billion-context 0.1.101 → 0.1.102-pr.641.518

package/README.md

@@ -158,7 +158,7 @@ rewrite until dsh gains a settings-path env or an upstream loopback opt-out.
 Overlay dirs created by older versions are left in place and never merged back
 into the real home.
 
-### Option 1 — Launcher (`bili pi` / `bili codex` / `bili claude` / `bili omp` / `bili opencode` / `bili hermes` / `bili dsh`)
+### Option 1 — Launcher (`bili pi` / `bili codex` / `bili claude` / `bili omp` / `bili opencode` / `bili hermes` / `bili dsh` / `bili codebuddy`)
 
 The launcher wraps a client in one command: it starts a proxy on an
 independent port (a fresh instance is always spawned — a port is never
@@ -177,6 +177,7 @@ bili omp                              # pi-style, file-free (#535): env + extens
 bili opencode                         # MITM for HTTPS + temp opencode.json (/bili/ for HTTP) + thin /acp plugin
 bili hermes                           # file-free (#535): hermes proxy env (HTTPS_PROXY + HERMES_CA_BUNDLE) — https via CONNECT MITM, http via absolute-form forward proxy; real ~/.hermes untouched
 bili dsh                              # deepseek-harness: non-loopback upstreams ride proxy envs (https MITM, http absolute-form), loopback keeps the overlay DSH_HOME (~/.dsh-bili) rewrite (#535), built-in deepseek route via DEEPSEEK_BASE_URL, native /acp command injected via --patch
+bili codebuddy                        # Tencent CodeBuddy Code CLI: CODEBUDDY_BASE_URL /bili/ rewrite (OpenAI chat completions wire), budget aligned via CODEBUDDY_AUTO_COMPACT_WINDOW; real ~/.codebuddy untouched
 bili pi --mitm-domain api.foo.com     # add a domain to the MITM whitelist
 ```
 

package/dist/index.js

@@ -43467,7 +43467,7 @@ var require_lib = __commonJS({
   }
 });
 
-// node_modules/acp-kernel/dist/chunk-JGNVAFFE.js
+// node_modules/acp-kernel/dist/chunk-6TAK7DSI.js
 import { createRequire } from "module";
 var require2 = createRequire(import.meta.url);
 function defaultCountTokens(text) {
@@ -43476,6 +43476,12 @@ function defaultCountTokens(text) {
   const cjkCount = cjk?.length ?? 0;
   return cjkCount + Math.ceil((text.length - cjkCount) / 4);
 }
+function thinkingTokenValue(thinking) {
+  return typeof thinking === "number" && Number.isFinite(thinking) && thinking > 0 ? thinking : 0;
+}
+function countMessageTokens(message, countTokens = defaultCountTokens) {
+  return countTokens(message.text ?? "") + thinkingTokenValue(message.thinkingTokens);
+}
 function estimateTokensFast(text) {
   if (!text) return 0;
   return Math.ceil(text.length / 4);
@@ -45283,7 +45289,8 @@ function renderMessage(message, map, countTokens, strategy, snapshot = null) {
     "^" + escapeRegex(TAG_OPEN) + "[^>]*" + GT + escapeRegex(ref) + escapeRegex(TAG_CLOSE) + "\\n?"
   );
   const cleanText = (message.text || "").replace(ownTagRe, "");
-  const tokens = snapshot ? snapshot[ref] ?? (snapshot[ref] = countTokens(cleanText)) : countTokens(cleanText);
+  const textTokens = snapshot ? snapshot[ref] ?? (snapshot[ref] = countTokens(cleanText)) : countTokens(cleanText);
+  const tokens = textTokens + thinkingTokenValue(message.thinkingTokens);
   const type = classifyType(message);
   const prefix = acpTag(ref, tokens, type) + "\n";
   if (!cleanText) return { ...message, text: prefix };
@@ -45362,8 +45369,12 @@ function adjustBoundariesForReasoningPairs(startIndex, endIndex, messages) {
         j2++;
       }
       const companion = messages[j2 + 1];
-      if (companion !== void 0 && companion.role === "assistant" && (companion.contentType === "text" || companion.contentType === "tool-call") && j2 + 1 > newEndIndex) {
-        newEndIndex = j2 + 1;
+      if (companion !== void 0 && companion.role === "assistant" && (companion.contentType === "text" || companion.contentType === "tool-call")) {
+        let e = j2 + 1;
+        while (e + 1 < messages.length && messages[e + 1].role === "assistant" && (messages[e + 1].contentType === "text" || messages[e + 1].contentType === "tool-call")) {
+          e++;
+        }
+        if (e > newEndIndex) newEndIndex = e;
       }
     }
     if (msg.role === "assistant" && (msg.contentType === "text" || msg.contentType === "tool-call")) {
@@ -45379,6 +45390,64 @@ function adjustBoundariesForReasoningPairs(startIndex, endIndex, messages) {
   }
   return { startIndex: newStartIndex, endIndex: newEndIndex };
 }
+function isAssistantAct(msg) {
+  return msg.role === "assistant" && (msg.contentType === "text" || msg.contentType === "tool-call");
+}
+function computeTurnGroups(messages) {
+  const resultIdByCallId = /* @__PURE__ */ new Map();
+  for (const msg of messages) {
+    if (msg.contentType === "tool-result" && typeof msg.toolCallId === "string" && msg.id) {
+      if (!resultIdByCallId.has(msg.toolCallId))
+        resultIdByCallId.set(msg.toolCallId, msg.id);
+    }
+  }
+  const grouped = /* @__PURE__ */ new Set();
+  const groups = [];
+  for (let i = 0; i < messages.length; i++) {
+    const msg = messages[i];
+    if (!msg.id || grouped.has(msg.id)) continue;
+    if (!(msg.contentType === "reasoning" || isAssistantAct(msg))) continue;
+    let reasoningStart = i;
+    if (msg.contentType === "reasoning") {
+      while (reasoningStart > 0 && messages[reasoningStart - 1].contentType === "reasoning") {
+        reasoningStart--;
+      }
+    } else {
+      let s3 = i;
+      while (s3 > 0 && isAssistantAct(messages[s3 - 1])) s3--;
+      reasoningStart = s3;
+      while (reasoningStart > 0 && messages[reasoningStart - 1].contentType === "reasoning") {
+        reasoningStart--;
+      }
+    }
+    const burstStart = (() => {
+      let s3 = reasoningStart;
+      while (s3 < messages.length && messages[s3].contentType === "reasoning")
+        s3++;
+      return s3;
+    })();
+    if (burstStart >= messages.length || !isAssistantAct(messages[burstStart])) {
+      continue;
+    }
+    let burstEnd = burstStart;
+    while (burstEnd + 1 < messages.length && isAssistantAct(messages[burstEnd + 1])) {
+      burstEnd++;
+    }
+    const members = /* @__PURE__ */ new Set();
+    for (let k2 = reasoningStart; k2 <= burstEnd; k2++) {
+      const m2 = messages[k2];
+      if (!m2.id) continue;
+      members.add(m2.id);
+      if (m2.role === "assistant" && m2.contentType === "tool-call" && typeof m2.toolCallId === "string") {
+        const rid = resultIdByCallId.get(m2.toolCallId);
+        if (rid) members.add(rid);
+      }
+    }
+    for (const id of members) grouped.add(id);
+    groups.push([...members]);
+  }
+  return groups;
+}
 function estimateTextTokens(text) {
   return Math.ceil(text.length / 4);
 }
@@ -45402,7 +45471,7 @@ function computeProtectedRefs(messages, state, config, countTokens = estimateTex
     if (isNeverPreserveRecent(msg)) continue;
     const ref = state.messageRefs.byRaw[msg.id];
     if (!ref || ref === "BLOCKED") continue;
-    visible.push({ ref, tokens: countTokens(msg.text ?? "") });
+    visible.push({ ref, tokens: countMessageTokens(msg, countTokens) });
   }
   if (preserveN > 0) {
     for (const m2 of visible.slice(-preserveN)) {
@@ -45445,7 +45514,7 @@ function buildCompressibleRanges(messages, state, config, protectedZoneRefs, cou
       protectedMsgs.push({
         ref,
         gapBefore: skipSinceProtected,
-        tokens: countTokens(msg.text ?? ""),
+        tokens: countMessageTokens(msg, countTokens),
         tools: msg.toolName ? [msg.toolName] : []
       });
       skipSinceProtected = false;
@@ -45460,7 +45529,7 @@ function buildCompressibleRanges(messages, state, config, protectedZoneRefs, cou
     compressibleMsgs.push({
       ref,
       gapBefore: skipSinceCompressible,
-      tokens: countTokens(msg.text ?? ""),
+      tokens: countMessageTokens(msg, countTokens),
       chars: (msg.text ?? "").length,
       isTool: isToolMessage(msg),
       isUser: msg.role === "user"
@@ -46093,6 +46162,42 @@ function applySingleRange(input) {
       )} from compression range (recent/last-user zone).`
     );
   }
+  {
+    const reasoningIds = /* @__PURE__ */ new Set();
+    const callIds = /* @__PURE__ */ new Set();
+    for (const m2 of input.messages) {
+      if (!m2.id) continue;
+      if (m2.contentType === "reasoning") reasoningIds.add(m2.id);
+      if (m2.role === "assistant" && m2.contentType === "tool-call") callIds.add(m2.id);
+    }
+    const withdrawIds = /* @__PURE__ */ new Set();
+    let splitTurnCount = 0;
+    for (const group of computeTurnGroups(input.messages)) {
+      const foldHasReasoning = group.some(
+        (id) => effectiveMessageIds.has(id) && reasoningIds.has(id)
+      );
+      if (!foldHasReasoning) continue;
+      const keptHasCall = group.some(
+        (id) => !effectiveMessageIds.has(id) && callIds.has(id)
+      );
+      if (!keptHasCall) continue;
+      splitTurnCount++;
+      for (const id of group) withdrawIds.add(id);
+    }
+    if (withdrawIds.size > 0) {
+      for (const id of withdrawIds) effectiveMessageIds.delete(id);
+      const beforeWithdraw = filteredIds.length;
+      filteredIds = filteredIds.filter((id) => !withdrawIds.has(id));
+      if (filteredIds.length === 0 && consumedBlockIds.length === 0) {
+        throw new Error(
+          `Range would split ${splitTurnCount} turn(s) at the protected-zone boundary: a visible tool-call must keep its reasoning run (strict-echo providers reject a rebuilt request that lost it). Shrink the range to end before the turn starts, or wait until the whole turn ages out of the protected zone.`
+        );
+      }
+      warnings.push(
+        `Withdrawn ${beforeWithdraw - filteredIds.length} message(s) from compression range to keep ${splitTurnCount} turn(s) intact (visible tool-call would lose its reasoning run).`
+      );
+    }
+  }
   if (!isBlockBoundary && filteredIds.length === 0 && consumedBlockIds.length > 0) {
     const first = consumedBlockIds[0];
     const last = consumedBlockIds[consumedBlockIds.length - 1];
@@ -46106,7 +46211,7 @@ function applySingleRange(input) {
   let compressedTokens = 0;
   for (const id of filteredIds) {
     const message = input.messages.find((entry) => entry.id === id);
-    compressedTokens += input.countTokens(message?.text ?? "");
+    compressedTokens += message ? countMessageTokens(message, input.countTokens) : 0;
   }
   for (const consumedId of consumedBlockIds) {
     const consumed = blockById(input.state, consumedId);
@@ -46436,7 +46541,7 @@ function computeContextBreakdown(messages, total, growth, countTokens) {
   const count = countTokens ?? ((t) => Math.ceil(t.length / 4));
   let system = 0, tool = 0, summaries = 0, code = 0, text = 0;
   for (const msg of messages) {
-    const tokens = count(msg.text ?? "");
+    const tokens = countMessageTokens(msg, count);
     if (msg.text?.startsWith("[Compressed conversation section]")) {
       summaries += tokens;
     } else if (msg.contentType === "tool-call" || msg.contentType === "tool-result") {
@@ -46622,7 +46727,7 @@ function collectVisible(messages, state, countTokens) {
     if (coveredIds.has(message.id)) return;
     const ref = refForRaw(state.messageRefs, message.id);
     if (!ref) return;
-    const tokens = countTokens(message.text ?? "");
+    const tokens = countMessageTokens(message, countTokens);
     const tool = isToolMessage(message) ? message.toolName ?? (message.toolCallId ? toolCallNames.get(message.toolCallId) : void 0) ?? "tool" : "text";
     if (tokens > 0) visible.push({ ref, tokens, tool, index });
   });
@@ -53721,6 +53826,11 @@ async function* runCompressLoop(upstream, ctx, requestBody, requestOptions, adap
         }
       } catch (e) {
         if (!(e instanceof UpstreamHttpError)) throw e;
+        if (e.status === 400 && /reasoning_content/i.test(e.body) && ctx.session.metadata.strictReasoningEcho !== true) {
+          ctx.session.metadata.strictReasoningEcho = true;
+          ctx.log(`[acp-loop] 400 mentions reasoning_content \u2014 learned strict reasoning echo for this session; #651 reasoning-drop disabled (#684)`);
+          log("warn", `[acp-loop] learned strictReasoningEcho (session ${ctx.session.id}); reasoning-drop disabled (#684)`);
+        }
         const suffix = e.attempts > 1 ? ` after ${e.attempts} attempt(s)` : "";
         ctx.log(`[acp-proxy: compress loop upstream error ${e.status}${suffix}: ${e.body.slice(0, 200)}]`);
         log("error", `[acp-loop] upstream error ${e.status}${suffix}: ${e.body.slice(0, 200)}`);
@@ -57204,6 +57314,98 @@ function resolveCodexHome(env) {
   const h = os.homedir();
   return nonEmpty2(env.CODEX_HOME) ? env.CODEX_HOME : path10.join(h, ".codex");
 }
+function resolveCodebuddyHome(env) {
+  const h = os.homedir();
+  return nonEmpty2(env.CODEBUDDY_CONFIG_DIR) ? env.CODEBUDDY_CONFIG_DIR : path10.join(h, ".codebuddy");
+}
+function parseCodebuddyModelsJson(obj) {
+  const out = { models: [], urls: [] };
+  const seenUrl = /* @__PURE__ */ new Set();
+  const collect = (id, entry) => {
+    if (!entry || typeof entry !== "object" || Array.isArray(entry)) return;
+    const e = entry;
+    const url = e.url;
+    if (typeof url === "string" && url.length > 0 && !seenUrl.has(url)) {
+      seenUrl.add(url);
+      out.urls.push(url);
+    }
+    const win = toModelWindow(id, e.maxInputTokens);
+    if (win) out.models.push(win);
+  };
+  if (!obj) return out;
+  if (Array.isArray(obj)) {
+    for (const item of obj) {
+      if (!item || typeof item !== "object" || Array.isArray(item)) continue;
+      const it2 = item;
+      collect(it2.id ?? it2.name, it2);
+    }
+    return out;
+  }
+  if (typeof obj !== "object") return out;
+  const root = obj;
+  const modelsField = root.models;
+  if (Array.isArray(modelsField)) {
+    for (const item of modelsField) {
+      if (!item || typeof item !== "object" || Array.isArray(item)) continue;
+      const it2 = item;
+      collect(it2.id ?? it2.name, it2);
+    }
+    return out;
+  }
+  if (modelsField && typeof modelsField === "object") {
+    for (const [id, val] of Object.entries(modelsField)) collect(id, val);
+    return out;
+  }
+  for (const [id, val] of Object.entries(root)) collect(id, val);
+  return out;
+}
+function readJsonFile(filePath) {
+  try {
+    return JSON.parse(fs6.readFileSync(filePath, "utf8"));
+  } catch {
+    return null;
+  }
+}
+function readCodebuddyConfig(codebuddyHome, cwd, env = process.env) {
+  let codebuddyBaseUrl;
+  let model;
+  let autoCompactWindow;
+  const settings = readJsonObject(path10.join(codebuddyHome, "settings.json"));
+  const settingsEnv = settings?.env;
+  if (settingsEnv && typeof settingsEnv === "object" && !Array.isArray(settingsEnv)) {
+    const e = settingsEnv;
+    const v2 = e.CODEBUDDY_BASE_URL;
+    if (nonEmpty2(v2)) codebuddyBaseUrl = v2;
+  }
+  const tm = settings?.model;
+  if (nonEmpty2(tm)) model = String(tm);
+  const tacw = Number(settings?.autoCompactWindow);
+  if (Number.isFinite(tacw) && tacw > 0) autoCompactWindow = tacw;
+  if (!codebuddyBaseUrl && nonEmpty2(env.CODEBUDDY_BASE_URL)) codebuddyBaseUrl = env.CODEBUDDY_BASE_URL;
+  const windowByModel = /* @__PURE__ */ new Map();
+  const urls = [];
+  const seenUrl = /* @__PURE__ */ new Set();
+  for (const f2 of [
+    path10.join(codebuddyHome, "models.json"),
+    path10.join(cwd, ".codebuddy", "models.json")
+  ]) {
+    const parsed = parseCodebuddyModelsJson(readJsonFile(f2));
+    for (const w2 of parsed.models) windowByModel.set(w2.id, w2.contextWindow);
+    for (const u2 of parsed.urls) {
+      if (!seenUrl.has(u2)) {
+        seenUrl.add(u2);
+        urls.push(u2);
+      }
+    }
+  }
+  return {
+    ...codebuddyBaseUrl ? { codebuddyBaseUrl } : {},
+    ...model ? { model } : {},
+    ...autoCompactWindow ? { autoCompactWindow } : {},
+    ...windowByModel.size > 0 ? { models: [...windowByModel.entries()].map(([id, contextWindow]) => ({ id, contextWindow })) } : {},
+    ...urls.length > 0 ? { modelUrls: urls } : {}
+  };
+}
 function parseDshSettingsYaml(text) {
   const out = [];
   const seen = /* @__PURE__ */ new Set();
@@ -57577,6 +57779,7 @@ function loadClientConfig(env, cwd) {
   config.opencode = readOpencodeConfig(resolveOpencodeConfigFile(env));
   config.hermes = readHermesConfig(resolveHermesHome(env));
   config.dsh = readDshConfig(resolveDshHome(env));
+  config.codebuddy = readCodebuddyConfig(resolveCodebuddyHome(env), cwd, env);
   return config;
 }
 function collectModelWindows(config, scope) {
@@ -57591,12 +57794,14 @@ function collectModelWindows(config, scope) {
     else if (scope === "pi") for (const p2 of Object.values(config.pi?.providers ?? {})) add(p2.models);
     else if (scope === "omp") for (const p2 of Object.values(config.omp?.providers ?? {})) add(p2.models);
     else if (scope === "opencode") for (const p2 of Object.values(config.opencode?.providers ?? {})) add(p2.models);
+    else if (scope === "codebuddy") add(config.codebuddy?.models);
     return out;
   }
   for (const p2 of Object.values(config.pi?.providers ?? {})) add(p2.models);
   for (const p2 of Object.values(config.omp?.providers ?? {})) add(p2.models);
   for (const p2 of Object.values(config.opencode?.providers ?? {})) add(p2.models);
   add(config.codex?.modelWindows);
+  add(config.codebuddy?.models);
   return out;
 }
 
@@ -57634,18 +57839,26 @@ function extractHttpsHosts(config) {
   if (config.zcode) {
     for (const prov of Object.values(config.zcode.providers)) push(prov.baseURL);
   }
+  if (config.codebuddy) {
+    push(config.codebuddy.codebuddyBaseUrl);
+    for (const u2 of config.codebuddy.modelUrls ?? []) push(u2);
+  }
   return out;
 }
 function configFilePaths(env) {
   const home = os2.homedir();
   const codexHome = nonEmpty2(env.CODEX_HOME) ? env.CODEX_HOME : path11.join(home, ".codex");
   const zcodeHome = nonEmpty2(env.ZCODE_DATA_BASE_DIR) ? env.ZCODE_DATA_BASE_DIR : path11.join(home, ".zcode");
+  const codebuddyHome = resolveCodebuddyHome(env);
   return [
     path11.join(home, ".claude", "settings.json"),
     path11.join(process.cwd(), ".claude", "settings.json"),
     path11.join(codexHome, "config.toml"),
     path11.join(resolvePiHome(env), "models.json"),
-    path11.join(zcodeHome, "v2", "config.json")
+    path11.join(zcodeHome, "v2", "config.json"),
+    path11.join(codebuddyHome, "settings.json"),
+    path11.join(codebuddyHome, "models.json"),
+    path11.join(process.cwd(), ".codebuddy", "models.json")
   ];
 }
 function readMtimes(paths) {
@@ -59536,7 +59749,7 @@ ${bodyBuffer.toString("utf8")}`);
             log2("info", `[debug] strip-images: dropped ${stripped.removed} historical image part(s), kept last ${keepRecent} (session=${session.id})`);
           }
           const work = stripped.body;
-          return countTokens ? prepareCountTokens(work, core, reqConfig, log2, session) : protocol === "anthropic" ? prepareAnthropic(work, req, opts, core, reqConfig, reqPrompts, log2, session, pluginMode, reasoningCfg) : protocol === "openai" ? prepareOpenai(work, req, opts, core, reqConfig, reqPrompts, log2, session, pluginMode, nativeWindow, reasoningCfg) : responsesCompact ? prepareResponsesCompact(stripped.removed > 0 ? Buffer.from(JSON.stringify(work)) : bodyBuffer, work, session, req, core, reqConfig, log2) : prepareResponses(work, req, opts, core, reqConfig, reqPrompts, log2, session, responsesIdentity, pluginMode, upstreamOrigin, nativeWindow, reasoningCfg);
+          return countTokens ? prepareCountTokens(work, core, reqConfig, log2, session) : protocol === "anthropic" ? prepareAnthropic(work, req, opts, core, reqConfig, reqPrompts, log2, session, pluginMode, upstreamOrigin, reasoningCfg) : protocol === "openai" ? prepareOpenai(work, req, opts, core, reqConfig, reqPrompts, log2, session, pluginMode, upstreamOrigin, nativeWindow, reasoningCfg) : responsesCompact ? prepareResponsesCompact(stripped.removed > 0 ? Buffer.from(JSON.stringify(work)) : bodyBuffer, work, session, req, core, reqConfig, log2) : prepareResponses(work, req, opts, core, reqConfig, reqPrompts, log2, session, responsesIdentity, pluginMode, upstreamOrigin, nativeWindow, reasoningCfg);
         };
         const isCodexCompactTrigger = protocol === "responses" && !responsesCompact && isCodexClient(req.headers) && hasCompactionTrigger(parsed.input);
         if (isCodexCompactTrigger) {
@@ -59624,13 +59837,64 @@ ${bodyBuffer.toString("utf8")}`);
   }
 }
 var ACP_TAG_MARK = "<acp ";
-function withReasoningDrop(msgs, reasoning, log2, sessionId) {
+function withReasoningDrop(msgs, reasoning, log2, sessionId, strictEcho) {
+  if (strictEcho) return msgs;
   const out = dropCompressReasoning(msgs, reasoning);
   if (out.length !== msgs.length) {
     log2("info", `[${sessionId}] compress-reasoning: dropped ${msgs.length - out.length} reasoning message(s) from closed compress turns (#651)`);
   }
   return out;
 }
+function isStrictReasoningEcho(session, upstreamOrigin) {
+  if (session.metadata.strictReasoningEcho === true) return true;
+  return upstreamOrigin !== void 0 && /deepseek/i.test(upstreamOrigin);
+}
+function warnReasoningPairs(wireMessages, log2, sessionId) {
+  let withRc = 0;
+  let split = 0;
+  for (const m2 of wireMessages) {
+    const msg = m2;
+    if (msg?.role !== "assistant") continue;
+    const hasRc = typeof msg.reasoning_content === "string" && msg.reasoning_content.length > 0;
+    if (hasRc) withRc++;
+    else if (Array.isArray(msg.tool_calls) && msg.tool_calls.length > 0) split++;
+  }
+  if (withRc > 0 && split > 0) {
+    log2("warn", `[${sessionId}] reasoning-pair-violated: ${split} assistant tool-call message(s) lack reasoning_content while ${withRc} carry it \u2014 strict-echo upstreams (DeepSeek thinking mode) will reject the request (#684)`);
+  }
+}
+function warnAnthropicThinkingPairs(wireMessages, log2, sessionId) {
+  let withThinking = 0;
+  let split = 0;
+  for (const m2 of wireMessages) {
+    const msg = m2;
+    if (msg?.role !== "assistant" || !Array.isArray(msg.content)) continue;
+    const hasThinking = msg.content.some((b2) => b2?.type === "thinking");
+    if (hasThinking) withThinking++;
+    else if (msg.content.some((b2) => b2?.type === "tool_use")) split++;
+  }
+  if (withThinking > 0 && split > 0) {
+    log2("warn", `[${sessionId}] thinking-pair-violated: ${split} assistant tool_use message(s) lack a thinking block while ${withThinking} carry one \u2014 extended-thinking tool use requires preserved thinking (#684)`);
+  }
+}
+function warnResponsesReasoningPairs(input, log2, sessionId) {
+  let withReasoning = 0;
+  let split = 0;
+  let prevWasReasoning = false;
+  for (const item of input) {
+    const it2 = item;
+    if (it2?.type === "reasoning") {
+      withReasoning++;
+      prevWasReasoning = true;
+      continue;
+    }
+    if (it2?.type === "function_call" && !prevWasReasoning) split++;
+    prevWasReasoning = false;
+  }
+  if (withReasoning > 0 && split > 0) {
+    log2("warn", `[${sessionId}] reasoning-pair-violated: ${split} function_call item(s) lack a preceding reasoning item while ${withReasoning} exist \u2014 strict-echo upstreams will reject the request (#684)`);
+  }
+}
 function stripKernelSummaries(messages, state) {
   const carried = /* @__PURE__ */ new Set();
   for (const b2 of state.blocks) {
@@ -59734,13 +59998,13 @@ function effectiveTokenCount(session, msgs) {
   if (!session.metadata.anonymousPrefixAffinity) return 0;
   return estimateCoreMessagesUpper(msgs);
 }
-function prepareAnthropic(parsed, req, opts, core, config, prompts, log2, session, pluginMode, reasoning) {
+function prepareAnthropic(parsed, req, opts, core, config, prompts, log2, session, pluginMode, upstreamOrigin, reasoning) {
   const sessionId = session.id;
   const stream2 = parsed.stream === true;
   ++session.stats.requests;
   session.hostCreditTokens = 0;
   const injectTools = opts.compress.injectTool && !pluginMode;
-  const stripReasoning = (msgs) => withReasoningDrop(msgs, reasoning, log2, sessionId);
+  const stripReasoning = (msgs) => withReasoningDrop(msgs, reasoning, log2, sessionId, isStrictReasoningEcho(session, upstreamOrigin));
   if (isAutoModeClassifier(parsed)) {
     log2("info", `[${sessionId}] auto-mode classifier passthrough (skipping compress injection)`);
     return { body: JSON.stringify(parsed), session, processedMessages: [], originalMessages: [], anthropicSystem: parsed.system, protocol: "anthropic", stream: stream2, compressInjected: false, pluginMode, nudge: void 0, prompts };
@@ -59803,6 +60067,7 @@ function prepareAnthropic(parsed, req, opts, core, config, prompts, log2, sessio
   snapshotMessages(session, originalMessages);
   markDirty(session);
   const rebuilt = { ...parsed, messages: rebuiltMessages, system: systemOut, tools: toolsOut };
+  warnAnthropicThinkingPairs(rebuiltMessages, log2, sessionId);
   delete rebuilt.prompt_cache_key;
   armHostUsageCredit(session, originalMessages, processedMessages, req.headers, opts.hostUsageCredit, log2);
   return { body: JSON.stringify(rebuilt), session, processedMessages, originalMessages, anthropicSystem: parsed.system, protocol: "anthropic", stream: stream2, compressInjected: injectTools, pluginMode, nudge, prompts, renderTags: "text-only" };
@@ -59860,13 +60125,13 @@ function clampOutgoingOutput(rebuilt, field, ctx, sessionId, log2) {
     log2("info", `[${sessionId}] output budget clamped ${raw} -> ${capped} (input~${inputEstimate}, window=${ctx.nativeWindow}); prevents input+output overflow (#453)`);
   }
 }
-function prepareOpenai(parsed, req, opts, core, config, prompts, log2, session, pluginMode, nativeWindow, reasoning) {
+function prepareOpenai(parsed, req, opts, core, config, prompts, log2, session, pluginMode, upstreamOrigin, nativeWindow, reasoning) {
   const sessionId = session.id;
   const stream2 = parsed.stream === true;
   ++session.stats.requests;
   session.hostCreditTokens = 0;
   let openaiSystemText = "";
-  const stripReasoning = (msgs) => withReasoningDrop(msgs, reasoning, log2, sessionId);
+  const stripReasoning = (msgs) => withReasoningDrop(msgs, reasoning, log2, sessionId, isStrictReasoningEcho(session, upstreamOrigin));
   let openaiOutboundSystem;
   let processedMessages = [];
   let originalMessages = [];
@@ -59931,6 +60196,7 @@ function prepareOpenai(parsed, req, opts, core, config, prompts, log2, session,
     processedMessages = [];
   }
   const rebuilt = { ...parsed, messages: rebuiltMessages, tools: toolsOut };
+  warnReasoningPairs(rebuiltMessages, log2, sessionId);
   clampOutgoingOutput(rebuilt, typeof parsed.max_completion_tokens === "number" ? "max_completion_tokens" : "max_tokens", { systemText: openaiSystemText, tools: toolsOut, processedMessages, lastInputTokens: session.stats.lastInputTokens, nativeWindow, imageTokens: imageTokensInParsedBody("openai", rebuilt) }, sessionId, log2);
   delete rebuilt.prompt_cache_retention;
   if (stream2 && rebuilt.stream_options === void 0) {
@@ -59949,7 +60215,7 @@ function prepareResponses(parsed, req, opts, core, config, prompts, log2, sessio
   const stream2 = parsed.stream === true;
   ++session.stats.requests;
   session.hostCreditTokens = 0;
-  const stripReasoning = (msgs) => withReasoningDrop(msgs, reasoning, log2, sessionId);
+  const stripReasoning = (msgs) => withReasoningDrop(msgs, reasoning, log2, sessionId, isStrictReasoningEcho(session, upstreamOrigin));
   if (reconcileNativeCompactionBoundary(session)) {
     log2("info", `[${sessionId}] reconciled ACP state after native Responses compact boundary`);
   }
@@ -60067,6 +60333,7 @@ function prepareResponses(parsed, req, opts, core, config, prompts, log2, sessio
     log2("info", `[${sessionId}] codex compact intercepted (trigger); forged SSE with ${summaries.length} block summary(s), upstream not contacted`);
   }
   const rebuilt = { ...parsed, input: rebuiltInput, tools: toolsOut };
+  warnResponsesReasoningPairs(Array.isArray(rebuiltInput) ? rebuiltInput : [], log2, sessionId);
   if (!isCompactionTrigger) {
     clampOutgoingOutput(rebuilt, "max_output_tokens", { systemText: (responsesProjection?.systemParts ?? []).join("\n"), tools: toolsOut, processedMessages, lastInputTokens: session.stats.lastInputTokens, nativeWindow, imageTokens: imageTokensInParsedBody("responses", rebuilt) }, sessionId, log2);
   }
@@ -65219,7 +65486,7 @@ function selfDistFile(name) {
   return path15.join(selfPackageRoot(), "dist", name);
 }
 var LAUNCHER_DEFAULT_HOST = "127.0.0.1";
-var LAUNCH_CLIENTS = ["pi", "codex", "claude", "omp", "opencode", "hermes", "dsh", "pi-test"];
+var LAUNCH_CLIENTS = ["pi", "codex", "claude", "omp", "opencode", "hermes", "dsh", "codebuddy", "pi-test"];
 var HEALTH_PATH = "/__bili/health";
 var HEALTH_POLL_INTERVAL_MS = 200;
 var SPAWN_WAIT_MS = 2e4;
@@ -65307,6 +65574,29 @@ function discoverRoutes(client, config) {
       }
     } catch {
     }
+  } else if (client === "codebuddy") {
+    const raw = nonEmpty2(config.codebuddy?.codebuddyBaseUrl) ? config.codebuddy.codebuddyBaseUrl : "https://tencent.sso.codebuddy.cn/v2";
+    const real = unwrapUpstream2(raw);
+    try {
+      const url = new URL(real);
+      if ((url.protocol === "https:" || url.protocol === "http:") && !rewriteKeys.has("CODEBUDDY_BASE_URL")) {
+        rewriteKeys.add("CODEBUDDY_BASE_URL");
+        httpRewrites.push({ key: "CODEBUDDY_BASE_URL", realUpstream: real });
+      }
+    } catch {
+    }
+    for (const rawModelUrl of config.codebuddy?.modelUrls ?? []) {
+      try {
+        const url = new URL(unwrapUpstream2(rawModelUrl));
+        if (url.protocol !== "https:") continue;
+        const host = url.hostname;
+        if (host && !httpsSeen.has(host.toLowerCase())) {
+          httpsSeen.add(host.toLowerCase());
+          httpsDomains.push(host);
+        }
+      } catch {
+      }
+    }
   } else if (client === "pi") {
     for (const [name, prov] of Object.entries(config.pi?.providers ?? {})) {
       classify(prov.baseUrl, name);
@@ -65450,6 +65740,21 @@ function buildClaudeEnv(origin, caPath, httpRewrites, httpsRewrites, baseEnv) {
   if (hr2) env.ANTHROPIC_BASE_URL = hr2.realUpstream;
   return env;
 }
+async function resolveCodebuddyBudgetEnv(opts) {
+  const { model, userAutoCompactWindow, shellAutoCompactWindow, routes, upstreamUrl } = opts;
+  if (!model || userAutoCompactWindow !== void 0 || nonEmpty2(shellAutoCompactWindow)) return {};
+  const window2 = await resolveLauncherWindow(model, routes, upstreamUrl);
+  if (!window2) return {};
+  return { CODEBUDDY_AUTO_COMPACT_WINDOW: String(window2) };
+}
+function buildCodebuddyEnv(origin, caPath, httpRewrites, httpsRewrites, baseEnv) {
+  const env = { ...baseEnv, HTTPS_PROXY: origin, NODE_EXTRA_CA_CERTS: caPath, BILLION_CONTEXT_PROXY: origin };
+  const r = httpRewrites.find((rw) => rw.key === "CODEBUDDY_BASE_URL");
+  if (r) env.CODEBUDDY_BASE_URL = wrapUpstream(origin, r.realUpstream);
+  const hr2 = httpsRewrites.find((rw) => rw.key === "CODEBUDDY_BASE_URL");
+  if (hr2) env.CODEBUDDY_BASE_URL = hr2.realUpstream;
+  return env;
+}
 function launcherDirectUrl(env) {
   return env.BILI_LAUNCHER_DIRECT === "1";
 }
@@ -65490,7 +65795,7 @@ function isPrivateIPv4(host) {
   return false;
 }
 function launcherInjectMcp(env, base, codexUpstream) {
-  if (base === "pi" || base === "omp" || base === "opencode" || base === "hermes" || base === "dsh") return false;
+  if (base === "pi" || base === "omp" || base === "opencode" || base === "hermes" || base === "dsh" || base === "codebuddy") return false;
   if (env.BILI_LAUNCHER_PLUGIN === "0") return false;
   if (base === "codex" && env.BILI_LAUNCHER_PLUGIN === void 0 && codexUpstream !== void 0 && isPrivateUpstreamHost(codexUpstream)) {
     return false;
@@ -66313,6 +66618,10 @@ function resolveClientCommand(client, env) {
     );
     return { command: process.execPath, prefixArgs: [cli] };
   }
+  if (client === "codebuddy") {
+    const resolved2 = resolveOnPath("codebuddy", env) ?? resolveOnPath("cbc", env);
+    return { command: resolved2 ?? "codebuddy", prefixArgs: [] };
+  }
   const resolved = resolveOnPath(client, env);
   return { command: resolved ?? client, prefixArgs: [] };
 }
@@ -66478,6 +66787,19 @@ async function runLaunch(params, deps = {}) {
       Object.assign(env, inj.envPatch);
       if (inj.warning) console.error(`bili: ${inj.warning}`);
     }
+  } else if (base === "codebuddy") {
+    env = buildCodebuddyEnv(origin, ca, routes.httpRewrites, routes.httpsRewrites, process.env);
+    const codebuddyBudget = await resolveCodebuddyBudgetEnv({
+      model: config.codebuddy?.model,
+      userAutoCompactWindow: config.codebuddy?.autoCompactWindow,
+      shellAutoCompactWindow: process.env.CODEBUDDY_AUTO_COMPACT_WINDOW,
+      routes: biliRoutes,
+      upstreamUrl: config.codebuddy?.codebuddyBaseUrl ?? "https://tencent.sso.codebuddy.cn/v2"
+    });
+    Object.assign(env, codebuddyBudget);
+    if (codebuddyBudget.CODEBUDDY_AUTO_COMPACT_WINDOW !== void 0) {
+      console.error(`bili: codebuddy budget aligned \u2014 CODEBUDDY_AUTO_COMPACT_WINDOW=${codebuddyBudget.CODEBUDDY_AUTO_COMPACT_WINDOW}`);
+    }
   } else {
     env = directUrl ? buildClaudePluginEnv(origin, true, process.env) : buildClaudeEnv(origin, ca, routes.httpRewrites, routes.httpsRewrites, process.env);
     if (directUrl) env.BILLION_CONTEXT_PROXY = origin;
@@ -66732,6 +67054,7 @@ Usage:
   bili opencode [opts --] [args]   start a proxy + launch opencode against it (cert-MITM)
   bili hermes [opts --] [args]     start a proxy + launch hermes-agent against it (/bili/ rewrite)
   bili dsh [opts --] [args]        start a proxy + launch deepseek-harness against it (/bili/ rewrite)
+  bili codebuddy [opts --] [args]  start a proxy + launch codebuddy against it (/bili/ rewrite)
   bili test pi                     non-polluting pi smoke test through the proxy
   bili export [session] [--full]   list sessions / export one as a Markdown handoff
                                     (--full includes original messages; --output FILE)
@@ -66746,7 +67069,7 @@ Usage:
   bili --version                   print version
   bili --help                      show this help
 
-Launcher (bili pi / bili codex / bili claude / bili omp / bili opencode / bili hermes / bili dsh):
+Launcher (bili pi / bili codex / bili claude / bili omp / bili opencode / bili hermes / bili dsh / bili codebuddy):
   Brings up a proxy on an independent port (a fresh instance every launch), then runs the client pointed at it via HTTPS_PROXY + the proxy's
   MITM CA \u2014 no config-file edits. Discovered HTTPS upstream domains are
   auto-whitelisted for MITM so the proxy TLS-terminates exactly the hosts the
@@ -66762,6 +67085,7 @@ Launcher (bili pi / bili codex / bili claude / bili omp / bili opencode / bili h
     bili omp                              # launch omp through the proxy (pi-based; /bili/ rewrite)
     bili hermes                           # launch hermes-agent through the proxy (/bili/ rewrite of ~/.hermes/config.yaml)
     bili dsh --profile web "task"         # launch deepseek-harness through the proxy (/bili/ rewrite of ~/.dsh/settings.yaml)
+    bili codebuddy                        # launch codebuddy through the proxy (CODEBUDDY_BASE_URL /bili/ rewrite)
     bili test pi                          # quick end-to-end check of the pi path
     bili --mitm-domain api.foo.com pi     # add a domain to the MITM whitelist (flags precede the client)
     bili -F http://127.0.0.1:7897 codex   # route bili's upstream through a proxy (gost-style -F)