billion-context 0.1.100-pr.685.510 → 0.1.100

package/dist/index.js

@@ -47777,14 +47777,7 @@ async function fetchWithTimeout(url, opts, timeoutMs, externalSignal) {
     const finalOpts = {
       ...opts,
       signal: controller.signal,
-      dispatcher: opts.dispatcher ?? directDispatcher(effective),
-      // Forward-proxy correctness: never silently follow a redirect.
-      // undici's default (follow) downgrades POST→GET and drops the body
-      // on 301/302/303, so a redirecting upstream (CDN/WAF) turns a valid
-      // POST into a 405 at the redirect target. Pass the 3xx through to
-      // the client, which follows it with its own policy. Internal bili
-      // fetches that want to follow (registry, upstream test) opt in.
-      redirect: opts.redirect ?? "manual"
+      dispatcher: opts.dispatcher ?? directDispatcher(effective)
     };
     const raw = await fetch(url, finalOpts);
     if (raw.body) {
@@ -48587,7 +48580,6 @@ function parseRouteEntry(v2) {
     if (obj.compress) route.compress = obj.compress;
     const compatRoles = parseCompatRoles(obj.compat?.roles);
     if (compatRoles) route.compat = { roles: compatRoles };
-    if (typeof obj.passthrough === "boolean") route.passthrough = obj.passthrough;
     return route;
   }
   if (v2 === null) return {};
@@ -49017,8 +49009,7 @@ async function fetchFresh() {
     try {
       const { response, clearTimer } = await fetchWithTimeout(REGISTRY_URL, {
         ...opts,
-        headers: { Accept: "application/json" },
-        redirect: "follow"
+        headers: { Accept: "application/json" }
       }, 15e3);
       const text = await response.text();
       clearTimer();
@@ -51511,44 +51502,20 @@ ${body}`;
 }
 
 // src/loop/tag-echo-filter.ts
-function buildAcplikeName() {
-  const cores = ["acp", "apc", "cap", "cpa", "pac", "pca"];
-  const names = new Set(cores);
-  for (const c of cores) {
-    for (const ch of ["a", "c", "p", "i"]) {
-      for (let pos = 0; pos <= c.length; pos++) names.add(c.slice(0, pos) + ch + c.slice(pos));
-    }
-  }
-  return [...names].sort((a, b2) => b2.length - a.length).join("|");
-}
-var ACP_NAME_ALT = `(?:${buildAcplikeName()})`;
-var NAME = ACP_NAME_ALT;
-var PAIRED = new RegExp("<" + NAME + "\\s[^<>]{0,256}>([^<>]{0,64})<\\/" + NAME + ">");
-var LONE_OPEN = new RegExp("<" + NAME + "(?:\\s[^<>]{0,256})?>");
-var LONE_CLOSE = new RegExp("<\\/" + NAME + "(?=[\\s>])[^<>]{0,32}>");
-var PARTIAL_TAIL = new RegExp("(<" + NAME + "\\s[^<>]*|<\\/" + NAME + "(?:\\s[^<>]{0,32})?|<\\/?[acip]*)$");
-var TRUNC_OPEN = new RegExp("<" + NAME + "\\s[^<>]*$");
-var TRUNC_CLOSE = new RegExp("<\\/" + NAME + "(?:\\s[^<>]{0,32})?$");
-var DEFINITE_TAIL = new RegExp("^<" + NAME + "\\s|^<\\/" + NAME);
-var OPEN_WITH_ATTRS = new RegExp("^<" + NAME + "\\s");
-var CLOSE_HEAD = "</";
-var CLOSE_NAME_ANCHORED = new RegExp("^" + NAME);
+var PAIRED = /\x3cacp\s[^<>]{0,256}>([^<>]{0,64})<\/acp>/;
+var LONE_OPEN = /\x3cacp(?:\s[^<>]{0,256})?>/;
+var LONE_CLOSE = /<\/acp(?=[\s>])[^<>]{0,32}>/;
+var PARTIAL_TAIL = /(\x3cacp\s[^<>]*|\x3c\/acp(?:\s[^<>]{0,32})?|<\/?a?c?p?)$/;
+var TRUNC_OPEN = /\x3cacp\s[^<>]*$/;
+var TRUNC_CLOSE = /\x3c\/acp(?:\s[^<>]{0,32})?$/;
+var CLOSE_TAG = "</acp";
 var HOLD_LIMIT = 128;
 var TAG_OPEN_CAP = 4096;
 var SWALLOW_CAP = 80;
-function looseCloseEnd(s3) {
-  let idx = s3.indexOf(CLOSE_HEAD);
-  while (idx >= 0) {
-    const m2 = CLOSE_NAME_ANCHORED.exec(s3.slice(idx + 2));
-    if (m2 && s3[idx + 2 + m2[0].length] === ">") return idx + 2 + m2[0].length + 1;
-    idx = s3.indexOf(CLOSE_HEAD, idx + 1);
-  }
-  return -1;
-}
 function stripAcpTags(text) {
   return text.replace(new RegExp(PAIRED.source, "g"), "").replace(new RegExp(LONE_OPEN.source, "g"), "").replace(new RegExp(LONE_CLOSE.source, "g"), "").replace(new RegExp(TRUNC_OPEN.source), "").replace(new RegExp(TRUNC_CLOSE.source), "");
 }
-var RENDER_TAG_DETECT = new RegExp("<\\/?" + NAME + "(?=[\\s>])|\\\\u003c\\/?" + NAME + "(?=[\\s>\\\\])");
+var RENDER_TAG_DETECT = /\x3c\/?acp(?=[\s>])|\\u003c\/?acp(?=[\s>\\])/;
 function containsRenderTagText(s3) {
   return RENDER_TAG_DETECT.test(s3);
 }
@@ -51565,8 +51532,6 @@ function createTagEchoFilter(onDrop) {
   let swallowed = "";
   let droppedAny = false;
   let notified = false;
-  let inputChars = 0;
-  let outputChars = 0;
   const drop = (snippet) => {
     droppedAny = true;
     if (onDrop && !notified) {
@@ -51580,9 +51545,10 @@ function createTagEchoFilter(onDrop) {
     for (; ; ) {
       if (swallowUntilClose) {
         const combined = swallowed + buf;
-        const end = looseCloseEnd(combined);
-        if (end >= 0) {
-          drop(combined.slice(0, end));
+        const closeIdx = combined.indexOf(CLOSE_TAG);
+        if (closeIdx >= 0 && combined[closeIdx + CLOSE_TAG.length] === ">") {
+          const end = closeIdx + CLOSE_TAG.length + 1;
+          drop(swallowed + combined.slice(0, end));
           swallowed = "";
           swallowUntilClose = false;
           buf = combined.slice(end);
@@ -51607,7 +51573,7 @@ function createTagEchoFilter(onDrop) {
       if (!m2) {
         const t = PARTIAL_TAIL.exec(buf);
         if (t) {
-          const definite = DEFINITE_TAIL.test(t[0]);
+          const definite = /^\x3cacp\s/.test(t[0]) || /^\x3c\/acp/.test(t[0]);
           const cap = definite ? TAG_OPEN_CAP : HOLD_LIMIT;
           if (t[0].length <= cap) {
             held = t[0];
@@ -51626,7 +51592,7 @@ function createTagEchoFilter(onDrop) {
       drop(m2[0]);
       out += buf.slice(0, m2.index);
       buf = buf.slice(m2.index + m2[0].length);
-      if (m2 === o && OPEN_WITH_ATTRS.test(m2[0])) {
+      if (m2 === o && /^\x3cacp\s/.test(m2[0])) {
         swallowUntilClose = true;
         swallowed = "";
       }
@@ -51635,12 +51601,9 @@ function createTagEchoFilter(onDrop) {
   };
   return {
     push(delta) {
-      inputChars += delta.length;
       const chunk = held + delta;
       held = "";
-      const r = process2(chunk);
-      outputChars += r.length;
-      return r;
+      return process2(chunk);
     },
     flush() {
       const rest = swallowed + held;
@@ -51648,36 +51611,27 @@ function createTagEchoFilter(onDrop) {
       swallowed = "";
       held = "";
       swallowUntilClose = false;
-      let result;
       if (wasSwallowing) {
         if (rest.length > 0) drop(rest);
-        result = "";
-      } else {
-        const t = new RegExp(TRUNC_OPEN.source).exec(rest);
-        if (t) {
-          drop(t[0]);
-          result = rest.slice(0, t.index);
-        } else {
-          const tc = new RegExp(TRUNC_CLOSE.source).exec(rest);
-          if (tc) {
-            drop(tc[0]);
-            result = rest.slice(0, tc.index);
-          } else {
-            result = rest;
-          }
-        }
+        return "";
       }
-      outputChars += result.length;
-      return result;
+      const t = new RegExp(TRUNC_OPEN.source).exec(rest);
+      if (t) {
+        drop(t[0]);
+        return rest.slice(0, t.index);
+      }
+      const tc = new RegExp(TRUNC_CLOSE.source).exec(rest);
+      if (tc) {
+        drop(tc[0]);
+        return rest.slice(0, tc.index);
+      }
+      return rest;
     },
     dropped() {
       return droppedAny;
     },
     pending() {
       return held.length > 0 || swallowUntilClose;
-    },
-    stats() {
-      return { inputChars, outputChars, dropped: droppedAny };
     }
   };
 }
@@ -51982,41 +51936,16 @@ function splitChunks(messages, startIdx, endIdx, budget, minUnits, countText = d
   }
   return chunks;
 }
-function summaryPayload(protocol, model, system, content, stream2, includeMaxOutputTokens) {
+function summaryPayload(protocol, model, system, content, stream2) {
   if (protocol === "anthropic") {
     return { model, max_tokens: MAX_SUMMARY_OUTPUT_TOKENS, system, messages: [{ role: "user", content }], stream: stream2 };
   }
   if (protocol === "openai") {
     return { model, max_tokens: MAX_SUMMARY_OUTPUT_TOKENS, messages: [{ role: "system", content: system }, { role: "user", content }], stream: stream2 };
   }
-  const payload = { model, instructions: system, input: [{ role: "user", content }], stream: stream2, store: false };
-  if (includeMaxOutputTokens) payload.max_output_tokens = MAX_SUMMARY_OUTPUT_TOKENS;
-  return payload;
+  return { model, max_output_tokens: MAX_SUMMARY_OUTPUT_TOKENS, instructions: system, input: [{ role: "user", content }], stream: stream2, store: false };
 }
 var STREAM_REQUIRED_RE = /\bstream\b[^\n]{0,60}\btrue\b/i;
-var MAX_OUTPUT_TOKENS_REJECTED_RE = /\bmax_output_tokens\b/i;
-function noMaxOutputTokensKey(deps) {
-  return `${deps.url}\0${deps.model}`;
-}
-function hasLearnedNoMaxOutputTokens(deps) {
-  const learned = deps.session.metadata.preflightNoMaxOutputTokens;
-  return typeof learned === "object" && learned !== null && learned[noMaxOutputTokensKey(deps)] === true;
-}
-function rememberNoMaxOutputTokens(deps) {
-  const learned = deps.session.metadata.preflightNoMaxOutputTokens;
-  const map = typeof learned === "object" && learned !== null ? learned : {};
-  map[noMaxOutputTokensKey(deps)] = true;
-  deps.session.metadata.preflightNoMaxOutputTokens = map;
-}
-var SUMMARY_STRIP_HEADERS = /* @__PURE__ */ new Set(["x-openai-internal-codex-responses-lite"]);
-function summaryHeaders(deps) {
-  const headers = {};
-  for (const [k2, v2] of Object.entries(deps.headers)) {
-    if (SUMMARY_STRIP_HEADERS.has(k2.toLowerCase())) continue;
-    headers[k2] = v2;
-  }
-  return headers;
-}
 function extractSummaryFromSse(protocol, text) {
   let out = "";
   for (const line of text.split("\n")) {
@@ -52080,39 +52009,25 @@ async function summarizeRange(deps, content, startRef, endRef) {
   const system = buildCompressSystemPrompt(deps.prompts) + `
 
 TASK: The conversation segment below (messages ${startRef}\u2013${endRef}) must be compressed because the session context exceeds the current model's window. Write a tier-1 compression summary of the segment following every rule above. Output ONLY the summary text \u2014 no preamble, no closing remarks, no tool calls.`;
-  let stream2 = deps.session.metadata.preflightStreamSummary === true;
-  let includeMaxOutputTokens = !(deps.protocol === "responses" && hasLearnedNoMaxOutputTokens(deps));
-  for (; ; ) {
-    try {
-      return await requestSummary(deps, system, content, stream2, includeMaxOutputTokens);
-    } catch (err2) {
-      if (err2 instanceof UpstreamHttpError && err2.status === 400) {
-        let adapted = false;
-        if (!stream2 && STREAM_REQUIRED_RE.test(err2.body)) {
-          deps.session.metadata.preflightStreamSummary = true;
-          stream2 = true;
-          adapted = true;
-          deps.log("info", "[preflight] upstream requires stream for summaries; retrying with SSE (learned for this session)");
-        }
-        if (deps.protocol === "responses" && includeMaxOutputTokens && MAX_OUTPUT_TOKENS_REJECTED_RE.test(err2.body)) {
-          rememberNoMaxOutputTokens(deps);
-          includeMaxOutputTokens = false;
-          adapted = true;
-          deps.log("info", `[preflight] upstream rejects max_output_tokens for summaries (model=${deps.model}); retrying without it (learned for this session+upstream+model)`);
-        }
-        if (adapted) continue;
-      }
-      throw err2;
+  const learned = deps.session.metadata.preflightStreamSummary === true;
+  try {
+    return await requestSummary(deps, system, content, learned);
+  } catch (err2) {
+    if (err2 instanceof UpstreamHttpError && err2.status === 400 && !learned && STREAM_REQUIRED_RE.test(err2.body)) {
+      deps.session.metadata.preflightStreamSummary = true;
+      deps.log("info", "[preflight] upstream requires stream for summaries; retrying with SSE (learned for this session)");
+      return await requestSummary(deps, system, content, true);
     }
+    throw err2;
   }
 }
-async function requestSummary(deps, system, content, stream2, includeMaxOutputTokens) {
+async function requestSummary(deps, system, content, stream2) {
   const { response, clearTimer } = await fetchWithRetry(
     deps.url,
     {
       method: "POST",
-      headers: { "content-type": "application/json", ...summaryHeaders(deps) },
-      body: JSON.stringify(summaryPayload(deps.protocol, deps.model, system, content, stream2, includeMaxOutputTokens)),
+      headers: { "content-type": "application/json", ...deps.headers },
+      body: JSON.stringify(summaryPayload(deps.protocol, deps.model, system, content, stream2)),
       dispatcher: proxyDispatcher(deps.proxyUrl)
     },
     void 0,
@@ -53749,18 +53664,6 @@ async function* runCompressLoop(upstream, ctx, requestBody, requestOptions, adap
   }
 }
 
-// src/degenerate-turn.ts
-function degenerateTurnWarning(o) {
-  if ((o.reason ?? o.terminalReason) !== o.terminalReason) return null;
-  if (o.toolCalls > 0) return null;
-  if (o.text.outputChars > 0) return null;
-  const bits = [];
-  if (o.sawThinking) bits.push("thinking present");
-  if (o.text.dropped && o.text.inputChars > 0) bits.push(`${o.text.inputChars} chars of emitted text stripped as render-tag echo`);
-  else bits.push("no visible text emitted");
-  return `[degenerate-turn] ${o.wire}: turn ended ${o.terminalReason} with zero visible text and zero tool calls (${bits.join("; ")}) \u2014 the agent receives an empty turn and may stall until nudged (#673)`;
-}
-
 // src/loop/adapter-responses.ts
 var RESPONSES_ITEM_ID_MAX = 64;
 function normalizeResponsesMessageItems(input) {
@@ -53824,7 +53727,7 @@ function dropWhitespaceResponsesMessages(input) {
   }
   return dropped;
 }
-var RENDER_TAG_RE = new RegExp("<" + ACP_NAME_ALT + "\\s[^>]*>[^<]*<\\/" + ACP_NAME_ALT + ">|<" + ACP_NAME_ALT + "\\s[^>]*\\/>", "g");
+var RENDER_TAG_RE = /\x3cacp\s[^>]*\x3e[^<]*\x3c\/acp\x3e|\x3cacp\s[^>]*\/\x3e/g;
 function stripRenderTags(text) {
   return text.replace(RENDER_TAG_RE, "");
 }
@@ -54039,17 +53942,6 @@ data: ${JSON.stringify({ type: "response.created", response: { id: createdRespId
           yield { kind: "text", delta: tail, ...raw ? { raw } : {} };
         }
       };
-      let sawReasoning = false;
-      let toolCallsEmitted = 0;
-      let degenerateWarned = false;
-      const maybeWarnDegenerate = (reason) => {
-        if (degenerateWarned) return;
-        const msg = degenerateTurnWarning({ reason, terminalReason: "completed", toolCalls: toolCallsEmitted, text: tagFilter.stats(), sawThinking: sawReasoning, wire: "responses" });
-        if (msg) {
-          degenerateWarned = true;
-          log("warn", msg);
-        }
-      };
       for await (const eventStr of iterSseEvents(upstream)) {
         const explicitType = extractEventType(eventStr);
         const dataLine = extractDataLine(eventStr);
@@ -54062,7 +53954,6 @@ data: ${JSON.stringify({ type: "response.created", response: { id: createdRespId
         }
         const type = explicitType ?? (typeof obj.type === "string" ? obj.type : null);
         if (!type) continue;
-        if (type.startsWith("response.reasoning")) sawReasoning = true;
         const rawBuf = Buffer.from(eventStr + "\n\n", "utf8");
         if (round === 1 && typeof obj.output_index === "number") {
           outputIndex = Math.max(outputIndex, obj.output_index + 1);
@@ -54146,7 +54037,6 @@ data: ${JSON.stringify({ type: "response.created", response: { id: createdRespId
             if (fc) {
               if (typeof item.arguments === "string" && item.arguments) fc.arguments = item.arguments;
               pending.delete(itemId);
-              toolCallsEmitted++;
               yield {
                 kind: "tool_call",
                 name: fc.name,
@@ -54155,7 +54045,6 @@ data: ${JSON.stringify({ type: "response.created", response: { id: createdRespId
               };
             } else {
               yield { kind: "meta", chunk: rawBuf, firstRoundOnly: false };
-              toolCallsEmitted++;
               yield {
                 kind: "tool_call",
                 name: typeof item.name === "string" ? item.name : "",
@@ -54192,7 +54081,6 @@ data: ${JSON.stringify({ type: "response.created", response: { id: createdRespId
             outputTokens: typeof respUsage?.output_tokens === "number" ? respUsage.output_tokens : void 0,
             cachedTokens: typeof pd?.cached_tokens === "number" ? pd.cached_tokens : void 0
           };
-          maybeWarnDegenerate("completed");
           yield { kind: "done", finishReason: "completed" };
         } else if (type === "response.incomplete") {
           yield* flushFilter();
@@ -54510,23 +54398,11 @@ function createOpenaiAdapter(requestBody, clientSystem, hostCredit = 0, absorbNa
           yield { kind: "text", delta: tail, raw: buildContent(tail) };
         }
       };
-      let sawReasoning = false;
-      let toolCallsEmitted = 0;
-      let degenerateWarned = false;
-      const maybeWarnDegenerate = (reason) => {
-        if (degenerateWarned) return;
-        const msg = degenerateTurnWarning({ reason, terminalReason: "stop", toolCalls: toolCallsEmitted, text: tagFilter.stats(), sawThinking: sawReasoning, wire: "openai" });
-        if (msg) {
-          degenerateWarned = true;
-          log("warn", msg);
-        }
-      };
       const rawToolChunks = [];
       let sawRealToolCall = false;
       const flushPendingAsStructured = function* () {
         for (const [, tc] of pending) {
           if (tc.name.length > 0 || tc.id.length > 0) {
-            toolCallsEmitted++;
             yield {
               kind: "tool_call",
               name: tc.name,
@@ -54549,7 +54425,6 @@ function createOpenaiAdapter(requestBody, clientSystem, hostCredit = 0, absorbNa
         }
         for (const [idx, tc] of pending) {
           if (realIndexes.has(idx) && (tc.name.length > 0 || tc.id.length > 0)) {
-            toolCallsEmitted++;
             yield {
               kind: "tool_call",
               name: tc.name,
@@ -54569,7 +54444,6 @@ function createOpenaiAdapter(requestBody, clientSystem, hostCredit = 0, absorbNa
         }
         for (const [idx, tc] of pending) {
           if (!realIndexes.has(idx) && tc.name.length > 0) {
-            toolCallsEmitted++;
             yield {
               kind: "tool_call",
               name: tc.name,
@@ -54594,7 +54468,6 @@ function createOpenaiAdapter(requestBody, clientSystem, hostCredit = 0, absorbNa
           if (sawRealToolCall) {
             yield { kind: "meta", chunk: Buffer.from(eventStr + "\n\n", "utf8") };
           }
-          maybeWarnDegenerate("stop");
           yield { kind: "done", finishReason: "stop", ...sawRealToolCall ? { suppressCompletion: true } : {} };
           continue;
         }
@@ -54640,11 +54513,9 @@ function createOpenaiAdapter(requestBody, clientSystem, hostCredit = 0, absorbNa
           if (sawRealToolCall) {
             const chunk = patchUsageChunk(eventStr, parsed, u2 ?? {}, hostCredit);
             yield { kind: "meta", chunk };
-            maybeWarnDegenerate(finishReason);
             yield { kind: "done", finishReason, suppressCompletion: true };
             continue;
           } else {
-            maybeWarnDegenerate(finishReason);
             yield {
               kind: "done",
               finishReason: hadToolCalls && finishReason === "stop" ? "tool_calls" : finishReason
@@ -54653,7 +54524,6 @@ function createOpenaiAdapter(requestBody, clientSystem, hostCredit = 0, absorbNa
         }
         if (!delta) continue;
         if (typeof delta.reasoning_content === "string" && delta.reasoning_content.length > 0) {
-          sawReasoning = true;
           yield { kind: "reasoning", delta: delta.reasoning_content, raw: finishReason ? stripFinishReasonChunk(rawBuf) : rawBuf };
         }
         if (delta.tool_calls) {
@@ -54931,17 +54801,6 @@ ${systemPrompt}` : systemPrompt;
         log("warn", `[tag-echo] stripped model-emitted render tag: ${snippet.slice(0, 80).replace(/\n/g, " ")}`);
       });
       let lastTextIndex = null;
-      let sawThinking = false;
-      let toolCallsEmitted = 0;
-      let degenerateWarned = false;
-      const maybeWarnDegenerate = (reason) => {
-        if (degenerateWarned) return;
-        const msg = degenerateTurnWarning({ reason, terminalReason: "end_turn", toolCalls: toolCallsEmitted, text: tagFilter.stats(), sawThinking, wire: "anthropic" });
-        if (msg) {
-          degenerateWarned = true;
-          log("warn", msg);
-        }
-      };
       for await (const eventStr of iterSseEvents2(upstream)) {
         const parsed = parseAnthropicSse(eventStr);
         if (!parsed) continue;
@@ -54978,10 +54837,7 @@ ${systemPrompt}` : systemPrompt;
             const id = typeof block.id === "string" ? block.id : `toolu_${upstreamIndex}`;
             pending.set(upstreamIndex, { id, name, json: "" });
           } else {
-            if (block.type === "thinking" || block.type === "redacted_thinking") {
-              thinkingIndexes.add(upstreamIndex);
-              sawThinking = true;
-            }
+            if (block.type === "thinking" || block.type === "redacted_thinking") thinkingIndexes.add(upstreamIndex);
             const ci2 = clientIndex++;
             indexMap.set(upstreamIndex, ci2);
             openBlocks.push(ci2);
@@ -55026,7 +54882,6 @@ ${systemPrompt}` : systemPrompt;
           const tb = pending.get(upstreamIndex);
           if (tb) {
             pending.delete(upstreamIndex);
-            toolCallsEmitted++;
             yield {
               kind: "tool_call",
               name: tb.name,
@@ -55073,7 +54928,6 @@ ${systemPrompt}` : systemPrompt;
               cachedTokens: roundCached
             };
           }
-          maybeWarnDegenerate(stopReason);
           yield { kind: "done", finishReason: stopReason };
         } else if (type === "message_stop") {
           if (lastTextIndex !== null) {
@@ -55092,7 +54946,6 @@ ${systemPrompt}` : systemPrompt;
               cachedTokens: roundCached
             };
           }
-          maybeWarnDegenerate(stopReason);
           yield { kind: "done", finishReason: stopReason ?? "end_turn" };
         } else if (round === 1) {
           yield { kind: "meta", chunk: rawBuf, firstRoundOnly: true };
@@ -56703,18 +56556,11 @@ data: ${JSON.stringify({ type: "content_block_delta", index, delta: { type: delt
 
 `;
   };
-  let sawToolUse = false;
-  let sawThinking = false;
-  let visibleTextChars = 0;
-  let finalFinishReason;
   const flushTails = () => {
     let out = "";
     for (const s3 of streams.values()) {
       const tail = s3.filter.flush();
-      if (tail.length > 0) {
-        out += syntheticTail(s3.field, s3.index, tail);
-        if (s3.field === "content" || s3.field === "text") visibleTextChars += tail.length;
-      }
+      if (tail.length > 0) out += syntheticTail(s3.field, s3.index, tail);
     }
     return out;
   };
@@ -56737,28 +56583,6 @@ data: ${JSON.stringify({ type: "content_block_delta", index, delta: { type: delt
       reason: "plugin chat passthrough stream ended without a completion event"
     });
   };
-  const maybeWarnDegenerate = () => {
-    if (!sawTerminal || res.destroyed || res.writableEnded) return;
-    let inputChars = 0;
-    let dropped = false;
-    for (const s3 of streams.values()) {
-      const st2 = s3.filter.stats();
-      inputChars += st2.inputChars;
-      dropped = dropped || st2.dropped;
-    }
-    const msg = degenerateTurnWarning({
-      reason: finalFinishReason,
-      terminalReason: protocol === "anthropic" ? "end_turn" : "stop",
-      toolCalls: sawToolUse ? 1 : 0,
-      text: { inputChars, outputChars: visibleTextChars, dropped },
-      sawThinking,
-      wire: `plugin-passthrough-${protocol}`
-    });
-    if (msg) {
-      log("warn", msg);
-      log2?.(msg);
-    }
-  };
   const pushField = (field, index, text) => {
     const s3 = filterFor(field, index);
     const clean = s3.filter.push(text);
@@ -56778,28 +56602,21 @@ data: ${JSON.stringify({ type: "content_block_delta", index, delta: { type: delt
     let hadText = false;
     for (let ci2 = 0; ci2 < choices.length; ci2++) {
       const ch = choices[ci2];
-      if (ch && typeof ch["finish_reason"] === "string") finalFinishReason = ch["finish_reason"];
       const d = ch?.["delta"];
       if (!d || typeof d !== "object") continue;
       const dd = d;
-      if (dd["tool_calls"] !== void 0) sawToolUse = true;
       for (const field of ["content", "reasoning_content", "reasoning"]) {
         const v2 = dd[field];
         if (typeof v2 !== "string") continue;
         hadText = true;
-        if (field !== "content" && v2.length > 0) sawThinking = true;
         if (!mayStartRenderTag(v2) && !anyPending()) {
           if (v2.length > 0) keptText = true;
-          if (field === "content") visibleTextChars += v2.length;
           continue;
         }
         const index = typeof ch?.["index"] === "number" ? ch["index"] : ci2;
         const [clean, changed] = pushField(field, index, v2);
         if (clean.length === 0) droppedText = true;
-        else {
-          keptText = true;
-          if (field === "content") visibleTextChars += clean.length;
-        }
+        else keptText = true;
         if (changed) {
           if (!rebuilt) {
             rebuilt = { ...ev, choices: choices.map((c) => ({ ...c, delta: { ...c["delta"] } })) };
@@ -56818,13 +56635,6 @@ data: ${JSON.stringify({ type: "content_block_delta", index, delta: { type: delt
     return rawEvent + "\n\n";
   };
   const processAnthropic = (ev, rawEvent) => {
-    if (ev["type"] === "content_block_start") {
-      const cb = ev["content_block"];
-      const bt2 = cb && typeof cb === "object" ? cb["type"] : void 0;
-      if (bt2 === "tool_use") sawToolUse = true;
-      else if (bt2 === "thinking" || bt2 === "redacted_thinking") sawThinking = true;
-      return anyPending() ? flushTails() + rawEvent + "\n\n" : rawEvent + "\n\n";
-    }
     if (ev["type"] !== "content_block_delta") {
       return anyPending() ? flushTails() + rawEvent + "\n\n" : rawEvent + "\n\n";
     }
@@ -56835,17 +56645,11 @@ data: ${JSON.stringify({ type: "content_block_delta", index, delta: { type: delt
       return rawEvent + "\n\n";
     }
     const raw = d[field];
-    if (field === "thinking" && raw.length > 0) sawThinking = true;
     if (!mayStartRenderTag(raw) && !anyPending()) {
-      if (field === "text" && raw.length > 0) visibleTextChars += raw.length;
       return rawEvent + "\n\n";
     }
     const [clean, changed] = pushField(field, index, raw);
-    if (!changed) {
-      if (field === "text" && raw.length > 0) visibleTextChars += raw.length;
-      return rawEvent + "\n\n";
-    }
-    if (field === "text" && clean.length > 0) visibleTextChars += clean.length;
+    if (!changed) return rawEvent + "\n\n";
     if (clean.length === 0 && Object.keys(d ?? {}).length <= 2) return "";
     return rebuildEvent(rawEvent, { ...ev, delta: { ...d, [field]: clean } });
   };
@@ -56876,10 +56680,6 @@ data: ${JSON.stringify({ type: "content_block_delta", index, delta: { type: delt
             continue;
           }
           if (ev["type"] === "message_stop") sawTerminal = true;
-          if (ev["type"] === "message_delta") {
-            const d = ev["delta"];
-            if (d && typeof d["stop_reason"] === "string") finalFinishReason = d["stop_reason"];
-          }
           const sample = usageFromSseEvent(ev);
           if (sample) mergeUsageSample(acc, sample);
           let backfilled = false;
@@ -56903,7 +56703,6 @@ data: ${JSON.stringify({ type: "content_block_delta", index, delta: { type: delt
     if (rest.length > 0 && !res.destroyed && !res.writableEnded) await write(rest);
     settleUsage();
     maybeNoteTruncated();
-    maybeWarnDegenerate();
   } catch (e) {
     settleUsage();
     maybeNoteTruncated();
@@ -56940,9 +56739,6 @@ async function pipePluginResponsesWithStrip(stream2, res, session, log2) {
     log("warn", `[tag-echo] stripped model-emitted render tag (plugin passthrough): ${snippet.slice(0, 80).replace(/\n/g, " ")}`);
     log2?.(`[tag-echo] stripped model-emitted render tag from plugin passthrough text`);
   });
-  let sawFunctionCall = false;
-  let sawReasoning = false;
-  let responseStatus;
   const write = (s3) => {
     if (!res.write(Buffer.from(s3, "utf8"))) {
       return new Promise((r) => res.once("drain", () => r()));
@@ -56963,22 +56759,6 @@ async function pipePluginResponsesWithStrip(stream2, res, session, log2) {
       reason: "plugin responses passthrough stream ended without a completion event"
     });
   };
-  const maybeWarnDegenerate = () => {
-    if (!sawTerminal || res.destroyed || res.writableEnded) return;
-    const st2 = tagFilter.stats();
-    const msg = degenerateTurnWarning({
-      reason: responseStatus,
-      terminalReason: "completed",
-      toolCalls: sawFunctionCall ? 1 : 0,
-      text: st2,
-      sawThinking: sawReasoning,
-      wire: "plugin-passthrough-responses"
-    });
-    if (msg) {
-      log("warn", msg);
-      log2?.(msg);
-    }
-  };
   let lastDeltaMeta = null;
   const flushTail = (after) => {
     const tail = tagFilter.flush();
@@ -57018,16 +56798,6 @@ async function pipePluginResponsesWithStrip(stream2, res, session, log2) {
           const sample = usageFromSseEvent(ev);
           if (sample) mergeUsageSample(acc, sample);
           const type = ev["type"];
-          if (typeof type === "string") {
-            if (type.startsWith("response.reasoning")) sawReasoning = true;
-            if (type === "response.output_item.added" || type === "response.output_item.done") {
-              const item = ev["item"];
-              const it2 = item?.["type"];
-              if (it2 === "function_call" || it2 === "custom_tool_call") sawFunctionCall = true;
-            }
-            const resp = ev["response"];
-            if (resp && typeof resp["status"] === "string") responseStatus = resp["status"];
-          }
           if (type === "response.output_text.done" || type === "response.content_part.done" || type === "response.output_item.done" || type === "response.completed" || type === "response.failed" || type === "response.incomplete") {
             if (type === "response.completed" || type === "response.failed" || type === "response.incomplete") sawTerminal = true;
             let evOut = ev;
@@ -57073,7 +56843,6 @@ async function pipePluginResponsesWithStrip(stream2, res, session, log2) {
       const rest = flushTail("");
       if (rest.length > 0) await write(rest);
     }
-    maybeWarnDegenerate();
     settleUsage();
     maybeNoteTruncated();
   } catch (e) {
@@ -59149,7 +58918,6 @@ async function handle(req, res, opts, core, config, log2, instanceId, instanceSt
     try {
       const result = await fetchWithTimeout(targetUrl2, {
         method: "HEAD",
-        redirect: "follow",
         ...proxyUrl ? { dispatcher: proxyDispatcher(proxyUrl, 15e3) } : {}
       }, 15e3);
       result.clearTimer();
@@ -59346,11 +59114,7 @@ ${bodyBuffer.toString("utf8")}`);
     }
   }
   let prepared = null;
-  const routePassthrough = !opts.passthrough && findRoute(opts.routes, route?.rewrittenUrl)?.passthrough === true;
-  if (routePassthrough && hopMarker === void 0 && protocol && parsed) {
-    log2("info", `[route-passthrough] ${maskUrlsInText(route?.rewrittenUrl ?? "")} matches a passthrough route \u2014 forwarding verbatim, kernel bypassed`);
-  }
-  if (!opts.passthrough && !routePassthrough && hopMarker === void 0 && protocol && parsed && typeof parsed === "object") {
+  if (!opts.passthrough && hopMarker === void 0 && protocol && parsed && typeof parsed === "object") {
     const sessionHeader = headerValue2(req, opts.sessionHeader);
     let pluginAgent = pluginAgentHeader(req.headers);
     let pluginConversation = pluginConversationHeader(req.headers);
@@ -59611,7 +59375,7 @@ ${bodyBuffer.toString("utf8")}`);
     }
   }
   if (!prepared) {
-    if (protocol === null && !opts.passthrough && !routePassthrough && !isModelDiscoveryPath(urlPath)) {
+    if (protocol === null && !opts.passthrough && !isModelDiscoveryPath(urlPath)) {
       logUnrecognizedPath(log2, req.url ?? "");
     }
     await forward(req, res, opts, bodyBuffer, null, core, reqConfig, log2, route, instanceId, void 0);
@@ -65096,22 +64860,17 @@ args = [${JSON.stringify(path14.join(selfPackageRoot(), "dist", "mcp.js"))}]
 env = { BILI_MCP_PROXY = ${JSON.stringify(proxyOriginForInstall())} }
 `;
 }
-function malformedCodexArgs(block) {
-  return /^[ \t]*args[ \t]*=[ \t]*["']/m.test(block);
-}
 function codexInstall() {
   const file = codexToml();
   const text = fs11.existsSync(file) ? fs11.readFileSync(file, "utf8") : "";
   const existing = /^[ \t]*\[mcp_servers\.bili\][ \t]*$/m.exec(text);
   if (existing !== null) {
     const block = text.slice(existing.index, text.indexOf("\n[", existing.index + 1) === -1 ? void 0 : text.indexOf("\n[", existing.index + 1));
-    const canonical = codexBlock().replace(/^\n/, "");
-    if (block.trimEnd() === canonical.trimEnd()) return `codex: already installed (${file})`;
-    const refreshed = text.slice(0, existing.index) + canonical + text.slice(existing.index + block.length);
+    if (block.includes(`BILI_MCP_PROXY = ${JSON.stringify(proxyOriginForInstall())}`)) return `codex: already installed (${file})`;
+    const refreshed = text.slice(0, existing.index) + codexBlock().replace(/^\n/, "") + text.slice(existing.index + block.length);
     backupOnce(file);
     fs11.writeFileSync(file, refreshed);
-    const healed = malformedCodexArgs(block) ? " (repaired args: was not an array)" : "";
-    return `codex: refreshed [mcp_servers.bili] -> ${file}${healed}`;
+    return `codex: refreshed proxy origin -> ${file} [mcp_servers.bili]`;
   }
   fs11.mkdirSync(path14.dirname(file), { recursive: true });
   backupOnce(file);
@@ -66193,31 +65952,10 @@ function stopProxy(handle2) {
   } catch {
   }
 }
-function quoteWinToken(token) {
-  if (!/\s/.test(token) || token.includes('"')) return token;
-  return `"${token}"`;
-}
-function buildWindowsCommandLine(cmd, args) {
-  return `"${[cmd, ...args].map(quoteWinToken).join(" ")}"`;
-}
-function planClientSpawn(cmd, args, env, platform = process.platform) {
-  if (platform !== "win32") return { command: cmd, args: [...args] };
-  const lower = cmd.toLowerCase();
-  const base = cmd.slice(Math.max(cmd.lastIndexOf("/"), cmd.lastIndexOf("\\")) + 1);
-  const needsCmd = lower.endsWith(".cmd") || lower.endsWith(".bat") || !path15.extname(base);
-  if (!needsCmd) return { command: cmd, args: [...args] };
-  const comspec = nonEmpty2(env.COMSPEC) ? env.COMSPEC : "cmd.exe";
-  return {
-    command: comspec,
-    args: ["/d", "/s", "/c", buildWindowsCommandLine(cmd, args)],
-    windowsVerbatimArguments: true
-  };
-}
 function runClient(cmd, args, env, deps) {
   const spawnImpl = deps?.spawnImpl ?? spawn;
-  const plan = planClientSpawn(cmd, args, env, deps?.platform);
   return new Promise((resolve, reject) => {
-    const child = spawnImpl(plan.command, plan.args, { stdio: "inherit", env, windowsVerbatimArguments: plan.windowsVerbatimArguments });
+    const child = spawnImpl(cmd, args, { stdio: "inherit", env, shell: process.platform === "win32" });
     child.on?.("error", (...rest) => reject(rest[0]));
     child.on?.("exit", (...rest) => {
       const code = rest[0];