billion-context 0.1.75 → 0.1.76-pr.495.317

package/dist/index.js

@@ -45360,6 +45360,26 @@ function numericBlockId(id) {
   const parsed = /^b(\d+)$/.exec(id);
   return parsed ? Number(parsed[1]) : 0;
 }
+function refGateDiagnostics(state, requestedRanges, unknownCount) {
+  const highest = highestUsedIndex(state.messageRefs);
+  const highestRef = highest > 0 ? indexToRef(highest) : "none";
+  return `[diagnostics: session highest ref=${highestRef}, unknown ranges in request=${unknownCount}/${requestedRanges}, session history=${state.stats.compressionCount} compression(s), ${state.blocks.length} block(s)]`;
+}
+function danglingMessageRefs(state, messages, spec) {
+  const visible = new Set(messages.map((m2) => m2.id));
+  const dangling = [];
+  for (const ref of [spec.startRef, spec.endRef]) {
+    const parsed = parseBoundary(ref);
+    if (!parsed || parsed.kind !== "message") continue;
+    const rawId = state.messageRefs.byRef[parsed.raw] ?? state.messageRefs.byRef[indexToRef(parsed.numericId)];
+    if (!rawId || visible.has(rawId)) continue;
+    const covered = state.blocks.some(
+      (block) => block.active && block.effectiveMessageIds.includes(rawId)
+    );
+    if (!covered) dangling.push(parsed.raw);
+  }
+  return dangling;
+}
 function createCore(ports = {}) {
   const countTokens = ports.countTokens ?? defaultCountTokens;
   function applyCompression(input) {
@@ -45460,7 +45480,15 @@ function createCore(ports = {}) {
       if (!hasBlockBoundaryRange && totalRangeChars < input.config.compress.minCompressRange) {
         const live = activeBlocks(state).map((b2) => b2.blockId).sort((x, y) => numericBlockId(x) - numericBlockId(y));
         const liveHint = live.length > 0 ? ` Current active blocks span ${live[0]}..${live[live.length - 1]} \u2014 retry with startId/endId set to active block IDs in that span.` : "";
-        const gateMessage = resolvableCount === 0 && consumedRanges.length === 0 && unknownCount > 0 ? `None of the ${input.ranges.length} requested range(s) resolved \u2014 every ref failed with "does not exist in this session". Refs recorded before an earlier compress are stale: each successful compress renumbers the remaining refs. Run acp_status, then call the compress tool again using only the refs it reports.` : consumedRanges.length > 0 ? `Requested range(s) already compressed (e.g. ${consumedRanges[0].startRef}..${consumedRanges[0].endRef}) \u2014 your refs are stale: a prior compress renumbered the remaining refs, so this range now falls inside an active block. Run acp_status, then call the compress tool again using only the CURRENT compressible ranges it reports.${liveHint}` : `Total compressible content too small (${totalRangeChars} chars across ${countedRanges} range(s), min ${input.config.compress.minCompressRange}). Combine more messages into your range(s) to meet the threshold.`;
+        const diagnostics = refGateDiagnostics(
+          state,
+          input.ranges.length,
+          unknownCount
+        );
+        const danglingRefs = consumedRanges.flatMap(
+          (spec) => danglingMessageRefs(state, input.messages, spec)
+        );
+        const gateMessage = resolvableCount === 0 && consumedRanges.length === 0 && unknownCount > 0 ? `None of the ${input.ranges.length} requested range(s) resolved \u2014 every ref is unknown to this session. Refs are per-session snapshots, assigned once when a message is first rendered; no compress reassigns them, so unknown refs cannot come from an earlier compress in this session. They come from a different generation: a previous session instance (switching model or upstream mid-conversation starts a fresh session whose refs restart at m00001), the generation before a native-compaction rebase (which also resets refs to m00001), or a typo. ${diagnostics} Run acp_status, then call the compress tool again using only the refs it reports.` : consumedRanges.length > 0 ? danglingRefs.length > 0 ? `Requested range(s) cannot be anchored (e.g. ${consumedRanges[0].startRef}..${consumedRanges[0].endRef}) \u2014 the refs exist in this session's ref map, but the messages they point to are no longer in the visible context and no active block covers them: the message content changed (or the message was filtered out of the view) and now carries a new ref, leaving your old refs dangling. ${diagnostics} Run acp_status, then call the compress tool again using only the refs it reports.` : `Requested range(s) already compressed (e.g. ${consumedRanges[0].startRef}..${consumedRanges[0].endRef}) \u2014 those refs no longer point to directly compressible content: the range is covered by active block(s) or the block ref(s) are stale (distilled or consumed). ${diagnostics} Run acp_status, then call the compress tool again using only the CURRENT compressible ranges it reports.${liveHint}` : `Total compressible content too small (${totalRangeChars} chars across ${countedRanges} range(s), min ${input.config.compress.minCompressRange}). Combine more messages into your range(s) to meet the threshold.`;
         return {
           state: input.state,
           result: {
@@ -47856,7 +47884,7 @@ function parseCompressSettings(v2) {
     }
     out[key] = typeof obj[key] === "string" ? obj[key].trim() : obj[key];
   }
-  for (const key of ["nudgeGrowthTokens", "preserveRecentMessages", "preserveRecentTokens", "minCompressRange"]) {
+  for (const key of ["nudgeGrowthTokens", "preserveRecentMessages", "preserveRecentTokens", "minCompressRange", "minCompressRangeChars"]) {
     takeNumber(key);
   }
   if ("tiers" in obj) {
@@ -47922,6 +47950,7 @@ function resolveContextLimitValue(raw, nativeLimit) {
 function mergeCompress(global2, provider, model) {
   const pick = (k2) => model?.[k2] ?? provider?.[k2] ?? global2?.[k2];
   const promptLevels = [global2?.prompts, provider?.prompts, model?.prompts].filter(Boolean);
+  const rangeOf = (s3) => s3?.minCompressRangeChars ?? s3?.minCompressRange;
   return {
     modelContextLimit: pick("modelContextLimit"),
     maxContextLimit: pick("maxContextLimit"),
@@ -47929,7 +47958,7 @@ function mergeCompress(global2, provider, model) {
     nudgeGrowthTokens: pick("nudgeGrowthTokens"),
     preserveRecentMessages: pick("preserveRecentMessages"),
     preserveRecentTokens: pick("preserveRecentTokens"),
-    minCompressRange: pick("minCompressRange"),
+    minCompressRangeChars: rangeOf(model) ?? rangeOf(provider) ?? rangeOf(global2),
     tiers: pick("tiers"),
     prompts: promptLevels.length > 0 ? Object.assign({}, ...promptLevels) : void 0,
     acknowledgePromptsRisk: pick("acknowledgePromptsRisk")
@@ -47983,7 +48012,7 @@ function applyCompressSettings(base, limit, s3) {
     preserveRecentTokens: s3.preserveRecentTokens ?? base.preserveRecentTokens,
     compress: {
       ...base.compress,
-      minCompressRange: s3.minCompressRange ?? base.compress.minCompressRange
+      minCompressRange: s3.minCompressRangeChars ?? s3.minCompressRange ?? base.compress.minCompressRange
     }
   };
 }
@@ -48008,12 +48037,22 @@ import path3 from "path";
 // src/fetch-util.ts
 var MAX_REQUEST_BYTES = 100 * 1024 * 1024;
 var UPSTREAM_TIMEOUT_MS = 10 * 60 * 1e3;
+var liveUpstreamTimers = /* @__PURE__ */ new Set();
 async function fetchWithTimeout(url, opts, timeoutMs = UPSTREAM_TIMEOUT_MS, externalSignal) {
   const controller = new AbortController();
-  let timer2 = setTimeout(() => controller.abort(), timeoutMs);
+  const armTimer = () => {
+    const t = setTimeout(() => {
+      liveUpstreamTimers.delete(t);
+      controller.abort();
+    }, timeoutMs);
+    liveUpstreamTimers.add(t);
+    return t;
+  };
+  let timer2 = armTimer();
   const rearm = () => {
     clearTimeout(timer2);
-    timer2 = setTimeout(() => controller.abort(), timeoutMs);
+    liveUpstreamTimers.delete(timer2);
+    timer2 = armTimer();
   };
   let onExternalAbort = null;
   if (externalSignal) {
@@ -48025,6 +48064,7 @@ async function fetchWithTimeout(url, opts, timeoutMs = UPSTREAM_TIMEOUT_MS, exte
   }
   const cleanup = () => {
     clearTimeout(timer2);
+    liveUpstreamTimers.delete(timer2);
     if (onExternalAbort && externalSignal) externalSignal.removeEventListener("abort", onExternalAbort);
   };
   try {
@@ -48154,6 +48194,7 @@ var CACHE_FILE = path3.join(cacheDir(), "models-dev.json");
 var TTL_MS = 24 * 60 * 60 * 1e3;
 var cache2 = null;
 var loading = null;
+var warnedConflicts = /* @__PURE__ */ new Set();
 var snapshotReg = null;
 var snapshotMs = 0;
 function bundledSnapshotRegistry() {
@@ -48326,15 +48367,22 @@ function registryLookup(reg, model, host) {
   }
   if (provider === void 0) {
     const suffix = `/${model}`;
-    let agreed;
+    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 (agreed === void 0) agreed = ctx;
-      else if (agreed !== ctx) return void 0;
+      if (max === void 0 || ctx > max) max = ctx;
+      distinct.add(ctx);
+      parts.push(`${key}=${ctx}`);
     }
-    return agreed;
+    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}`);
+    }
+    return max;
   }
   return void 0;
 }
@@ -48459,11 +48507,14 @@ ${summary}` }]
   }
   return { items, replaced, dropped };
 }
-function codexCompactGate(session, effectiveLimit, transformOk) {
-  if (!transformOk || effectiveLimit <= 0) return false;
+function codexCompactGatePre(session, effectiveLimit) {
+  if (effectiveLimit <= 0) return false;
   if (session.stats.lastInputTokens >= effectiveLimit * 0.9) return false;
   return session.state.blocks.some((b2) => b2.active);
 }
+function codexCompactGate(session, effectiveLimit, transformOk) {
+  return transformOk && codexCompactGatePre(session, effectiveLimit);
+}
 function buildTriggerForgeBody(summaryText, usage, stream2) {
   const compactionItem = {
     type: "compaction",
@@ -49940,7 +49991,13 @@ function buildSession(parsed) {
     metadata: parsed.metadata ?? {},
     state: mergeState(parsed.state),
     createdAt: parsed.createdAt ?? Date.now(),
-    lastSeen: Date.now(),
+    // #404: lastSeen reflects the on-disk savedAt (the last real
+    // activity), NOT the restore moment — a restart must not fabricate
+    // activity for every session (broke fallback=latest ties, panel
+    // freshness, and eviction ordering). Consumers that need "has this
+    // session been used since boot" read the restored flag.
+    lastSeen: parsed.savedAt ?? Date.now(),
+    restored: true,
     blockContents,
     lastMessages: Array.isArray(parsed.messages) ? parsed.messages : void 0,
     inFlight: 0,
@@ -49995,7 +50052,8 @@ async function initSessions() {
   await store.migrateLegacyIds();
   const loaded = await store.loadAll();
   if (loaded.size > MAX_SESSIONS) {
-    const entries = [...loaded.entries()].sort((a, b2) => (b2[1].createdAt ?? 0) - (a[1].createdAt ?? 0));
+    const freshness = (s3) => Math.max(s3.createdAt ?? 0, s3.lastSeen ?? 0);
+    const entries = [...loaded.entries()].sort((a, b2) => freshness(b2[1]) - freshness(a[1]));
     for (const [id, s3] of entries) {
       if (sessions.size >= MAX_SESSIONS) break;
       sessions.set(id, s3);
@@ -50008,6 +50066,7 @@ function getSession(id, meta) {
   const existing = sessions.get(id);
   if (existing) {
     existing.lastSeen = Date.now();
+    existing.restored = false;
     if (meta?.protocol && !existing.meta.protocol) existing.meta.protocol = meta.protocol;
     if (meta?.upstreamOrigin && !existing.meta.upstreamOrigin) existing.meta.upstreamOrigin = meta.upstreamOrigin;
     if (meta?.label && !existing.meta.label) existing.meta.label = meta.label;
@@ -50017,6 +50076,7 @@ function getSession(id, meta) {
   const reloaded = store.loadSync(id, meta);
   if (reloaded) {
     reloaded.lastSeen = Date.now();
+    reloaded.restored = false;
     reloaded.persisted = true;
     sessions.set(id, reloaded);
     return reloaded;
@@ -51027,7 +51087,7 @@ async function saveProviders(){const button=byId("save-providers");busy(button,t
 async function saveCompress(){const button=byId("save-compress");busy(button,true,"\u4FDD\u5B58\u4E2D\u2026");try{const text=byId("compress-json").value.trim();const compress=text?JSON.parse(text):null;await json("/__bili/config",{method:"PUT",headers:{"content-type":"application/json"},body:JSON.stringify({compress})});toast("\u538B\u7F29\u53C2\u6570\u5DF2\u70ED\u66F4\u65B0")}catch(error){toast(String(error),true)}finally{busy(button,false)}}
 async function loadOverrides(){try{const data=await loadConfig();const providers=data.providers||{};const box=byId("upstream-overrides");const entries=Object.entries(providers);if(entries.length===0){box.innerHTML='<p class="status">\u6682\u65E0 Provider\u3002\u5728"\u9AD8\u7EA7\u8BBE\u7F6E"\u9875\u6DFB\u52A0 Provider \u540E\u53EF\u5728\u6B64\u6309 URL \u914D\u7F6E\u4EE3\u7406\u3002</p>';return}box.innerHTML='<table><thead><tr><th>Provider URL</th><th>\u4E0A\u6E38\u4EE3\u7406\uFF08\u7A7A=\u7EE7\u627F\u5168\u5C40\uFF09</th></tr></thead><tbody>'+entries.map(([url,route])=>'<tr><td class="mono">'+escapeHtml(url)+'</td><td><input class="mono override-proxy" data-url="'+escapeHtml(url)+'" value="'+escapeHtml((route&&route.proxy)||"")+'" placeholder="\u7EE7\u627F\u5168\u5C40"></td></tr>').join("")+'</tbody></table>'}catch(error){byId("upstream-overrides").textContent=String(error)}}
 async function saveOverrides(){const button=byId("save-overrides");busy(button,true,"\u4FDD\u5B58\u4E2D\u2026");try{const data=await loadConfig();const providers=data.providers||{};document.querySelectorAll(".override-proxy").forEach((input)=>{const url=input.dataset.url;if(!url)return;if(!providers[url])providers[url]={};const val=input.value.trim();if(val)providers[url].proxy=val;else delete providers[url].proxy});await json("/__bili/config",{method:"PUT",headers:{"content-type":"application/json"},body:JSON.stringify({providers})});await json("/__bili/config/reload",{method:"POST"});toast("\u6309 URL \u8986\u76D6\u5DF2\u70ED\u66F4\u65B0");await loadOverrides()}catch(error){toast(String(error),true)}finally{busy(button,false)}}
-async function loadSessions(){try{const data=await json("/__bili/stats");const rows=(data.sessions||[]).map((item)=>"<tr><td>"+escapeHtml(item.title||"\u2014")+"</td><td>"+escapeHtml(item.protocol||"\u2014")+"</td><td class=mono>"+escapeHtml(item.label||"\u2014")+"</td><td>"+escapeHtml(item.requests)+"</td><td>"+escapeHtml(item.contextTokens)+"</td><td>"+escapeHtml(new Date(item.lastSeen).toLocaleString())+"</td></tr>").join("");byId("sessions-body").innerHTML=rows||'<tr><td colspan="6">\u6682\u65E0\u4F1A\u8BDD</td></tr>'}catch(error){toast(String(error),true);throw error}}
+async function loadSessions(){try{const data=await json("/__bili/stats");const rows=(data.sessions||[]).map((item)=>"<tr><td>"+escapeHtml(item.title||"\u2014")+"</td><td>"+escapeHtml(item.protocol||"\u2014")+"</td><td class=mono>"+escapeHtml(item.label||"\u2014")+"</td><td>"+escapeHtml(item.requests)+"</td><td>"+escapeHtml(item.contextTokens)+"</td><td>"+escapeHtml(new Date(item.lastSeen).toLocaleString())+(item.restored?" <span class=muted>(\u6062\u590D\uFF0C\u5F00\u673A\u540E\u672A\u6D3B\u52A8)</span>":"")+"</td></tr>").join("");byId("sessions-body").innerHTML=rows||'<tr><td colspan="6">\u6682\u65E0\u4F1A\u8BDD</td></tr>'}catch(error){toast(String(error),true);throw error}}
 async function refreshSessions(){const button=byId("refresh-sessions");busy(button,true,"\u5237\u65B0\u4E2D\u2026");try{await loadSessions();toast("\u4F1A\u8BDD\u5217\u8868\u5DF2\u5237\u65B0")}catch{}finally{busy(button,false)}}
 document.querySelectorAll(".copy-btn").forEach((btn)=>btn.addEventListener("click",async()=>{busy(btn,true,"\u590D\u5236\u4E2D\u2026");try{await navigator.clipboard.writeText(btn.dataset.copy||"");toast("\u5DF2\u590D\u5236")}catch(e){toast(String(e),true)}finally{busy(btn,false)}}));byId("save-upstream").addEventListener("click",saveUpstream);byId("test-upstream").addEventListener("click",testUpstream);byId("save-providers").addEventListener("click",saveProviders);byId("save-compress").addEventListener("click",saveCompress);byId("save-overrides").addEventListener("click",saveOverrides);byId("refresh-sessions").addEventListener("click",refreshSessions);
 Promise.all([loadConfig(),loadUpstream()]).catch((error)=>toast(String(error),true));setInterval(()=>{if(byId("page-sessions").classList.contains("active"))loadSessions().catch(()=>{})},5000);
@@ -51037,7 +51097,7 @@ Promise.all([loadConfig(),loadUpstream()]).catch((error)=>toast(String(error),tr
 var WEB_STYLES = `
 :root{--bg:#0f1115;--panel:#171a21;--panel2:#1e222c;--line:#2b313d;--text:#e7eaf0;--muted:#8f98a8;--blue:#6ea8fe;--green:#55c98b;--yellow:#e9b949;--red:#ef6b73;--radius:10px}
 *{box-sizing:border-box}body{margin:0;background:var(--bg);color:var(--text);font:14px/1.5 -apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;min-height:100vh}.app{display:grid;grid-template-columns:220px 1fr;min-height:100vh}.sidebar{border-right:1px solid var(--line);padding:22px 14px;background:#12151b;position:sticky;top:0;height:100vh}.brand{font-size:18px;font-weight:700;padding:0 10px 22px}.brand span{color:var(--blue)}.version{display:block;color:var(--muted);font-size:11px;font-weight:400;margin-top:2px}.nav{display:grid;gap:4px}.nav button{appearance:none;border:0;background:transparent;color:var(--muted);text-align:left;padding:10px 12px;border-radius:7px;cursor:pointer;font:inherit}.nav button:hover,.nav button.active{background:var(--panel2);color:var(--text)}.main{padding:32px;max-width:1050px;width:100%}.page{display:none}.page.active{display:block}h1{font-size:23px;margin:0 0 4px}h2{font-size:15px;margin:0}.lead{color:var(--muted);margin:0 0 24px}.grid{display:grid;grid-template-columns:repeat(2,minmax(0,1fr));gap:14px}.card{background:var(--panel);border:1px solid var(--line);border-radius:var(--radius);padding:18px;margin-bottom:14px}.card-head{display:flex;justify-content:space-between;align-items:center;margin-bottom:16px}.badge{border-radius:99px;padding:3px 9px;font-size:12px;background:var(--panel2);color:var(--muted)}.badge.ok{color:var(--green)}.badge.warn{color:var(--yellow)}.badge.err{color:var(--red)}.kv{display:grid;grid-template-columns:135px 1fr;gap:10px 14px;margin:0}.kv dt{color:var(--muted)}.kv dd{margin:0;word-break:break-all}.mono{font-family:"Cascadia Code",Consolas,monospace;font-size:12px}.actions{display:flex;gap:9px;flex-wrap:wrap;margin-top:18px}.btn{border:1px solid var(--line);background:var(--panel2);color:var(--text);border-radius:7px;padding:8px 13px;cursor:pointer;font:inherit}.btn:hover{border-color:#4b5668}.btn.primary{background:var(--blue);border-color:var(--blue);color:#0b1220;font-weight:600}.btn.danger{color:var(--red)}.btn:disabled{opacity:.55;cursor:wait}.field{display:grid;gap:6px;margin:12px 0}.field label{color:var(--muted);font-size:12px}input,textarea,select{width:100%;background:#101319;color:var(--text);border:1px solid var(--line);border-radius:7px;padding:9px 10px;font:inherit}textarea{min-height:250px;font-family:"Cascadia Code",Consolas,monospace;font-size:12px;resize:vertical}.modes{display:flex;gap:18px;margin:12px 0}.modes label{display:flex;gap:7px;align-items:center;color:var(--muted)}.modes input{width:auto}.notice{border-left:3px solid var(--yellow);padding:10px 12px;background:rgba(233,185,73,.08);color:#d8c185;margin-top:14px}.status{color:var(--muted);white-space:pre-wrap}.status.ok{color:var(--green)}.status.err{color:var(--red)}table{width:100%;border-collapse:collapse}th,td{text-align:left;border-bottom:1px solid var(--line);padding:9px 8px;font-size:12px}th{color:var(--muted);font-weight:500}.toast{position:fixed;right:24px;bottom:24px;background:var(--green);color:#07140d;padding:10px 15px;border-radius:8px;opacity:0;transform:translateY(8px);transition:.2s;pointer-events:none}.toast.show{opacity:1;transform:none}.toast.err{background:var(--red);color:#1b0608}
-.section-title{font-size:17px;margin:28px 0 4px}.hint{color:var(--muted);margin:0 0 14px;font-size:13px}.btn.small{padding:4px 10px;font-size:12px}.config-block{background:var(--panel2);border:1px solid var(--line);border-radius:6px;padding:10px 14px;margin:8px 0 12px;overflow-x:auto;font-size:13px}.config-block code{font-family:"Cascadia Code",Consolas,monospace;color:var(--text)}.fork-ribbon{position:fixed;top:0;right:0;background:#1a1a2e;color:#fff;padding:6px 16px;font-size:12px;text-decoration:none;border-radius:0 0 0 8px;z-index:100;opacity:.8;transition:.2s}.fork-ribbon:hover{opacity:1}
+.section-title{font-size:17px;margin:28px 0 4px}.hint{color:var(--muted);margin:0 0 14px;font-size:13px}.muted{color:var(--muted);font-size:11px}.btn.small{padding:4px 10px;font-size:12px}.config-block{background:var(--panel2);border:1px solid var(--line);border-radius:6px;padding:10px 14px;margin:8px 0 12px;overflow-x:auto;font-size:13px}.config-block code{font-family:"Cascadia Code",Consolas,monospace;color:var(--text)}.fork-ribbon{position:fixed;top:0;right:0;background:#1a1a2e;color:#fff;padding:6px 16px;font-size:12px;text-decoration:none;border-radius:0 0 0 8px;z-index:100;opacity:.8;transition:.2s}.fork-ribbon:hover{opacity:1}
 @media(max-width:760px){.app{grid-template-columns:1fr}.sidebar{height:auto;position:static;border-right:0;border-bottom:1px solid var(--line)}.nav{display:flex;overflow:auto}.main{padding:20px}.grid{grid-template-columns:1fr}.kv{grid-template-columns:110px 1fr}}
 `;
 
@@ -51066,7 +51126,7 @@ function renderPage(origin, version2) {
 <div class="card"><div class="card-head"><h2>ZCode\uFF08\u7F16\u7A0B\u5957\u9910\uFF09</h2><span class="badge">MITM</span><button class="btn small copy-btn" data-copy="${caPathEsc}">\u590D\u5236 CA \u8DEF\u5F84</button></div><dl class="kv"><dt>\u65B9\u5F0F</dt><dd>MITM \u900F\u660E\u4EE3\u7406</dd><dt>\u8BBE\u7F6E</dt><dd>Settings \u2192 Network \u2192 HTTP Proxy = <span class="mono">${origin}</span></dd><dt>CA \u8BC1\u4E66</dt><dd class="mono">${caPathEsc}${caReady ? "" : "\uFF08\u5C1A\u672A\u751F\u6210\uFF0C\u542F\u52A8\u4EE3\u7406\u540E\u81EA\u52A8\u521B\u5EFA\uFF09"}</dd></dl><p class="hint">CA \u8DEF\u5F84\u662F\u672C\u673A bili \u5B9E\u9645\u751F\u6210\u7684\u7EDD\u5BF9\u8DEF\u5F84\uFF0C\u70B9"\u590D\u5236 CA \u8DEF\u5F84"\u540E\u76F4\u63A5\u7C98\u8D34\u8FDB ZCode\u3002\u26A0\uFE0F Windows \u4E0A ZCode \u4E0D\u4F1A\u5C55\u5F00 <span class="mono">~</span>\uFF0C\u8BF7\u52FF\u624B\u586B <span class="mono">~/\u2026</span> \u5F62\u5F0F\u7684\u8DEF\u5F84\u3002</p></div></section>
 <section id="page-upstream" class="page"><h1>\u4E0A\u6E38\u7F51\u7EDC</h1><p class="lead">\u63A7\u5236 bili \u5982\u4F55\u8BBF\u95EE\u771F\u5B9E Provider\uFF0C\u4E0D\u4F1A\u6539\u53D8\u5BA2\u6237\u7AEF\u7684\u672C\u5730\u8DEF\u7531\u5730\u5740\u3002</p><div class="card"><div class="card-head"><h2>\u5168\u5C40\u4EE3\u7406</h2></div><div class="modes"><label><input type="radio" name="proxy-mode" value="direct" checked>\u76F4\u8FDE\uFF08\u9ED8\u8BA4\uFF09</label><label><input type="radio" name="proxy-mode" value="manual">\u624B\u52A8\u4EE3\u7406</label><label><input type="radio" name="proxy-mode" value="auto">\u81EA\u52A8\uFF08\u8DDF\u968F\u7CFB\u7EDF\uFF09</label></div><div class="field"><label>HTTP / HTTPS Proxy</label><input id="proxy-url" class="mono" placeholder="http://127.0.0.1:7897"></div><dl class="kv"><dt>\u5F53\u524D\u6765\u6E90</dt><dd id="upstream-source">\u76F4\u8FDE</dd><dt>\u6709\u6548\u4EE3\u7406</dt><dd id="upstream-effective" class="mono">direct</dd><dt>\u7CFB\u7EDF PAC</dt><dd id="upstream-pac" class="mono">\u2014</dd><dt>\u72B6\u6001</dt><dd id="upstream-state" class="status">\u5C1A\u672A\u6D4B\u8BD5</dd></dl><div class="actions"><button id="save-upstream" class="btn primary">\u4FDD\u5B58\u5E76\u70ED\u66F4\u65B0</button><button id="test-upstream" class="btn">\u6D4B\u8BD5\u8FDE\u63A5</button></div></div><div class="card"><div class="card-head"><h2>\u6309 URL \u8986\u76D6</h2></div><p class="status">\u4E3A\u7279\u5B9A Provider \u5355\u72EC\u8BBE\u7F6E\u4E0A\u6E38\u4EE3\u7406\uFF0C\u8986\u76D6\u5168\u5C40\u8BBE\u7F6E\u3002\u7559\u7A7A = \u7EE7\u627F\u5168\u5C40\u3002</p><div id="upstream-overrides" class="status">\u52A0\u8F7D\u4E2D</div><div class="actions"><button id="save-overrides" class="btn primary">\u4FDD\u5B58\u8986\u76D6\u5E76\u70ED\u66F4\u65B0</button></div></div></section>
 <section id="page-sessions" class="page"><div class="card-head"><div><h1>\u4F1A\u8BDD</h1><p class="lead">ACP \u538B\u7F29\u72B6\u6001\u4E0E\u4E0A\u6E38\u7528\u91CF\u3002</p></div><button id="refresh-sessions" class="btn">\u5237\u65B0</button></div><div class="card"><table><thead><tr><th>\u6807\u9898</th><th>\u534F\u8BAE</th><th>\u6807\u8BC6</th><th>\u8BF7\u6C42</th><th>\u4E0A\u4E0B\u6587</th><th>\u6700\u540E\u6D3B\u52A8</th></tr></thead><tbody id="sessions-body"></tbody></table></div></section>
-<section id="page-settings" class="page"><h1>\u9AD8\u7EA7\u8BBE\u7F6E</h1><p class="lead">\u76F4\u63A5\u7F16\u8F91 providers JSON\uFF08\u6DFB\u52A0\u65B0 Provider\u3001\u914D\u7F6E\u6A21\u578B\u4E0A\u4E0B\u6587\u7A97\u53E3\u3001\u6309 URL \u4EE3\u7406\u7B49\uFF09\uFF1B\u4FDD\u5B58\u540E\u7ACB\u5373\u70ED\u66F4\u65B0\u3002</p><div class="card"><div class="field"><label>providers JSON</label><textarea id="providers-json">{}</textarea></div><div class="actions"><button id="save-providers" class="btn primary">\u4FDD\u5B58\u5E76\u5E94\u7528</button></div></div><div class="card"><div class="card-head"><h2>\u538B\u7F29\u53C2\u6570\uFF08\u5168\u5C40\uFF09</h2></div><p class="status">\u5168\u5C40 compress \u914D\u7F6E\uFF08modelContextLimit / maxContextLimit / emergencyThresholdPercent / nudgeGrowthTokens / preserveRecentMessages / preserveRecentTokens / minCompressRange / tiers / injectTool / injectNudge\uFF09\u3002\u4FDD\u5B58\u540E\u70ED\u66F4\u65B0\uFF0C\u65E0\u9700\u91CD\u542F\u3002\u7F6E\u7A7A = \u6062\u590D\u9ED8\u8BA4\u3002</p><div class="field"><label>compress JSON</label><textarea id="compress-json" placeholder='{"modelContextLimit": 200000}'></textarea></div><div class="actions"><button id="save-compress" class="btn primary">\u4FDD\u5B58\u5E76\u70ED\u66F4\u65B0</button></div></div></section></main></div><div id="toast" class="toast"></div><script>${WEB_CLIENT}</script></body></html>`;
+<section id="page-settings" class="page"><h1>\u9AD8\u7EA7\u8BBE\u7F6E</h1><p class="lead">\u76F4\u63A5\u7F16\u8F91 providers JSON\uFF08\u6DFB\u52A0\u65B0 Provider\u3001\u914D\u7F6E\u6A21\u578B\u4E0A\u4E0B\u6587\u7A97\u53E3\u3001\u6309 URL \u4EE3\u7406\u7B49\uFF09\uFF1B\u4FDD\u5B58\u540E\u7ACB\u5373\u70ED\u66F4\u65B0\u3002</p><div class="card"><div class="field"><label>providers JSON</label><textarea id="providers-json">{}</textarea></div><div class="actions"><button id="save-providers" class="btn primary">\u4FDD\u5B58\u5E76\u5E94\u7528</button></div></div><div class="card"><div class="card-head"><h2>\u538B\u7F29\u53C2\u6570\uFF08\u5168\u5C40\uFF09</h2></div><p class="status">\u5168\u5C40 compress \u914D\u7F6E\uFF08modelContextLimit / maxContextLimit / emergencyThresholdPercent / nudgeGrowthTokens / preserveRecentMessages / preserveRecentTokens / minCompressRangeChars\uFF08\u5355\u4F4D\uFF1A\u5B57\u7B26\uFF0C\u65E7\u540D minCompressRange \u517C\u5BB9\uFF09 / tiers / injectTool / injectNudge\uFF09\u3002\u4FDD\u5B58\u540E\u70ED\u66F4\u65B0\uFF0C\u65E0\u9700\u91CD\u542F\u3002\u7F6E\u7A7A = \u6062\u590D\u9ED8\u8BA4\u3002</p><div class="field"><label>compress JSON</label><textarea id="compress-json" placeholder='{"modelContextLimit": 200000}'></textarea></div><div class="actions"><button id="save-compress" class="btn primary">\u4FDD\u5B58\u5E76\u70ED\u66F4\u65B0</button></div></div></section></main></div><div id="toast" class="toast"></div><script>${WEB_CLIENT}</script></body></html>`;
 }
 
 // src/web/api.ts
@@ -54353,10 +54413,14 @@ function handlePluginStatus(conversationId2, res, fallbackLatest = false) {
   let session = entry ? peekSession(entry.sessionId) : void 0;
   let viaFallback = false;
   if ((!entry || !session) && fallbackLatest) {
-    const latest = listSessions().filter((s3) => typeof s3.lastSeen === "number").sort((a, b2) => (b2.lastSeen ?? 0) - (a.lastSeen ?? 0))[0];
+    const latest = listSessions().filter((s3) => s3.restored !== true).sort((a, b2) => (b2.lastSeen ?? 0) - (a.lastSeen ?? 0))[0];
     if (latest) {
       session = latest;
       viaFallback = true;
+    } else {
+      res.writeHead(404, { "content-type": "application/json" });
+      res.end(JSON.stringify({ ok: false, error: "no session with activity since boot \u2014 issue a model request or pass the conversation id" }));
+      return;
     }
   }
   if (!session) {
@@ -54563,6 +54627,12 @@ data: ${JSON.stringify({ type: "content_block_delta", index, delta: { type: delt
       return new Promise((r) => res.once("drain", () => r()));
     }
   };
+  const settleUsage = () => {
+    if (session && (acc.inputTokens !== void 0 || acc.outputTokens !== void 0 || acc.cachedTokens !== void 0)) {
+      applyUsageSample(session, acc, protocol);
+      markDirty(session);
+    }
+  };
   const pushField = (field, index, text) => {
     const s3 = filterFor(field, index);
     const clean = s3.filter.push(text);
@@ -54668,10 +54738,14 @@ data: ${JSON.stringify({ type: "content_block_delta", index, delta: { type: delt
     }
     const rest = flushTails();
     if (rest.length > 0 && !res.destroyed && !res.writableEnded) await write(rest);
-    if (session && (acc.inputTokens !== void 0 || acc.outputTokens !== void 0 || acc.cachedTokens !== void 0)) {
-      applyUsageSample(session, acc, protocol);
-      markDirty(session);
+    settleUsage();
+  } catch (e) {
+    settleUsage();
+    if (res.destroyed || res.writableEnded) {
+      log2?.("client aborted mid-stream");
+      return;
     }
+    throw e;
   } finally {
     reader.releaseLock();
     res.end();
@@ -54705,6 +54779,12 @@ async function pipePluginResponsesWithStrip(stream2, res, session, log2) {
       return new Promise((r) => res.once("drain", () => r()));
     }
   };
+  const settleUsage = () => {
+    if (session && (acc.inputTokens !== void 0 || acc.outputTokens !== void 0 || acc.cachedTokens !== void 0)) {
+      applyUsageSample(session, acc, "responses");
+      markDirty(session);
+    }
+  };
   let lastDeltaMeta = null;
   const flushTail = (after) => {
     const tail = tagFilter.flush();
@@ -54777,10 +54857,14 @@ async function pipePluginResponsesWithStrip(stream2, res, session, log2) {
       const rest = flushTail("");
       if (rest.length > 0) await write(rest);
     }
-    if (session && (acc.inputTokens !== void 0 || acc.outputTokens !== void 0 || acc.cachedTokens !== void 0)) {
-      applyUsageSample(session, acc, "responses");
-      markDirty(session);
+    settleUsage();
+  } catch (e) {
+    settleUsage();
+    if (res.destroyed || res.writableEnded) {
+      log2?.("client aborted mid-stream");
+      return;
     }
+    throw e;
   } finally {
     reader.releaseLock();
     res.end();
@@ -54806,10 +54890,16 @@ function rebuildEvent(rawEvent, ev) {
 async function pipePluginJson(stream2, res, session, protocol) {
   const reader = stream2.getReader();
   const chunks = [];
-  for (; ; ) {
-    const { done, value } = await reader.read();
-    if (done) break;
-    if (value && value.length > 0) chunks.push(Buffer.from(value));
+  try {
+    for (; ; ) {
+      const { done, value } = await reader.read();
+      if (done) break;
+      if (value && value.length > 0) chunks.push(Buffer.from(value));
+    }
+  } catch (e) {
+    if (!res.writableEnded) res.end();
+    if (res.destroyed || res.writableEnded) return;
+    throw e;
   }
   reader.releaseLock();
   const text = Buffer.concat(chunks).toString("utf8");
@@ -56460,7 +56550,10 @@ async function startServer(opts) {
       await handle(req, res, opts, core, config, log2, instanceId, instanceStartedAt);
     } catch (err2) {
       const msg = String(err2);
-      log2("error", msg);
+      const e = err2;
+      const clientAbort = (e?.name === "AbortError" || /abort/i.test(String(e?.message ?? ""))) && (res.destroyed || res.writableEnded);
+      if (clientAbort) log2("info", `client aborted mid-stream: ${msg}`);
+      else log2("error", msg);
       if (!res.headersSent) {
         const status = msg.includes("exceeds") ? 413 : 502;
         res.writeHead(status, { "content-type": "application/json" });
@@ -57012,6 +57105,21 @@ ${bodyBuffer.toString("utf8")}`);
     try {
       await withSessionLock(session, async () => {
         const runPrepare = () => countTokens ? prepareCountTokens(parsed, core, reqConfig, log2, session) : protocol === "anthropic" ? prepareAnthropic(parsed, req, opts, core, reqConfig, reqPrompts, log2, session, pluginMode) : protocol === "openai" ? prepareOpenai(parsed, req, opts, core, reqConfig, reqPrompts, log2, session, pluginMode, nativeWindow) : responsesCompact ? prepareResponsesCompact(bodyBuffer, parsed, session, req, core, reqConfig, log2) : prepareResponses(parsed, req, opts, core, reqConfig, reqPrompts, log2, session, responsesIdentity, pluginMode, upstreamOrigin, nativeWindow);
+        const isCodexCompactTrigger = protocol === "responses" && !responsesCompact && isCodexClient(req.headers) && hasCompactionTrigger(parsed.input);
+        if (isCodexCompactTrigger) {
+          const mode = codexCompactMode();
+          const gatePre = codexCompactGatePre(session, reqConfig.modelContextLimit);
+          if (mode === "intercept" && gatePre) prepared = runPrepare();
+          if (prepared?.codexForge) {
+            await forward(req, res, opts, prepared.body, prepared, core, reqConfig, log2, route, instanceId, affinity);
+            rememberPluginMessages(sessionId, prepared.processedMessages, prepared.originalMessages, prepared.nudge);
+            return;
+          }
+          const why = mode !== "intercept" ? "BILI_CODEX_COMPACT=pass" : !gatePre ? "gate preconditions not met" : "transform/forge failed";
+          log2("info", `[${session.id}] codex compaction_trigger request not intercepted (${why}) \u2014 forwarding verbatim (no preflight, no rebuild, no window clamp)`);
+          await forward(req, res, opts, bodyBuffer, null, core, reqConfig, log2, route, instanceId, affinity);
+          return;
+        }
         prepared = runPrepare();
         if (!countTokens && !responsesCompact) {
           const outcome = await preflightCompressIfNeeded(
@@ -57457,7 +57565,7 @@ function prepareResponsesCompact(body, parsed, session, req, core, config, log2)
     compressInjected: false,
     resetAfterSuccess: true
   };
-  if (codexCompactMode() !== "intercept" || !isCodexClient(req.headers) || !Array.isArray(parsed.input)) {
+  if (codexCompactMode() !== "intercept" || !isCodexClient(req.headers) || !Array.isArray(parsed.input) || !codexCompactGatePre(session, config.modelContextLimit)) {
     return base;
   }
   const prevState = session.state;
@@ -57849,187 +57957,205 @@ ${hdrText}
     return;
   }
   if (prepared?.pluginMode) {
-    if (prepared.stream) {
-      if (prepared.protocol === "responses") {
-        await pipePluginResponsesWithStrip(upstream.body, res, prepared.session, (msg) => log2("info", `[${prepared.session.id}] ${msg}`));
-      } else {
-        await pipePluginChatWithStrip(upstream.body, res, prepared.protocol, prepared.session, (msg) => log2("info", `[${prepared.session.id}] ${msg}`));
-      }
-    } else {
-      await pipePluginJson(upstream.body, res, prepared.session, prepared.protocol);
-    }
-    clearUpstreamTimer();
-    return;
-  }
-  let responseBody = upstream.body;
-  if (prepared !== null && prepared.stream && maxFakeCompletionRetries() > 0) {
-    const resolvedBuf = await resolveFakeCompletion(upstream.body, {
-      protocol: prepared.protocol,
-      body,
-      upstreamUrl,
-      reqHeaders: buildForwardHeaders(headers),
-      proxyUrl,
-      signal: clientAbort.signal,
-      session: prepared.session,
-      log: log2
-    });
-    responseBody = bufferToStream(resolvedBuf);
-  }
-  const useRewriter = prepared !== null && prepared.compressInjected && prepared.processedMessages.length > 0;
-  if (!useRewriter || prepared === null) {
-    if (prepared && prepared.resetAfterSuccess) {
-      const [toClient, toObserve] = responseBody.tee();
-      const observed = observeResponsesTerminalState(toObserve, prepared.stream);
-      const tagLog = (msg) => log2("info", `[${prepared.session.id}] ${msg}`);
-      if ((upstream.headers.get("content-type") ?? "").includes("text/event-stream")) {
-        await pipePluginResponsesWithStrip(toClient, res, void 0, tagLog);
-      } else {
-        await pipeThrough(toClient, res);
-      }
-      clearUpstreamTimer();
-      const terminal = await observed;
-      if (terminal === "completed") {
-        markNativeCompactionBoundary(prepared.session);
-        log2("info", `[${prepared.session.id}] native Responses compact completed; rebase scheduled for next Responses turn`);
-      } else {
-        log2("warn", `[${prepared.session.id}] native compact response terminal=${terminal}; rebase NOT scheduled`);
+    try {
+      let pluginBody = upstream.body;
+      if (prepared.stream && maxFakeCompletionRetries() > 0) {
+        const resolvedBuf = await resolveFakeCompletion(pluginBody, {
+          protocol: prepared.protocol,
+          body,
+          upstreamUrl,
+          reqHeaders: buildForwardHeaders(headers),
+          proxyUrl,
+          signal: clientAbort.signal,
+          session: prepared.session,
+          log: log2
+        });
+        pluginBody = bufferToStream(resolvedBuf);
       }
-    } else if (prepared && prepared.stream && (upstream.headers.get("content-type") ?? "").includes("text/event-stream")) {
-      const p2 = prepared;
-      const tagLog = (msg) => log2("info", `[${p2.session.id}] ${msg}`);
-      if (p2.protocol === "responses") {
-        await pipePluginResponsesWithStrip(responseBody, res, void 0, tagLog);
+      if (prepared.stream) {
+        if (prepared.protocol === "responses") {
+          await pipePluginResponsesWithStrip(pluginBody, res, prepared.session, (msg) => log2("info", `[${prepared.session.id}] ${msg}`));
+        } else {
+          await pipePluginChatWithStrip(pluginBody, res, prepared.protocol, prepared.session, (msg) => log2("info", `[${prepared.session.id}] ${msg}`));
+        }
       } else {
-        await pipePluginChatWithStrip(responseBody, res, p2.protocol, void 0, tagLog);
+        await pipePluginJson(pluginBody, res, prepared.session, prepared.protocol);
       }
-      clearUpstreamTimer();
-    } else if (prepared && (upstream.headers.get("content-type") ?? "").includes("application/json")) {
-      await pipePluginJson(responseBody, res, void 0, prepared.protocol);
-      clearUpstreamTimer();
-    } else {
-      await pipeThrough(responseBody, res);
+    } finally {
       clearUpstreamTimer();
     }
     return;
   }
-  const ctx = {
-    core,
-    config,
-    messages: prepared.originalMessages,
-    session: prepared.session,
-    log: (msg) => log2("info", `[${prepared.session.id}] ${msg}`),
-    debug: opts.debug
-  };
-  if (prepared.stream) {
-    let streamToRead = responseBody;
-    let dumpRaw;
-    if (opts.dumpSse) {
-      const [a, b2] = responseBody.tee();
-      streamToRead = a;
-      dumpRaw = dumpStreamToFile(b2, opts.dumpSse, `${Date.now()}-${safeSessionId(prepared.session.id)}-raw.sse`);
-    }
-    try {
-      const parsedReq = JSON.parse(typeof body === "string" ? body : body.toString("utf8"));
-      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 refreshFolded = (current) => {
-        const turn = core.processTurn({
-          messages: prepared.originalMessages,
-          state: prepared.session.state,
-          config,
-          tokenCount: prepared.session.stats.lastInputTokens,
-          renderTags: prepared.renderTags ?? "text-only"
-        });
-        prepared.session.state = turn.state;
-        const records = current.filter((m2) => typeof m2.id === "string" && m2.id.startsWith("acp_loop_"));
-        return stripKernelSummaries([...turn.messages, ...records], turn.state);
-      };
-      const loop = runCompressLoop(
-        streamToRead,
-        { core, config, messages: prepared.processedMessages.length > 0 ? prepared.processedMessages : prepared.originalMessages, compressMessages: prepared.originalMessages, session: prepared.session, log: ctx.log, proxyUrl, protocol: prepared.protocol, textProtocol, debug: opts.debug, nudge: prepared.nudge, refreshFolded },
-        parsedReq,
-        { url: upstreamUrl, headers: reqHeaders },
-        adapter,
-        systemPrompt,
-        clientAbort.signal
-      );
-      let protocolFragmentWarned = false;
-      for await (const chunk of loop) {
-        if (res.destroyed || res.writableEnded) break;
-        {
-          const s3 = chunk.toString("utf8");
-          if (s3.includes("<acp ") || s3.includes("</acp")) {
-            log2("warn", `[${prepared.session.id}] tag echo: ${prepared.protocol} response stream contains <acp tag`);
-          } else if (!protocolFragmentWarned && containsToolCallXmlFragment(s3)) {
-            protocolFragmentWarned = true;
-            log2("warn", `[${prepared.session.id}] tag echo: ${prepared.protocol} response stream contains tool-call XML fragment (possible tag echo; not stripped)`);
-          }
-        }
-        res.write(chunk);
-        if (res.writableNeedDrain) {
-          await Promise.race([
-            new Promise((r) => res.once("drain", () => r())),
-            new Promise((r) => res.once("close", () => r()))
-          ]);
+  let responseBody = upstream.body;
+  try {
+    if (prepared !== null && prepared.stream && maxFakeCompletionRetries() > 0) {
+      const resolvedBuf = await resolveFakeCompletion(upstream.body, {
+        protocol: prepared.protocol,
+        body,
+        upstreamUrl,
+        reqHeaders: buildForwardHeaders(headers),
+        proxyUrl,
+        signal: clientAbort.signal,
+        session: prepared.session,
+        log: log2
+      });
+      responseBody = bufferToStream(resolvedBuf);
+    }
+    const useRewriter = prepared !== null && prepared.compressInjected && prepared.processedMessages.length > 0;
+    if (!useRewriter || prepared === null) {
+      if (prepared && prepared.resetAfterSuccess) {
+        const [toClient, toObserve] = responseBody.tee();
+        const observed = observeResponsesTerminalState(toObserve, prepared.stream);
+        const tagLog = (msg) => log2("info", `[${prepared.session.id}] ${msg}`);
+        if ((upstream.headers.get("content-type") ?? "").includes("text/event-stream")) {
+          await pipePluginResponsesWithStrip(toClient, res, void 0, tagLog);
+        } else {
+          await pipeThrough(toClient, res);
         }
-        if (res.destroyed || res.writableEnded) break;
+        const terminal = await observed;
+        if (terminal === "completed") {
+          markNativeCompactionBoundary(prepared.session);
+          log2("info", `[${prepared.session.id}] native Responses compact completed; rebase scheduled for next Responses turn`);
+        } else {
+          log2("warn", `[${prepared.session.id}] native compact response terminal=${terminal}; rebase NOT scheduled`);
+        }
+      } else if (prepared && prepared.stream && (upstream.headers.get("content-type") ?? "").includes("text/event-stream")) {
+        const p2 = prepared;
+        const tagLog = (msg) => log2("info", `[${p2.session.id}] ${msg}`);
+        if (p2.protocol === "responses") {
+          await pipePluginResponsesWithStrip(responseBody, res, void 0, tagLog);
+        } else {
+          await pipePluginChatWithStrip(responseBody, res, p2.protocol, void 0, tagLog);
+        }
+      } else if (prepared && (upstream.headers.get("content-type") ?? "").includes("application/json")) {
+        await pipePluginJson(responseBody, res, void 0, prepared.protocol);
+      } else {
+        await pipeThrough(responseBody, res);
       }
-      res.end();
-    } catch (e) {
-      emitStreamError(res, prepared.protocol, e?.message ?? String(e), (m2) => log2("error", `[${prepared.session.id}] ${m2}`));
-    } finally {
-      clearUpstreamTimer();
-      if (dumpRaw) await dumpRaw;
+      return;
     }
-  } else {
-    try {
-      const buf = await upstream.arrayBuffer();
-      const text = Buffer.from(buf).toString("utf8");
+    const ctx = {
+      core,
+      config,
+      messages: prepared.originalMessages,
+      session: prepared.session,
+      log: (msg) => log2("info", `[${prepared.session.id}] ${msg}`),
+      debug: opts.debug
+    };
+    if (prepared.stream) {
+      let streamToRead = responseBody;
+      let dumpRaw;
+      if (opts.dumpSse) {
+        const [a, b2] = responseBody.tee();
+        streamToRead = a;
+        dumpRaw = dumpStreamToFile(b2, opts.dumpSse, `${Date.now()}-${safeSessionId(prepared.session.id)}-raw.sse`);
+      }
       try {
-        let json = JSON.parse(text);
-        if (prepared.protocol === "responses" && prepared.responsesTextProtocol) {
-          const requestBody = JSON.parse(typeof body === "string" ? body : body.toString("utf8"));
-          const requestHeaders = buildForwardHeaders(headers);
-          json = await compressLoopResponsesJson(
-            json,
-            { core, config, messages: prepared.originalMessages, session: prepared.session, log: ctx.log, proxyUrl, textProtocol: true },
-            requestBody,
-            { url: upstreamUrl, headers: requestHeaders }
-          );
-        }
-        const u2 = json.usage ?? {};
-        const { total, cached } = usageTotals(prepared.protocol, u2);
-        if (typeof total === "number") {
-          prepared.session.stats.inputTokens += total;
-          prepared.session.stats.lastInputTokens = Math.max(
-            0,
-            total - (prepared.session.stats.compressCreditTokens ?? 0)
-          );
-          if (typeof cached === "number") {
-            prepared.session.stats.cachedTokens += cached;
-            prepared.session.stats.cacheSamples += 1;
+        const parsedReq = JSON.parse(typeof body === "string" ? body : body.toString("utf8"));
+        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 refreshFolded = (current) => {
+          const turn = core.processTurn({
+            messages: prepared.originalMessages,
+            state: prepared.session.state,
+            config,
+            tokenCount: prepared.session.stats.lastInputTokens,
+            renderTags: prepared.renderTags ?? "text-only"
+          });
+          prepared.session.state = turn.state;
+          const records = current.filter((m2) => typeof m2.id === "string" && m2.id.startsWith("acp_loop_"));
+          return stripKernelSummaries([...turn.messages, ...records], turn.state);
+        };
+        const loop = runCompressLoop(
+          streamToRead,
+          { core, config, messages: prepared.processedMessages.length > 0 ? prepared.processedMessages : prepared.originalMessages, compressMessages: prepared.originalMessages, session: prepared.session, log: ctx.log, proxyUrl, protocol: prepared.protocol, textProtocol, debug: opts.debug, nudge: prepared.nudge, refreshFolded },
+          parsedReq,
+          { url: upstreamUrl, headers: reqHeaders },
+          adapter,
+          systemPrompt,
+          clientAbort.signal
+        );
+        let protocolFragmentWarned = false;
+        for await (const chunk of loop) {
+          if (res.destroyed || res.writableEnded) break;
+          {
+            const s3 = chunk.toString("utf8");
+            if (s3.includes("<acp ") || s3.includes("</acp")) {
+              log2("warn", `[${prepared.session.id}] tag echo: ${prepared.protocol} response stream contains <acp tag`);
+            } else if (!protocolFragmentWarned && containsToolCallXmlFragment(s3)) {
+              protocolFragmentWarned = true;
+              log2("warn", `[${prepared.session.id}] tag echo: ${prepared.protocol} response stream contains tool-call XML fragment (possible tag echo; not stripped)`);
+            }
+          }
+          res.write(chunk);
+          if (res.writableNeedDrain) {
+            await Promise.race([
+              new Promise((r) => res.once("drain", () => r())),
+              new Promise((r) => res.once("close", () => r()))
+            ]);
           }
-          const out = u2.completion_tokens ?? u2.output_tokens;
-          if (typeof out === "number") prepared.session.stats.outputTokens += out;
+          if (res.destroyed || res.writableEnded) break;
         }
-        if (prepared.protocol === "openai") {
-          rewriteOpenaiJsonResponse(json, ctx);
-        } else if (prepared.protocol === "responses") {
-          rewriteResponsesJsonResponse(json, ctx);
-        } else {
-          rewriteJsonResponse(json, ctx);
+        res.end();
+      } catch (e) {
+        emitStreamError(res, prepared.protocol, e?.message ?? String(e), (m2) => log2("error", `[${prepared.session.id}] ${m2}`));
+      } finally {
+        clearUpstreamTimer();
+        if (dumpRaw) await dumpRaw;
+        markDirty(prepared.session);
+      }
+    } else {
+      try {
+        const buf = await upstream.arrayBuffer();
+        const text = Buffer.from(buf).toString("utf8");
+        try {
+          let json = JSON.parse(text);
+          if (prepared.protocol === "responses" && prepared.responsesTextProtocol) {
+            const requestBody = JSON.parse(typeof body === "string" ? body : body.toString("utf8"));
+            const requestHeaders = buildForwardHeaders(headers);
+            json = await compressLoopResponsesJson(
+              json,
+              { core, config, messages: prepared.originalMessages, session: prepared.session, log: ctx.log, proxyUrl, textProtocol: true },
+              requestBody,
+              { url: upstreamUrl, headers: requestHeaders }
+            );
+          }
+          const u2 = json.usage ?? {};
+          const { total, cached } = usageTotals(prepared.protocol, u2);
+          if (typeof total === "number") {
+            prepared.session.stats.inputTokens += total;
+            prepared.session.stats.lastInputTokens = Math.max(
+              0,
+              total - (prepared.session.stats.compressCreditTokens ?? 0)
+            );
+            if (typeof cached === "number") {
+              prepared.session.stats.cachedTokens += cached;
+              prepared.session.stats.cacheSamples += 1;
+            }
+            const out = u2.completion_tokens ?? u2.output_tokens;
+            if (typeof out === "number") prepared.session.stats.outputTokens += out;
+          }
+          if (prepared.protocol === "openai") {
+            rewriteOpenaiJsonResponse(json, ctx);
+          } else if (prepared.protocol === "responses") {
+            rewriteResponsesJsonResponse(json, ctx);
+          } else {
+            rewriteJsonResponse(json, ctx);
+          }
+          res.end(JSON.stringify(json));
+        } catch {
+          res.end(text);
         }
-        res.end(JSON.stringify(json));
-      } catch {
-        res.end(text);
+      } finally {
+        clearUpstreamTimer();
+        markDirty(prepared.session);
       }
-    } finally {
-      clearUpstreamTimer();
     }
+  } finally {
+    clearUpstreamTimer();
   }
-  markDirty(prepared.session);
 }
 async function readStreamToBuffer(stream2, maxBytes = 1 << 20) {
   const reader = stream2.getReader();
@@ -58174,7 +58300,8 @@ function sendStats(res) {
     outputTokens: s3.stats.outputTokens,
     cacheSamples: s3.stats.cacheSamples,
     cacheHitPct: s3.stats.cacheSamples > 0 && s3.stats.inputTokens > 0 ? Math.round(s3.stats.cachedTokens / s3.stats.inputTokens * 100) : null,
-    lastSeen: new Date(s3.lastSeen).toISOString()
+    lastSeen: new Date(s3.lastSeen).toISOString(),
+    restored: s3.restored === true
   }));
   res.writeHead(200, { "content-type": "application/json" });
   res.end(JSON.stringify({ sessions: sessions2 }, null, 2));