@wrongstack/core 0.264.0 → 0.265.1

package/dist/models/index.js

@@ -842,12 +842,9 @@ function getCachedEstimate(key, compute) {
   const existing = ESTIMATE_CACHE.get(key);
   if (existing !== void 0) return existing;
   if (ESTIMATE_CACHE.size >= ESTIMATE_CACHE_MAX_SIZE) {
-    let evicted = 0;
-    const maxEvict = Math.floor(ESTIMATE_CACHE_MAX_SIZE / 4);
     for (const k of ESTIMATE_CACHE.keys()) {
-      if (evicted >= maxEvict) break;
+      if (ESTIMATE_CACHE.size <= Math.floor(ESTIMATE_CACHE_MAX_SIZE / 2)) break;
       ESTIMATE_CACHE.delete(k);
-      evicted++;
     }
   }
   const estimate = compute(key);
@@ -919,9 +916,9 @@ Rules:
 - If unsure, keep rather than collapse (errors are more costly than waste)
 
 Return ONLY the JSON object, no markdown, no explanation outside the JSON.`;
-function formatMessages(messages, maxChars = 8e3) {
+function formatMessages(messages, maxTokens = 2048) {
   const lines = [];
-  let used = 0;
+  let usedTokens = 0;
   for (let i = 0; i < messages.length; i++) {
     const m = expectDefined(messages[i]);
     const role = m.role.padEnd(10, " ");
@@ -933,13 +930,14 @@ function formatMessages(messages, maxChars = 8e3) {
       text = content.filter(isTextBlock).map((b) => b.text).join(" ");
       const toolUses = content.filter((b) => b.type === "tool_use");
       if (toolUses.length > 0) {
-        text += ` [tools: ${toolUses.map((b) => b.name).join(", ")}]`;
+        text += ` [tools: ${toolUses.map((b) => b.name).filter(Boolean).join(", ")}]`;
       }
     }
     const line = `[${i}][${role}]: ${text}`;
-    if (used + line.length > maxChars) break;
+    const lineTokens = estimateTextTokens(line);
+    if (usedTokens + lineTokens > maxTokens) break;
     lines.push(line);
-    used += line.length;
+    usedTokens += lineTokens;
   }
   return lines.join("\n");
 }
@@ -948,20 +946,29 @@ var LLMSelector = class {
   model;
   maxContextTokens;
   systemPrompt;
+  maxOutputTokens;
   constructor(opts) {
     this.provider = opts.provider;
     this.model = opts.model ?? "unknown";
+    if (this.model === "unknown" && (process.env["NODE_ENV"] === "development" || process.env["WRONGSTACK_DEBUG"] === "1")) {
+      console.warn(
+        "[LLMSelector] model not set \u2014 selector will use the provider default. Set `model` explicitly in LLMSelectorOptions to silence this warning."
+      );
+    }
     this.maxContextTokens = opts.maxContextTokens ?? 4e4;
     this.systemPrompt = opts.systemPrompt ?? DEFAULT_SYSTEM_PROMPT;
+    this.maxOutputTokens = opts.maxOutputTokens ?? 1024;
   }
   async select(messages, maxToKeep) {
     const effectiveBudget = Math.min(maxToKeep, this.maxContextTokens);
-    const historyText = formatMessages(messages);
     const totalTokens = estimateMessageTokens(messages);
     const systemText = `${this.systemPrompt}
 
 Conversation (${messages.length} messages, ~${totalTokens} tokens, budget: ${effectiveBudget}):
 `;
+    const systemTokens = estimateTextTokens(systemText);
+    const historyBudget = Math.max(512, effectiveBudget - systemTokens - this.maxOutputTokens);
+    const historyText = formatMessages(messages, historyBudget);
     const budgetInstruction = totalTokens > effectiveBudget ? `
 
 IMPORTANT: Total conversation (${totalTokens} tokens) exceeds budget (${effectiveBudget}). You MUST collapse enough to fit. Prefer collapsing older/lower-importance ranges.` : "";
@@ -969,18 +976,26 @@ IMPORTANT: Total conversation (${totalTokens} tokens) exceeds budget (${effectiv
       model: this.model,
       system: [{ type: "text", text: systemText + budgetInstruction }],
       messages: [{ role: "user", content: historyText }],
-      maxTokens: 1024
+      maxTokens: this.maxOutputTokens
     };
     let raw;
+    const ac = new AbortController();
     try {
-      const ac = new AbortController();
-      const res = await this.provider.complete(req, { signal: ac.signal });
+      const timeoutSignal = AbortSignal.timeout(3e4);
+      const res = await this.provider.complete(req, {
+        signal: AbortSignal.any([ac.signal, timeoutSignal])
+      });
       const textBlocks = res.content.filter(isTextBlock);
       raw = textBlocks.map((b) => b.text).join("\n").trim();
-    } catch (_err) {
+    } catch (err) {
+      if (err instanceof Error) {
+        console.warn("[LLMSelector] selector call failed, using recency fallback:", err.message);
+      }
       return this.fallbackSelect(messages, effectiveBudget);
+    } finally {
+      ac.abort();
     }
-    return this.parseSelectorOutput(raw, messages.length);
+    return this.parseSelectorOutput(raw, messages);
   }
   fallbackSelect(messages, budget) {
     const toKeep = [];
@@ -1007,34 +1022,63 @@ IMPORTANT: Total conversation (${totalTokens} tokens) exceeds budget (${effectiv
       reasoning: `Fallback: kept last ${messages.length - startIdx} messages within ${budget} token budget`
     };
   }
-  parseSelectorOutput(raw, messageCount) {
+  /**
+   * Parse and validate the raw LLM output into a SelectorResult.
+   * Falls back to recency-based selection if the LLM output is malformed,
+   * out-of-bounds, or internally inconsistent.
+   */
+  parseSelectorOutput(raw, messages) {
+    const messageCount = messages.length;
+    if (messageCount === 0) {
+      return { kept: [], collapsed: [], reasoning: "empty session" };
+    }
     const jsonStart = raw.indexOf("{");
     const jsonEnd = raw.lastIndexOf("}");
     if (jsonStart === -1 || jsonEnd === -1) {
-      return this.fallbackSelect(
-        Array.from({ length: messageCount }, () => ({ role: "user", content: "" })),
-        this.maxContextTokens
-      );
+      return this.fallbackSelect(messages, this.maxContextTokens);
     }
     let parsed;
     try {
       parsed = JSON.parse(raw.slice(jsonStart, jsonEnd + 1));
     } catch {
-      return this.fallbackSelect(
-        Array.from({ length: messageCount }, () => ({ role: "user", content: "" })),
-        this.maxContextTokens
-      );
+      return this.fallbackSelect(messages, this.maxContextTokens);
     }
     const obj = parsed;
-    const kept = obj.kept ?? [];
-    const collapsed = obj.collapsed ?? [];
-    return {
-      kept: kept.map((k) => ({
+    const keptRaw = obj.kept ?? [];
+    const collapsedRaw = obj.collapsed ?? [];
+    const kept = [];
+    for (const k of keptRaw) {
+      if (typeof k.from !== "number" || typeof k.to !== "number" || k.from < 0 || k.to >= messageCount || k.from > k.to) {
+        return this.fallbackSelect(messages, this.maxContextTokens);
+      }
+      kept.push({
         from: k.from,
         to: k.to,
         importance: k.importance ?? "medium"
-      })),
-      collapsed: collapsed.map((c) => ({ from: c.from, to: c.to, summary: c.summary })),
+      });
+    }
+    const collapsed = [];
+    for (const c of collapsedRaw) {
+      if (typeof c.from !== "number" || typeof c.to !== "number" || c.from < 0 || c.to >= messageCount || c.from > c.to) {
+        return this.fallbackSelect(messages, this.maxContextTokens);
+      }
+      collapsed.push({ from: c.from, to: c.to, summary: c.summary });
+    }
+    const allRanges = [...kept, ...collapsed];
+    for (let i = 0; i < allRanges.length; i++) {
+      const a = allRanges[i];
+      if (!a) continue;
+      for (let j = i + 1; j < allRanges.length; j++) {
+        const b = allRanges[j];
+        if (!b) continue;
+        if (a.from <= b.to && a.to >= b.from) {
+          return this.fallbackSelect(messages, this.maxContextTokens);
+        }
+      }
+    }
+    return {
+      kept,
+      collapsed,
       reasoning: typeof obj.reasoning === "string" ? obj.reasoning : ""
     };
   }