billion-context 0.1.82 → 0.1.83-pr.425.342

package/dist/index.js

@@ -48360,33 +48360,64 @@ async function contextFromRegistry(model, host) {
 function peekRegistryContext(model, host) {
   return registryLookup(cache2, model, host);
 }
+var VARIANT_SUFFIXES = [
+  "-interleaved-thinking",
+  "-thinking",
+  "-reasoning",
+  "-extended",
+  "-fast",
+  "-high",
+  "-medium",
+  "-low"
+];
+function modelVariants(name) {
+  const variants = [name];
+  let current = name;
+  for (; ; ) {
+    let stripped = false;
+    for (const suffix of VARIANT_SUFFIXES) {
+      if (current.length > suffix.length && current.endsWith(suffix)) {
+        current = current.slice(0, -suffix.length);
+        variants.push(current);
+        stripped = true;
+        break;
+      }
+    }
+    if (!stripped) break;
+  }
+  return variants;
+}
 function registryLookup(reg, model, host) {
   if (!reg || !model) return void 0;
   const provider = host ? providerFromHost(host) : void 0;
-  const candidates = provider ? [`${provider}/${model}`, model] : [model];
-  for (const key of candidates) {
-    const entry = reg[key];
-    const ctx = entry?.limit?.context;
-    if (typeof ctx === "number" && ctx > 0) return ctx;
-  }
-  if (provider === void 0) {
-    const suffix = `/${model}`;
-    let max;
-    const distinct = /* @__PURE__ */ new Set();
-    const parts = [];
-    for (const key of Object.keys(reg)) {
-      if (!key.endsWith(suffix)) continue;
-      const ctx = reg[key].limit?.context;
-      if (typeof ctx !== "number" || ctx <= 0) continue;
-      if (max === void 0 || ctx > max) max = ctx;
-      distinct.add(ctx);
-      parts.push(`${key}=${ctx}`);
-    }
-    if (max !== void 0 && distinct.size > 1 && !warnedConflicts.has(model)) {
-      warnedConflicts.add(model);
-      log("warn", `[acp-registry] conflicting context windows for "${model}" (${parts.join(", ")}) \u2014 using max ${max}`);
+  for (const name of modelVariants(model)) {
+    const candidates = provider ? [`${provider}/${name}`, name] : [name];
+    for (const key of candidates) {
+      const entry = reg[key];
+      const ctx = entry?.limit?.context;
+      if (typeof ctx === "number" && ctx > 0) return ctx;
+    }
+    if (provider === void 0) {
+      const suffix = `/${name}`;
+      let max;
+      const distinct = /* @__PURE__ */ new Set();
+      const parts = [];
+      for (const key of Object.keys(reg)) {
+        if (!key.endsWith(suffix)) continue;
+        const ctx = reg[key].limit?.context;
+        if (typeof ctx !== "number" || ctx <= 0) continue;
+        if (max === void 0 || ctx > max) max = ctx;
+        distinct.add(ctx);
+        parts.push(`${key}=${ctx}`);
+      }
+      if (max !== void 0) {
+        if (distinct.size > 1 && !warnedConflicts.has(name)) {
+          warnedConflicts.add(name);
+          log("warn", `[acp-registry] conflicting context windows for "${name}" (${parts.join(", ")}) \u2014 using max ${max}`);
+        }
+        return max;
+      }
     }
-    return max;
   }
   return void 0;
 }
@@ -49829,8 +49860,17 @@ var SessionStore = class {
   async loadAll() {
     const out = /* @__PURE__ */ new Map();
     if (!this.enabled) return out;
+    let clamped = 0;
     for (const [id, envelope] of await this.store.loadAll()) {
-      out.set(id, buildSession(envelope.payload));
+      const session = buildSession(envelope.payload);
+      if (hasNegativePersistedTokens(envelope.payload)) {
+        await this.store.writeNow(id, () => buildRecord(session));
+        clamped++;
+      }
+      out.set(id, session);
+    }
+    if (clamped > 0) {
+      log("info", `[persist] one-time migration (#408): clamped negative lastInputTokens/contextTokens in ${clamped} session(s) to 0`);
     }
     return out;
   }
@@ -49911,7 +49951,14 @@ var SessionStore = class {
       meta?.protocol ? this.store.loadSync(id, relPathFor(id)) : null
     ];
     for (const envelope of envelopes) {
-      if (envelope) return buildSession(envelope.payload);
+      if (envelope) {
+        const session = buildSession(envelope.payload);
+        if (hasNegativePersistedTokens(envelope.payload)) {
+          this.scheduleSave(session);
+          log("info", `[persist] clamped negative token stats on reload for ${id} (#408)`);
+        }
+        return session;
+      }
     }
     return null;
   }
@@ -49988,10 +50035,14 @@ function buildSession(parsed) {
       cachedTokens: stats.cachedTokens ?? parsed.cachedTokens ?? 0,
       outputTokens: stats.outputTokens ?? parsed.outputTokens ?? 0,
       cacheSamples: stats.cacheSamples ?? parsed.cacheSamples ?? 0,
-      lastInputTokens: stats.lastInputTokens ?? parsed.lastInputTokens ?? 0,
+      // #408: clamp at restore — pre-clamp versions persisted negative
+      // values (lastInputTokens = total − credit before the Math.max
+      // guard existed) which would otherwise revive after upgrade and
+      // feed the /acp panel + web stats as negative percentages.
+      lastInputTokens: Math.max(0, stats.lastInputTokens ?? parsed.lastInputTokens ?? 0),
       // In-memory only — a fresh process has no pending compress fold.
       compressCreditTokens: 0,
-      contextTokens: stats.contextTokens ?? parsed.contextTokens ?? 0
+      contextTokens: Math.max(0, stats.contextTokens ?? parsed.contextTokens ?? 0)
     },
     metadata: parsed.metadata ?? {},
     state: mergeState(parsed.state),
@@ -50014,6 +50065,12 @@ function isValidRecord(parsed) {
   const r = parsed;
   return typeof r.id === "string" && typeof r.state === "object" && r.state !== null && Array.isArray(r.state.blocks);
 }
+function hasNegativePersistedTokens(parsed) {
+  const stats = parsed.stats ?? {};
+  const last = stats.lastInputTokens ?? parsed.lastInputTokens;
+  const ctx = stats.contextTokens ?? parsed.contextTokens;
+  return typeof last === "number" && last < 0 || typeof ctx === "number" && ctx < 0;
+}
 function defaultDir() {
   return sessionsDir();
 }
@@ -50147,6 +50204,8 @@ function resetSessionCompression(session) {
   session.blockContents.clear();
   session.stats.lastInputTokens = 0;
   session.stats.contextTokens = 0;
+  session.hostCreditTokens = 0;
+  session.hostContextTokens = 0;
   session.metadata.nativeCompactionAt = Date.now();
   markDirty(session);
 }
@@ -51736,6 +51795,7 @@ function recordUsage(ctx, usage, round) {
   const total = (typeof prompt === "number" ? prompt : 0) + (!includesCached && typeof cached === "number" ? cached : 0);
   if (total > 0) ctx.session.stats.inputTokens += total;
   ctx.session.stats.lastInputTokens = Math.max(0, total - (ctx.session.stats.compressCreditTokens ?? 0));
+  ctx.session.hostContextTokens = total + (ctx.session.hostCreditTokens ?? 0);
   if (typeof cached === "number") {
     ctx.session.stats.cachedTokens += cached;
     ctx.session.stats.cacheSamples += 1;
@@ -51873,6 +51933,10 @@ async function* runCompressLoop(upstream, ctx, requestBody, requestOptions, adap
       if (usage.inputTokens !== void 0 || usage.outputTokens !== void 0 || usage.cachedTokens !== void 0) {
         recordUsage(ctx, usage, round);
       }
+      const hostCredit = ctx.session.hostCreditTokens ?? 0;
+      if (hostCredit > 0 && typeof usage.inputTokens === "number") {
+        usage.inputTokens += hostCredit;
+      }
       let resolvedText = assistantText;
       let allCalls = calls;
       if (ctx.textProtocol && assistantText.length > 0 && adapter.extractTextTriggers) {
@@ -52117,6 +52181,23 @@ function usageTotals(protocol, usage) {
     cached: num3(usage["input_tokens_details"]?.["cached_tokens"])
   };
 }
+function backfillHostUsage(protocol, usage, credit) {
+  if (!Number.isFinite(credit) || credit <= 0) return false;
+  let patched = false;
+  const add = (key) => {
+    if (typeof usage[key] === "number" && Number.isFinite(usage[key])) {
+      usage[key] = usage[key] + credit;
+      patched = true;
+    }
+  };
+  if (protocol === "openai") {
+    add("prompt_tokens");
+    add("total_tokens");
+  } else {
+    add("input_tokens");
+  }
+  return patched;
+}
 var CONTEXT_OVERFLOW_PATTERNS = [
   /context_length_exceeded/i,
   /context_window_exceeded/i,
@@ -52805,7 +52886,7 @@ function stripFinishReasonChunk(buf) {
     return buf;
   }
 }
-function createOpenaiAdapter(requestBody, clientSystem) {
+function createOpenaiAdapter(requestBody, clientSystem, hostCredit = 0) {
   const model = requestBody.model ?? "unknown";
   let responseId = `chatcmpl-proxy-${Date.now()}`;
   let toolIndex = 0;
@@ -52999,7 +53080,24 @@ function createOpenaiAdapter(requestBody, clientSystem) {
             cachedTokens: typeof pd?.cached_tokens === "number" ? pd.cached_tokens : void 0
           };
           if (sawRealToolCall) {
-            yield { kind: "meta", chunk: rawBuf };
+            let chunk = rawBuf;
+            if (hostCredit > 0 && u2) {
+              const pu = typeof u2.prompt_tokens === "number" ? u2.prompt_tokens : void 0;
+              const tu = typeof u2.total_tokens === "number" ? u2.total_tokens : void 0;
+              if (pu !== void 0 || tu !== void 0) {
+                const patched = {
+                  ...parsed,
+                  usage: {
+                    ...u2,
+                    ...pu !== void 0 ? { prompt_tokens: pu + hostCredit } : {},
+                    ...tu !== void 0 ? { total_tokens: tu + hostCredit } : {}
+                  }
+                };
+                const out = eventStr.split("\n").map((l) => l.startsWith("data:") ? `data: ${JSON.stringify(patched)}` : l).join("\n");
+                chunk = Buffer.from(out + "\n\n", "utf8");
+              }
+            }
+            yield { kind: "meta", chunk };
             yield { kind: "done", finishReason, suppressCompletion: true };
             continue;
           } else {
@@ -53185,7 +53283,7 @@ data: ${JSON.stringify({ type: "content_block_delta", index, delta: { type: "tex
     "utf8"
   );
 }
-function createAnthropicAdapter(requestBody, originalSystem) {
+function createAnthropicAdapter(requestBody, originalSystem, hostCredit = 0) {
   const model = requestBody.model ?? void 0;
   let messageId;
   let clientIndex = 0;
@@ -53299,8 +53397,19 @@ ${systemPrompt}` : systemPrompt;
           if (typeof u2.input_tokens === "number") roundInput = u2.input_tokens;
           if (typeof u2.cache_read_input_tokens === "number") roundCached = u2.cache_read_input_tokens;
           if (round === 1) {
+            let chunk = rawBuf;
+            if (hostCredit > 0 && typeof u2.input_tokens === "number") {
+              const patched = structuredClone(data);
+              const pmsg = patched["message"];
+              const pu = pmsg?.["usage"] ?? {};
+              if (typeof pu.input_tokens === "number") {
+                pu.input_tokens += hostCredit;
+                const out = eventStr.split("\n").map((l) => l.startsWith("data:") ? `data: ${JSON.stringify(patched)}` : l).join("\n");
+                chunk = Buffer.from(out + "\n\n", "utf8");
+              }
+            }
             messageStartForwarded = true;
-            yield { kind: "meta", chunk: rawBuf, firstRoundOnly: true };
+            yield { kind: "meta", chunk, firstRoundOnly: true };
           }
         } else if (type === "ping") {
           yield { kind: "meta", chunk: rawBuf };
@@ -53483,10 +53592,10 @@ data: ${JSON.stringify({ type: "content_block_stop", index })}
 }
 
 // src/loop/index.ts
-function pickAdapter(protocol, requestBody, textProtocol, responsesProjection, anthropicSystem, openaiSystem) {
+function pickAdapter(protocol, requestBody, textProtocol, responsesProjection, anthropicSystem, openaiSystem, hostCredit = 0) {
   if (protocol === "responses") return createResponsesAdapter(textProtocol, responsesProjection);
-  if (protocol === "openai") return createOpenaiAdapter(requestBody, openaiSystem);
-  if (protocol === "anthropic") return createAnthropicAdapter(requestBody, anthropicSystem);
+  if (protocol === "openai") return createOpenaiAdapter(requestBody, openaiSystem, hostCredit);
+  if (protocol === "anthropic") return createAnthropicAdapter(requestBody, anthropicSystem, hostCredit);
   throw new Error(`[acp-loop] unknown protocol: ${protocol}`);
 }
 
@@ -54705,12 +54814,12 @@ function handlePluginStatus(conversationId2, res, fallbackLatest = false) {
   if (entry) entry.lastSeen = Date.now();
   const limit = session.metadata.effectiveContextLimit;
   const mem = remembered.get(session.id);
-  const modelContextLimit = typeof limit === "number" && limit > 0 ? limit : 2e5;
+  const modelContextLimit = typeof limit === "number" && limit > 0 ? limit : 0;
   let panel;
   try {
     panel = buildStatusPanel({
       version: `billion-context@${PROXY_VERSION}`,
-      tokenCount: session.stats.lastInputTokens,
+      tokenCount: session.hostContextTokens ?? session.stats.lastInputTokens,
       systemPromptTokens: 0,
       state: session.state,
       nudge: mem?.nudge,
@@ -54728,7 +54837,8 @@ function handlePluginStatus(conversationId2, res, fallbackLatest = false) {
     label: session.meta.label ?? null,
     pluginAgent: session.metadata.pluginAgent ?? null,
     contextLimit: typeof limit === "number" ? limit : null,
-    contextTokens: session.stats.lastInputTokens,
+    contextTokens: session.hostContextTokens ?? session.stats.lastInputTokens,
+    hostCredit: session.hostCreditTokens ?? 0,
     inputTokens: session.stats.inputTokens,
     outputTokens: session.stats.outputTokens,
     cachedTokens: session.stats.cachedTokens,
@@ -54845,6 +54955,7 @@ function applyUsageSample(session, sample, protocol) {
     session.stats.inputTokens += total;
     session.stats.lastInputTokens = Math.max(0, total - (session.stats.compressCreditTokens ?? 0));
     warnCacheCollapse(session, total, sample.cachedTokens ?? 0);
+    session.hostContextTokens = total + (session.hostCreditTokens ?? 0);
   }
   if (sample.outputTokens !== void 0) session.stats.outputTokens += sample.outputTokens;
 }
@@ -54858,6 +54969,7 @@ async function pipePluginChatWithStrip(stream2, res, protocol, session, log2) {
   const decoder = new TextDecoder("utf-8");
   let buf = "";
   const acc = {};
+  const credit = session?.hostCreditTokens ?? 0;
   const onDrop = (snippet) => {
     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`);
@@ -55015,12 +55127,23 @@ data: ${JSON.stringify({ type: "content_block_delta", index, delta: { type: delt
           if (ev["type"] === "message_stop") sawTerminal = true;
           const sample = usageFromSseEvent(ev);
           if (sample) mergeUsageSample(acc, sample);
+          let backfilled = false;
+          if (credit > 0 && protocol) {
+            const usage = ev["type"] === "message_start" ? ev["message"]?.["usage"] : ev["usage"];
+            const deltaEcho = ev["type"] === "message_delta" && (num2(usage?.["input_tokens"]) ?? 0) <= 0;
+            if (usage && !deltaEcho && backfillHostUsage(protocol, usage, credit)) backfilled = true;
+          }
           const out = protocol === "anthropic" ? processAnthropic(ev, rawEvent) : processOpenai(ev, rawEvent);
-          if (out.length > 0) await write(out);
+          if (backfilled && out === rawEvent + "\n\n") {
+            await write(rebuildEvent(rawEvent, ev));
+          } else if (out.length > 0) {
+            await write(out);
+          }
         }
       }
       if (res.destroyed || res.writableEnded) break;
     }
+    if (buf.length > 0 && !res.destroyed && !res.writableEnded) await write(buf);
     const rest = flushTails();
     if (rest.length > 0 && !res.destroyed && !res.writableEnded) await write(rest);
     settleUsage();
@@ -55065,6 +55188,7 @@ async function pipePluginResponsesWithStrip(stream2, res, session, log2) {
     if (!res.write(Buffer.from(s3, "utf8"))) {
       return new Promise((r) => res.once("drain", () => r()));
     }
+    return Promise.resolve();
   };
   const settleUsage = () => {
     if (session && (acc.inputTokens !== void 0 || acc.outputTokens !== void 0 || acc.cachedTokens !== void 0)) {
@@ -55121,7 +55245,17 @@ async function pipePluginResponsesWithStrip(stream2, res, session, log2) {
           const type = ev["type"];
           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;
-            const out = containsRenderTagText(jsonStr) ? rebuildEvent(rawEvent, stripResponsesText(ev)) : rawEvent + "\n\n";
+            let evOut = ev;
+            let rebuild = containsRenderTagText(jsonStr);
+            if (rebuild) evOut = stripResponsesText(ev);
+            if (type === "response.completed") {
+              const credit = session?.hostCreditTokens ?? 0;
+              const usage = evOut["response"]?.["usage"];
+              if (credit > 0 && usage && backfillHostUsage("responses", usage, credit)) {
+                rebuild = true;
+              }
+            }
+            const out = rebuild ? rebuildEvent(rawEvent, evOut) : rawEvent + "\n\n";
             await write(flushTail(out));
             continue;
           }
@@ -55202,8 +55336,10 @@ async function pipePluginJson(stream2, res, session, protocol) {
   }
   reader.releaseLock();
   const text = Buffer.concat(chunks).toString("utf8");
+  let json;
+  let mutated = false;
   try {
-    const json = JSON.parse(text);
+    json = JSON.parse(text);
     const usage = json["usage"];
     if (session && usage) {
       const input = num2(usage["prompt_tokens"]) ?? num2(usage["input_tokens"]);
@@ -55214,18 +55350,21 @@ async function pipePluginJson(stream2, res, session, protocol) {
           cachedTokens: num2(usage["prompt_tokens_details"]?.["cached_tokens"]) ?? num2(usage["input_tokens_details"]?.["cached_tokens"]) ?? num2(usage["cache_read_input_tokens"])
         }, protocol);
         markDirty(session);
+        const credit = session.hostCreditTokens ?? 0;
+        if (credit > 0 && protocol && backfillHostUsage(protocol, usage, credit)) {
+          mutated = true;
+        }
       }
     }
   } catch {
   }
-  if (containsRenderTagText(text)) {
-    try {
-      const json = JSON.parse(text);
-      const stripped = protocol === "responses" ? stripResponsesText(json) : protocol === "anthropic" ? stripAnthropicText(json) : stripOpenaiChatText(json);
-      res.end(Buffer.from(JSON.stringify(stripped), "utf8"));
-      return;
-    } catch {
-    }
+  if (json && containsRenderTagText(text)) {
+    json = protocol === "responses" ? stripResponsesText(json) : protocol === "anthropic" ? stripAnthropicText(json) : stripOpenaiChatText(json);
+    mutated = true;
+  }
+  if (mutated && json) {
+    res.end(Buffer.from(JSON.stringify(json), "utf8"));
+    return;
   }
   res.end(text);
 }
@@ -56729,6 +56868,9 @@ function logUnrecognizedPath(log2, url) {
     log2("info", `unrecognized path ${key}: forwarding unchanged; further occurrences suppressed`);
   }
 }
+function isModelDiscoveryPath(path18) {
+  return path18.replace(/\/+$/, "").endsWith("/models");
+}
 var BILI_HOP_HEADER = "x-bili-hop";
 function parseLauncherModelWindows(raw) {
   if (!raw) return {};
@@ -56930,7 +57072,7 @@ Content-Length: ${Buffer.byteLength(body)}\r
       );
     }
     if (opts.debug) {
-      log2("info", `[debug] build features: raw-HTTP-capture(on) | remote_compaction_v2-strip(on) | cert-MITM-launcher(on) | strip-acp-summary(on) \u2014 seeing this line confirms the launcher build (not registry 0.1.34)`);
+      log2("info", `[debug] build features: raw-HTTP-capture(${bodyDumpEnabled() ? "on" : "off"}) | remote_compaction_v2-strip(on) | cert-MITM-launcher(on) | strip-acp-summary(on) \u2014 seeing this line confirms the launcher build (not registry 0.1.34)`);
     }
   };
   const attemptListen = (port) => {
@@ -56998,6 +57140,13 @@ Content-Length: ${Buffer.byteLength(body)}\r
   }
   return server;
 }
+var SIDE_REQUEST_MAX_TOKENS = 200;
+function isSideRequest(parsed) {
+  if (!parsed || typeof parsed !== "object") return false;
+  const p2 = parsed;
+  const raw = p2.max_tokens ?? p2.max_completion_tokens ?? p2.max_output_tokens;
+  return typeof raw === "number" && raw > 0 && raw <= SIDE_REQUEST_MAX_TOKENS;
+}
 function isTrustedAdminOrigin(origin, host, trustedHosts) {
   if (!host || !trustedHosts.has(host.toLowerCase())) return false;
   if (!origin) return true;
@@ -57274,11 +57423,11 @@ ${bodyBuffer.toString("utf8")}`);
       const pluginWindow = pluginReportedContextWindow(req.headers);
       const launcherWindow = launcherContextWindow(model);
       const configuredWindow = resolveConfiguredContextLimit(opts.routes, embeddedUrl, model);
-      const peekWindow = host ? peekRegistryContext(model, host) : void 0;
+      const peekWindow = peekRegistryContext(model, host);
       let native = betaWindow ?? pluginWindow ?? launcherWindow ?? configuredWindow ?? peekWindow ?? lookupContextLimit(model);
       const operatorWindowTuned = resolveCompress(opts.routes, embeddedUrl, model, opts.compress).modelContextLimit !== void 0;
       nativeFromFallback = !betaWindow && !pluginWindow && !launcherWindow && !peekWindow && !configuredWindow && !operatorWindowTuned;
-      if (!native && host) {
+      if (!native) {
         native = await contextFromRegistry(model, host);
         if (native) nativeFromFallback = false;
       }
@@ -57379,7 +57528,6 @@ ${bodyBuffer.toString("utf8")}`);
     if (pluginAgent && !pluginConversation) pluginConversation = conversation;
     if (pluginAgent) {
       if (session.metadata.pluginAgent !== pluginAgent) session.metadata.pluginAgent = pluginAgent;
-      session.metadata.effectiveContextLimit = reqConfig.modelContextLimit;
       recordPluginSession(pluginConversation ?? conversation, session.id);
     }
     if (protocol === "responses" || protocol === "openai" || protocol === "anthropic") {
@@ -57389,6 +57537,21 @@ ${bodyBuffer.toString("utf8")}`);
       }
     }
     const pluginMode = pluginAgent !== void 0;
+    if (!countTokens && !responsesCompact && protocol !== null && isSideRequest(parsed)) {
+      log2("info", `[${session.id}] side request (max_tokens<=${SIDE_REQUEST_MAX_TOKENS}) \u2192 passthrough + tag strip only, kernel state untouched`);
+      const sidePrepared = {
+        body: bodyBuffer,
+        session,
+        processedMessages: [],
+        originalMessages: [],
+        protocol,
+        stream: parsed.stream === true,
+        compressInjected: false,
+        sidePassthrough: true
+      };
+      await forward(req, res, opts, bodyBuffer, sidePrepared, core, reqConfig, log2, route, instanceId, affinity);
+      return;
+    }
     const reqModel = parsed.model;
     const learnedMap = session.metadata.learnedContextLimits;
     const learnedLimit = (reqModel && learnedMap ? learnedMap[reqModel] : void 0) ?? session.metadata.learnedContextLimit;
@@ -57397,6 +57560,22 @@ ${bodyBuffer.toString("utf8")}`);
       reqConfig = { ...reqConfig, modelContextLimit: learnedLimit };
       nativeFromFallback = false;
       log2("info", `[${session.id}] self-healed context window: ${resolved} \u2192 ${learnedLimit} (learned from an upstream overflow)`);
+    } else if (nativeFromFallback && reqModel) {
+      const prevInput = session.stats.lastInputTokens ?? 0;
+      const prevWindow = session.metadata.lastTurnWindow;
+      const resolved = reqConfig.modelContextLimit;
+      if (prevWindow !== void 0 && prevInput > prevWindow && prevInput > resolved && prevInput >= 1e3) {
+        const map = session.metadata.learnedContextLimits ?? {};
+        const prev = map[reqModel];
+        if (prev === void 0 || prevInput > prev) {
+          map[reqModel] = prevInput;
+          session.metadata.learnedContextLimits = map;
+          markDirty(session);
+        }
+        reqConfig = { ...reqConfig, modelContextLimit: prevInput };
+        nativeFromFallback = false;
+        log2("info", `[${session.id}] self-healed context window upward: ${resolved} \u2192 ${prevInput} (a prior turn used ${prevInput} input tokens, exceeding the fallback window)`);
+      }
     }
     const nativeWindow = reqConfig.modelContextLimit;
     if (shouldReserveOutputHeadroom(protocol)) {
@@ -57410,6 +57589,8 @@ ${bodyBuffer.toString("utf8")}`);
       }
       if (reserved !== reqConfig.modelContextLimit) reqConfig = { ...reqConfig, modelContextLimit: reserved };
     }
+    session.metadata.effectiveContextLimit = reqConfig.modelContextLimit;
+    session.metadata.lastTurnWindow = reqConfig.modelContextLimit;
     acquireInFlight(session);
     try {
       await withSessionLock(session, async () => {
@@ -57474,7 +57655,7 @@ ${bodyBuffer.toString("utf8")}`);
     }
   }
   if (!prepared) {
-    if (protocol === null && !opts.passthrough) {
+    if (protocol === null && !opts.passthrough && !isModelDiscoveryPath(urlPath)) {
       logUnrecognizedPath(log2, req.url ?? "");
     }
     await forward(req, res, opts, bodyBuffer, null, core, reqConfig, log2, route, instanceId, void 0);
@@ -57520,6 +57701,7 @@ function prepareAnthropic(parsed, req, opts, core, config, prompts, log2, sessio
   const sessionId = session.id;
   const stream2 = parsed.stream === true;
   ++session.stats.requests;
+  session.hostCreditTokens = 0;
   const injectTools = opts.compress.injectTool && !pluginMode;
   if (isAutoModeClassifier(parsed)) {
     log2("info", `[${sessionId}] auto-mode classifier passthrough (skipping compress injection)`);
@@ -57575,6 +57757,10 @@ function prepareAnthropic(parsed, req, opts, core, config, prompts, log2, sessio
   markDirty(session);
   const rebuilt = { ...parsed, messages: rebuiltMessages, system: systemOut, tools: toolsOut };
   delete rebuilt.prompt_cache_key;
+  session.hostCreditTokens = processedMessages.length > 0 ? Math.max(0, estimateCoreMessages(originalMessages) - estimateCoreMessages(processedMessages)) : 0;
+  if (session.hostCreditTokens > 0) {
+    log2("info", `[${sessionId}] host usage backfill armed: +${session.hostCreditTokens} tok (forwarded view is folded); host usage will report the uncompressed baseline`);
+  }
   return { body: JSON.stringify(rebuilt), session, processedMessages, originalMessages, anthropicSystem: parsed.system, protocol: "anthropic", stream: stream2, compressInjected: injectTools, pluginMode, nudge, prompts, renderTags: "text-only" };
 }
 var OUTPUT_CLAMP_MARGIN_PCT = 0.05;
@@ -57610,6 +57796,7 @@ function prepareOpenai(parsed, req, opts, core, config, prompts, log2, session,
   const sessionId = session.id;
   const stream2 = parsed.stream === true;
   ++session.stats.requests;
+  session.hostCreditTokens = 0;
   let openaiSystemText = "";
   let processedMessages = [];
   let originalMessages = [];
@@ -57669,6 +57856,10 @@ function prepareOpenai(parsed, req, opts, core, config, prompts, log2, session,
   if (stream2 && rebuilt.stream_options === void 0) {
     rebuilt.stream_options = { include_usage: true };
   }
+  session.hostCreditTokens = processedMessages.length > 0 ? Math.max(0, estimateCoreMessages(originalMessages) - estimateCoreMessages(processedMessages)) : 0;
+  if (session.hostCreditTokens > 0) {
+    log2("info", `[${sessionId}] host usage backfill armed: +${session.hostCreditTokens} tok (forwarded view is folded); host usage will report the uncompressed baseline`);
+  }
   snapshotMessages(session, originalMessages);
   markDirty(session);
   return { body: JSON.stringify(rebuilt), session, processedMessages, originalMessages, protocol: "openai", stream: stream2, compressInjected: injectTools, pluginMode, nudge, prompts, openaiSystemText, renderTags: "text-only" };
@@ -57677,6 +57868,7 @@ function prepareResponses(parsed, req, opts, core, config, prompts, log2, sessio
   const sessionId = session.id;
   const stream2 = parsed.stream === true;
   ++session.stats.requests;
+  session.hostCreditTokens = 0;
   if (reconcileNativeCompactionBoundary(session)) {
     log2("info", `[${sessionId}] reconciled ACP state after native Responses compact boundary`);
   }
@@ -57802,6 +57994,10 @@ function prepareResponses(parsed, req, opts, core, config, prompts, log2, sessio
     });
     log2("info", `[${sessionId}] responses forward tools=[${fwdTools.join(",")}] injectTool=${injectTools}${pluginMode ? " (plugin mode: wire injection suppressed)" : ""} NO_INJECT_TOOL=${!!process.env.ACP_NO_INJECT_TOOL} NO_COMPRESS_PROMPT=${!!process.env.ACP_NO_COMPRESS_PROMPT}`);
   }
+  session.hostCreditTokens = processedMessages.length > 0 ? Math.max(0, estimateCoreMessages(originalMessages) - estimateCoreMessages(processedMessages)) : 0;
+  if (session.hostCreditTokens > 0) {
+    log2("info", `[${sessionId}] host usage backfill armed: +${session.hostCreditTokens} tok (forwarded view is folded); host usage will report the uncompressed baseline`);
+  }
   snapshotMessages(session, originalMessages);
   markDirty(session);
   return {
@@ -58311,7 +58507,7 @@ ${hdrText}
   }
   let responseBody = upstream.body;
   try {
-    if (prepared !== null && prepared.stream && maxFakeCompletionRetries() > 0) {
+    if (prepared !== null && prepared.stream && !prepared.sidePassthrough && maxFakeCompletionRetries() > 0) {
       const resolvedBuf = await resolveFakeCompletion(upstream.body, {
         protocol: prepared.protocol,
         body,
@@ -58378,7 +58574,7 @@ ${hdrText}
         const reqHeaders = buildForwardHeaders(headers);
         const textProtocol = prepared.protocol === "responses" && !!prepared.responsesTextProtocol;
         const systemPrompt = textProtocol ? buildCompressHybridSystemPrompt(prepared.prompts ?? defaultPrompts) : buildCompressSystemPrompt(prepared.prompts ?? defaultPrompts);
-        const adapter = pickAdapter(prepared.protocol, parsedReq, textProtocol, prepared.responsesProjection, prepared.anthropicSystem, prepared.openaiSystemText);
+        const adapter = pickAdapter(prepared.protocol, parsedReq, textProtocol, prepared.responsesProjection, prepared.anthropicSystem, prepared.openaiSystemText, prepared.session.hostCreditTokens ?? 0);
         const refreshFolded = (current) => {
           const turn = core.processTurn({
             messages: prepared.originalMessages,
@@ -58460,6 +58656,10 @@ ${hdrText}
             const out = u2.completion_tokens ?? u2.output_tokens;
             if (typeof out === "number") prepared.session.stats.outputTokens += out;
           }
+          const credit = prepared.session.hostCreditTokens ?? 0;
+          if (credit > 0 && backfillHostUsage(prepared.protocol, u2, credit)) {
+            prepared.session.hostContextTokens = (typeof total === "number" ? total : 0) + credit;
+          }
           if (prepared.protocol === "openai") {
             rewriteOpenaiJsonResponse(json, ctx);
           } else if (prepared.protocol === "responses") {