billion-context 0.1.97-pr.671.489 → 0.1.97

package/dist/index.js

@@ -48726,7 +48726,6 @@ function mergeCompress(global2, provider, model) {
   const promptLevels = [global2?.prompts, provider?.prompts, model?.prompts].filter(Boolean);
   const rangeOf = (s3) => s3?.minCompressRangeChars ?? s3?.minCompressRange;
   const absorbLevels = [global2?.absorb, provider?.absorb, model?.absorb].filter(Boolean);
-  const reasoningLevels = [global2?.reasoning, provider?.reasoning, model?.reasoning].filter(Boolean);
   return {
     modelContextLimit: pick("modelContextLimit"),
     maxContextLimit: pick("maxContextLimit"),
@@ -48740,11 +48739,7 @@ function mergeCompress(global2, provider, model) {
     acknowledgePromptsRisk: pick("acknowledgePromptsRisk"),
     absorb: absorbLevels.length > 0 ? Object.assign({}, ...absorbLevels) : void 0,
     stripImages: pick("stripImages"),
-    stripImagesKeepRecent: pick("stripImagesKeepRecent"),
-    // `reasoning` is a third nested-object field merged sub-field-wise
-    // exactly like `absorb`/`prompts`: a model-level `threshold` must not
-    // discard a provider-level `drop: false`.
-    reasoning: reasoningLevels.length > 0 ? Object.assign({}, ...reasoningLevels) : void 0
+    stripImagesKeepRecent: pick("stripImagesKeepRecent")
   };
 }
 function resolveCompress(routes, upstreamUrl, model, global2) {
@@ -48824,53 +48819,6 @@ function resolveRequestConfig(base, routes, embeddedUrl, model, native, globalCo
   return applyCompressSettings(base, limit, compress);
 }
 
-// src/reasoning-drop.ts
-var DEFAULT_COMPRESS_REASONING = { drop: true, threshold: 2048 };
-function resolveReasoningDrop(cfg) {
-  let threshold = DEFAULT_COMPRESS_REASONING.threshold;
-  if (cfg?.threshold !== void 0) {
-    const t = cfg.threshold;
-    if (typeof t === "number" && Number.isFinite(t) && t >= 0) {
-      threshold = Math.floor(t);
-    }
-  }
-  return { drop: cfg?.drop !== false, threshold };
-}
-function dropCompressReasoning(messages, cfg) {
-  const { drop, threshold } = resolveReasoningDrop(cfg);
-  if (!drop || messages.length === 0) return messages;
-  try {
-    const last = messages.length - 1;
-    const resultAt = /* @__PURE__ */ new Map();
-    for (let i = 0; i <= last; i++) {
-      const m2 = messages[i];
-      if (m2.contentType === "tool-result" && typeof m2.toolCallId === "string" && !resultAt.has(m2.toolCallId)) {
-        resultAt.set(m2.toolCallId, i);
-      }
-    }
-    const dropIdx = /* @__PURE__ */ new Set();
-    for (let i = 0; i <= last; i++) {
-      const m2 = messages[i];
-      if (m2.contentType !== "tool-call" || m2.toolName !== "compress") continue;
-      const ri2 = typeof m2.toolCallId === "string" ? resultAt.get(m2.toolCallId) : void 0;
-      if (ri2 === void 0 || ri2 <= i || ri2 >= last) continue;
-      let total = 0;
-      let j2 = i - 1;
-      while (j2 >= 0 && messages[j2].contentType === "reasoning") {
-        total += (messages[j2].text ?? "").length;
-        j2--;
-      }
-      if (total > threshold) {
-        for (let k2 = j2 + 1; k2 < i; k2++) dropIdx.add(k2);
-      }
-    }
-    if (dropIdx.size === 0) return messages;
-    return messages.filter((_2, idx) => !dropIdx.has(idx));
-  } catch {
-    return messages;
-  }
-}
-
 // src/strip-images.ts
 var DEFAULT_STRIP_IMAGES_KEEP_RECENT = 5;
 var IMAGE_PLACEHOLDER = "[image]";
@@ -56307,28 +56255,15 @@ function handlePluginManifest(res) {
     statusEndpoint: "/__bili/plugin/status"
   }));
 }
-function conversationIdForSession(sessionId) {
-  let bestId;
-  let bestSeen = -Infinity;
-  for (const [cid, entry] of conversations) {
-    if (entry.sessionId === sessionId && entry.lastSeen > bestSeen) {
-      bestId = cid;
-      bestSeen = entry.lastSeen;
-    }
-  }
-  return bestId;
-}
 function handlePluginStatus(conversationId2, res, deps, fallbackLatest = false) {
   let entry = conversations.get(conversationId2);
   let session = entry ? peekSession(entry.sessionId) : void 0;
   let viaFallback = false;
-  let resolvedConversationId = conversationId2;
   if ((!entry || !session) && fallbackLatest) {
     const latest = listSessions().filter((s3) => s3.restored !== true).sort((a, b2) => (b2.lastSeen ?? 0) - (a.lastSeen ?? 0))[0];
     if (latest) {
       session = latest;
       viaFallback = true;
-      resolvedConversationId = conversationIdForSession(latest.id) ?? conversationId2;
     } 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" }));
@@ -56378,7 +56313,7 @@ function handlePluginStatus(conversationId2, res, deps, fallbackLatest = false)
   res.writeHead(200, { "content-type": "application/json" });
   res.end(JSON.stringify({
     ok: true,
-    conversationId: resolvedConversationId,
+    conversationId: conversationId2,
     fallback: viaFallback || void 0,
     label: session.meta.label ?? null,
     pluginAgent: session.metadata.pluginAgent ?? null,
@@ -56413,12 +56348,8 @@ async function handlePluginTool(payload, res, deps) {
   const entry = conversations.get(conversationId2);
   const session = entry ? peekSession(entry.sessionId) : void 0;
   if (!entry || !session) {
-    deps.log("warn", `[plugin] tool "${tool}" rejected for conversation ${conversationId2}: ${entry ? "id registered but session not resident in this proxy instance" : "id never registered (stale shim session id after host resume?)"}`);
     res.writeHead(404, { "content-type": "application/json" });
-    res.end(JSON.stringify({
-      ok: false,
-      error: !entry ? "unknown plugin conversation (no model request has arrived with this conversation id yet)" : "unknown plugin conversation (id registered but its session is not resident in this proxy instance \u2014 a fresh model request re-binds it)"
-    }));
+    res.end(JSON.stringify({ ok: false, error: "unknown plugin conversation (no model request has arrived with this conversation id yet)" }));
     return;
   }
   if (!isProxyToolFor(tool, session, deps.config)) {
@@ -59283,14 +59214,13 @@ ${bodyBuffer.toString("utf8")}`);
       await withSessionLock(session, async () => {
         const runPrepare = () => {
           const cs2 = resolveCompress(opts.routes, route?.rewrittenUrl, parsed.model, opts.compress);
-          const reasoningCfg = cs2.reasoning;
           const keepRecent = cs2.stripImagesKeepRecent ?? DEFAULT_STRIP_IMAGES_KEEP_RECENT;
           const stripped = cs2.stripImages ? stripHistoricalImages(parsed, protocol, keepRecent) : { body: parsed, removed: 0 };
           if (opts.debug && stripped.removed > 0) {
             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) : protocol === "openai" ? prepareOpenai(work, req, opts, core, reqConfig, reqPrompts, log2, session, pluginMode, nativeWindow) : 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);
         };
         const isCodexCompactTrigger = protocol === "responses" && !responsesCompact && isCodexClient(req.headers) && hasCompactionTrigger(parsed.input);
         if (isCodexCompactTrigger) {
@@ -59378,13 +59308,6 @@ ${bodyBuffer.toString("utf8")}`);
   }
 }
 var ACP_TAG_MARK = "<acp ";
-function withReasoningDrop(msgs, reasoning, log2, sessionId) {
-  const out = dropCompressReasoning(msgs, reasoning);
-  if (out.length !== msgs.length) {
-    log2("info", `[${sessionId}] compress-reasoning: dropped ${msgs.length - out.length} reasoning message(s) from closed compress turns (#651)`);
-  }
-  return out;
-}
 function stripKernelSummaries(messages, state) {
   const carried = /* @__PURE__ */ new Set();
   for (const b2 of state.blocks) {
@@ -59486,13 +59409,12 @@ function effectiveTokenCount(session, msgs) {
   if (!session.metadata.anonymousPrefixAffinity) return 0;
   return estimateCoreMessagesUpper(msgs);
 }
-function prepareAnthropic(parsed, req, opts, core, config, prompts, log2, session, pluginMode, reasoning) {
+function prepareAnthropic(parsed, req, opts, core, config, prompts, log2, session, pluginMode) {
   const sessionId = session.id;
   const stream2 = parsed.stream === true;
   ++session.stats.requests;
   session.hostCreditTokens = 0;
   const injectTools = opts.compress.injectTool && !pluginMode;
-  const stripReasoning = (msgs) => withReasoningDrop(msgs, reasoning, log2, sessionId);
   if (isAutoModeClassifier(parsed)) {
     log2("info", `[${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 };
@@ -59530,7 +59452,7 @@ function prepareAnthropic(parsed, req, opts, core, config, prompts, log2, sessio
     log2("info", diagTagSummary(turn.messages, sessionId, "text-only"));
     const willInjectNudge = opts.compress.injectNudge && !!turn.nudge && (turn.nudge.shouldInject || emergencyNudge(turn.nudge));
     log2("info", diagNudge(turn, sessionId, tokenCount, config.modelContextLimit, parsed.model, willInjectNudge));
-    processedMessages = stripReasoning(stripKernelSummaries(turn.messages, turn.state));
+    processedMessages = stripKernelSummaries(turn.messages, turn.state);
     applyCompactionArchive(session, activeBefore, new Set(msgs.map((m2) => m2.id)), log2);
     reapOrphanBlocks(session, msgs, deactivateBlock);
     rebuiltMessages = coreToAnthropic(processedMessages, cacheControls);
@@ -59612,13 +59534,12 @@ 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, nativeWindow) {
   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);
   let openaiOutboundSystem;
   let processedMessages = [];
   let originalMessages = [];
@@ -59656,7 +59577,7 @@ function prepareOpenai(parsed, req, opts, core, config, prompts, log2, session,
     log2("info", diagTagSummary(turn.messages, sessionId, "text-only"));
     const willInjectNudge = opts.compress.injectNudge && !!turn.nudge && shouldInject && (turn.nudge.shouldInject || emergencyNudge(turn.nudge));
     log2("info", diagNudge(turn, sessionId, tokenCount, config.modelContextLimit, parsed.model, willInjectNudge));
-    processedMessages = stripReasoning(stripKernelSummaries(turn.messages, turn.state));
+    processedMessages = stripKernelSummaries(turn.messages, turn.state);
     applyCompactionArchive(session, activeBefore, new Set(msgs.map((m2) => m2.id)), log2);
     reapOrphanBlocks(session, msgs, deactivateBlock);
     rebuiltMessages = systemToUser(coreToOpenai(processedMessages));
@@ -59696,12 +59617,11 @@ function prepareOpenai(parsed, req, opts, core, config, prompts, log2, session,
   markDirty(session);
   return { body: JSON.stringify(rebuilt), session, processedMessages, originalMessages, protocol: "openai", stream: stream2, compressInjected: injectTools, pluginMode, nudge, prompts, openaiSystemText, renderTags: "text-only" };
 }
-function prepareResponses(parsed, req, opts, core, config, prompts, log2, session, identity, pluginMode, upstreamOrigin, nativeWindow, reasoning) {
+function prepareResponses(parsed, req, opts, core, config, prompts, log2, session, identity, pluginMode, upstreamOrigin, nativeWindow) {
   const sessionId = session.id;
   const stream2 = parsed.stream === true;
   ++session.stats.requests;
   session.hostCreditTokens = 0;
-  const stripReasoning = (msgs) => withReasoningDrop(msgs, reasoning, log2, sessionId);
   if (reconcileNativeCompactionBoundary(session)) {
     log2("info", `[${sessionId}] reconciled ACP state after native Responses compact boundary`);
   }
@@ -59766,7 +59686,7 @@ function prepareResponses(parsed, req, opts, core, config, prompts, log2, sessio
     log2("info", diagTagSummary(turn.messages, sessionId, "text-only"));
     const willInjectNudge = opts.compress.injectNudge && !!turn.nudge && shouldInject && !isCompactionTrigger && (turn.nudge.shouldInject || emergencyNudge(turn.nudge));
     log2("info", diagNudge(turn, sessionId, tokenCount, config.modelContextLimit, parsed.model, willInjectNudge));
-    processedMessages = repairResponsesAssistantOrdering(stripReasoning(stripKernelSummaries(turn.messages, turn.state)), originalMessages);
+    processedMessages = repairResponsesAssistantOrdering(stripKernelSummaries(turn.messages, turn.state), originalMessages);
     reapOrphanBlocks(session, msgs, deactivateBlock);
     rebuiltInput = patchResponsesInput(projection, processedMessages);
     const forgedSummaries = echoReplaced ? [] : session.metadata.codexForgedSummaries ?? [];
@@ -64401,7 +64321,6 @@ function resolveProxyOrigin() {
 var TOOL_TIMEOUT_MS = 6e4;
 var CONVERSATION_FROM_ENV = process.env.CLAUDE_CODE_SESSION_ID?.trim() || process.env.BILI_CONVERSATION_ID?.trim() || void 0;
 var IDENTITY_BINDING = Boolean(process.env.CLAUDE_CODE_SESSION_ID?.trim());
-var ORPHAN_ADOPT = IDENTITY_BINDING && process.env.BILI_MCP_NO_ORPHAN_ADOPT !== "1";
 var manifestTools = [];
 var conversationId = CONVERSATION_FROM_ENV;
 var registered = false;
@@ -64434,45 +64353,21 @@ function ensureManifest() {
   return manifestPromise;
 }
 async function forwardTool(tool, args, timeoutMs = TOOL_TIMEOUT_MS) {
-  for (let attempt = 0; ; attempt++) {
-    let res;
-    try {
-      res = await fetch(`${resolveProxyOrigin()}/__bili/plugin/tool`, {
-        method: "POST",
-        headers: { "content-type": "application/json" },
-        body: JSON.stringify({ conversationId, tool, args }),
-        signal: AbortSignal.timeout(timeoutMs)
-      });
-    } catch (err2) {
-      if (err2 instanceof Error && err2.name === "TimeoutError") throw new Error(`tool forward timed out after ${timeoutMs}ms: ${tool}`);
-      throw err2;
-    }
-    const data = await res.json();
-    if (res.ok && data.ok) return data.result ?? "";
-    if (res.status === 404 && attempt === 0 && ORPHAN_ADOPT && conversationId && typeof data.error === "string" && data.error.includes("no model request has arrived")) {
-      if (await adoptLatestActiveConversation()) continue;
-    }
-    throw new Error(data.error ?? `tool forward failed: ${res.status}`);
-  }
-}
-async function adoptLatestActiveConversation() {
+  let res;
   try {
-    const res = await fetch(
-      `${resolveProxyOrigin()}/__bili/plugin/status?conversationId=${encodeURIComponent(conversationId ?? "")}&fallback=latest`,
-      { signal: AbortSignal.timeout(5e3) }
-    );
-    if (!res.ok) return false;
-    const data = await res.json();
-    if (data.ok !== true || data.fallback !== true || !data.conversationId || data.conversationId === conversationId) return false;
-    process.stderr.write(
-      `[bili-mcp] conversation id no longer known by the proxy (host resumed its session?); adopting latest active conversation ${data.conversationId}
-`
-    );
-    conversationId = data.conversationId;
-    return true;
-  } catch {
-    return false;
+    res = await fetch(`${resolveProxyOrigin()}/__bili/plugin/tool`, {
+      method: "POST",
+      headers: { "content-type": "application/json" },
+      body: JSON.stringify({ conversationId, tool, args }),
+      signal: AbortSignal.timeout(timeoutMs)
+    });
+  } catch (err2) {
+    if (err2 instanceof Error && err2.name === "TimeoutError") throw new Error(`tool forward timed out after ${timeoutMs}ms: ${tool}`);
+    throw err2;
   }
+  const data = await res.json();
+  if (!res.ok || !data.ok) throw new Error(data.error ?? `tool forward failed: ${res.status}`);
+  return data.result ?? "";
 }
 var ERR_TOOL = -32602;
 async function handleMessage(msg) {