 @@ Your previous response was invalid (${lastError}). Return ONLY corrected JSON ma
     `kraken planner: failed to produce a valid task graph after ${MAX_PLAN_ATTEMPTS} attempts \u2014 ${lastError}`
   );
 }
-var LLM_TIMEOUT_MS, MAX_PLAN_ATTEMPTS, DEFAULT_LLM_MAX_TOKENS, DEFAULT_MAX_RETRIES, KRAKEN_PLANNER_SYSTEM_PROMPT, PlannedNodeSchema, PlannedGraphSchema;
+var MAX_PLAN_ATTEMPTS, DEFAULT_LLM_TIMEOUT_MS, PlannerTransportError, DEFAULT_LLM_MAX_TOKENS, DEFAULT_MAX_RETRIES, KRAKEN_PLANNER_SYSTEM_PROMPT, PlannedNodeSchema, PlannedGraphSchema;
 var init_planner = __esm({
   "src/cli/kraken/planner.ts"() {
     "use strict";
@@ -35587,8 +35632,14 @@ var init_planner = __esm({
     init_providerConfig();
     init_keyStore();
     init_openai_compatible();
-    LLM_TIMEOUT_MS = 9e4;
     MAX_PLAN_ATTEMPTS = 2;
+    DEFAULT_LLM_TIMEOUT_MS = 3e5;
+    PlannerTransportError = class extends Error {
+      constructor(message) {
+        super(message);
+        this.name = "PlannerTransportError";
+      }
+    };
     DEFAULT_LLM_MAX_TOKENS = 8192;
     DEFAULT_MAX_RETRIES = {
       explore: 0,
@@ -35645,6 +35696,7 @@ __export(executor_exports, {
   DEFAULT_FIX_BUDGET: () => DEFAULT_FIX_BUDGET,
   DEFAULT_MAX_PARALLEL: () => DEFAULT_MAX_PARALLEL,
   DEFAULT_NODE_TIMEOUT_MS: () => DEFAULT_NODE_TIMEOUT_MS,
+  DEFAULT_WRITER_NODE_TIMEOUT_MS: () => DEFAULT_WRITER_NODE_TIMEOUT_MS,
   KrakenGraphExecutor: () => KrakenGraphExecutor,
   isKrakenGraphEnabled: () => isKrakenGraphEnabled,
   isWorldModelGateEnabled: () => isWorldModelGateEnabled,
@@ -35666,11 +35718,21 @@ function resolveFixBudget(env = process.env) {
   const n = Number.parseInt(raw, 10);
   return Number.isFinite(n) && n >= 0 ? n : DEFAULT_FIX_BUDGET;
 }
-function resolveNodeTimeoutMs(env = process.env) {
+function resolveNodeTimeoutMs(env = process.env, agent) {
   const raw = env.ZELARI_KRAKEN_NODE_TIMEOUT_MS;
-  if (raw === void 0 || raw === "") return DEFAULT_NODE_TIMEOUT_MS;
-  const n = Number.parseInt(raw, 10);
-  return Number.isFinite(n) && n >= 0 ? n : DEFAULT_NODE_TIMEOUT_MS;
+  if (raw !== void 0 && raw !== "") {
+    const n = Number.parseInt(raw, 10);
+    if (Number.isFinite(n) && n >= 0) return n;
+  }
+  if (agent === "general") {
+    const rawWriter = env.ZELARI_KRAKEN_WRITER_NODE_TIMEOUT_MS;
+    if (rawWriter !== void 0 && rawWriter !== "") {
+      const n = Number.parseInt(rawWriter, 10);
+      if (Number.isFinite(n) && n >= 0) return n;
+    }
+    return DEFAULT_WRITER_NODE_TIMEOUT_MS;
+  }
+  return DEFAULT_NODE_TIMEOUT_MS;
 }
 function isKrakenGraphEnabled(env = process.env) {
   return env.ZELARI_KRAKEN_GRAPH !== "0";
@@ -35686,7 +35748,7 @@ function defaultChecksExists(cwd) {
     return false;
   }
 }
-var DEFAULT_MAX_PARALLEL, DEFAULT_FIX_BUDGET, DEFAULT_NODE_TIMEOUT_MS, KrakenGraphExecutor;
+var DEFAULT_MAX_PARALLEL, DEFAULT_FIX_BUDGET, DEFAULT_NODE_TIMEOUT_MS, DEFAULT_WRITER_NODE_TIMEOUT_MS, KrakenGraphExecutor;
 var init_executor = __esm({
   "src/cli/kraken/executor.ts"() {
     "use strict";
@@ -35699,11 +35761,13 @@ var init_executor = __esm({
     DEFAULT_MAX_PARALLEL = 12;
     DEFAULT_FIX_BUDGET = 3;
     DEFAULT_NODE_TIMEOUT_MS = 3e5;
+    DEFAULT_WRITER_NODE_TIMEOUT_MS = 9e5;
     KrakenGraphExecutor = class {
       deps;
       parentCwd;
       sessionId;
       maxParallel;
+      /** Explicit all-kinds override; when undefined the budget is per-kind. */
       nodeTimeoutMs;
       fixBudgetRemaining;
       worldModelGateOverride;
@@ -35717,7 +35781,7 @@ var init_executor = __esm({
         this.parentCwd = opts.parentCwd;
         this.sessionId = opts.sessionId;
         this.maxParallel = opts.maxParallel ?? resolveMaxParallel();
-        this.nodeTimeoutMs = opts.nodeTimeoutMs ?? resolveNodeTimeoutMs();
+        this.nodeTimeoutMs = opts.nodeTimeoutMs;
         this.fixBudgetRemaining = opts.fixBudget ?? resolveFixBudget();
         this.worldModelGateOverride = opts.worldModelGate;
         this.runTentacleFn = opts.runTentacleFn ?? runTentacle;
@@ -35818,7 +35882,8 @@ var init_executor = __esm({
         return res;
       }
       /**
-       * Bound a tentacle run to `nodeTimeoutMs` wall-clock. On timeout, resolves
+       * Bound a tentacle run to its wall-clock budget (explicit `nodeTimeoutMs`
+       * option, else per-kind — writers get more than readers). On timeout, resolves
        * to a synthetic `TentacleFailure` so the normal retry/fix/cascade-skip
        * path handles it — this only bounds how long the EXECUTOR waits, it does
        * not (cannot) forcibly cancel the underlying sub-agent run; the point is
@@ -35826,8 +35891,8 @@ var init_executor = __esm({
        * can exit instead of hanging forever on one stuck node.
        */
       withNodeTimeout(promise2, agent) {
-        if (this.nodeTimeoutMs <= 0) return promise2;
-        const ms = this.nodeTimeoutMs;
+        const ms = this.nodeTimeoutMs ?? resolveNodeTimeoutMs(process.env, agent);
+        if (ms <= 0) return promise2;
         let timer;
         const timeout = new Promise((resolve3) => {
           timer = setTimeout(() => {
@@ -48698,7 +48763,7 @@ init_keyStore();
 init_openai_compatible();
 var MAX_PAGE_CHARS = 24e3;
 var FETCH_TIMEOUT_MS2 = 25e3;
-var LLM_TIMEOUT_MS2 = 9e4;
+var LLM_TIMEOUT_MS = 9e4;
 var SYSTEM = `You convert web page content into a Zelari Code coding skill (SKILL.md style).
 
 Return ONLY a single JSON object (no markdown fences) with keys:
@@ -48841,7 +48906,7 @@ async function generateSkillFromUrl(opts) {
     model: opts.model
   });
   const controller = new AbortController();
-  const t = setTimeout(() => controller.abort(), LLM_TIMEOUT_MS2);
+  const t = setTimeout(() => controller.abort(), LLM_TIMEOUT_MS);
   try {
     const url2 = `${llm.baseUrl.replace(/\/$/, "")}/chat/completions`;
     const res = await fetch(url2, {