billion-context 0.1.127 → 0.1.128

package/dist/index.js

@@ -60741,6 +60741,9 @@ var MARKER_TAIL = /(?:^|\n)[^\x00-\x7F](?:[ \t])?(?:\[ACP|\[AC|\[A|\[)?$/u;
 function mayStartMarkerLine(s3) {
   return s3.includes("[ACP]") || MARKER_TAIL.test(s3);
 }
+function stripMarkerLines(text) {
+  return text.replace(MARKER_LINE, "");
+}
 function stripAcpTags(text) {
   let out = text;
   for (; ; ) {
@@ -61168,6 +61171,19 @@ function executeAnthropicProxyTool(toolName, args, ctx) {
 function refNum3(ref) {
   return Number(ref.replace(/\D/g, "")) || 0;
 }
+function compressibleSpanHint(state) {
+  const refs = Object.keys(state.messageRefs?.byRef ?? {});
+  const highest = refs.reduce((m2, r) => Math.max(m2, r.startsWith("m") ? Number(r.slice(1)) || 0 : 0), 0);
+  const boundary = state.blocks.reduce((m2, b2) => b2.active && b2.endRef?.startsWith("m") ? Math.max(m2, Number(b2.endRef.slice(1)) || 0) : m2, 0);
+  const fmt = (n) => `m${String(n).padStart(5, "0")}`;
+  if (highest <= boundary) {
+    const actives = state.blocks.filter((b2) => b2.active).map((b2) => b2.blockId);
+    const span = actives.length >= 2 ? ` (e.g. startId ${actives[0]}, endId ${actives[actives.length - 1]})` : "";
+    return ` No raw refs are directly compressible right now \u2014 compress a run of ACTIVE blocks instead${span}: fold their summaries into one higher-tier block. acp_status lists the current active blocks.`;
+  }
+  const covered = boundary > 0 ? ` (everything up to ${fmt(boundary)} is already inside active blocks)` : "";
+  return ` Live compressible refs: ${fmt(boundary + 1)}\u2013${fmt(highest)}${covered}. Retry NOW in this same turn with startId/endId inside that span.`;
+}
 var M_REF_NUM_RE = /^m0*(\d{1,7})$/i;
 function normalizeRangeOrder(ranges) {
   let swapped = 0;
@@ -61192,7 +61208,7 @@ var FAIL_STREAK_CAP = 5;
 function normalizedSpecKey(ranges) {
   return ranges.map((r) => refNum3(r.startRef) <= refNum3(r.endRef) ? `${r.startRef}..${r.endRef}` : `${r.endRef}..${r.startRef}`).sort().join(",");
 }
-function recordCompressFailure(session, key) {
+function recordCompressFailure(session, key, repeatAdvice) {
   if (!key) return "";
   const prev = session.metadata[FAIL_STREAK_KEY];
   const keys = Array.isArray(prev) ? prev.filter((k2) => typeof k2 === "string") : [];
@@ -61202,7 +61218,8 @@ function recordCompressFailure(session, key) {
   session.metadata[FAIL_STREAK_KEY] = keys;
   markDirty(session);
   if (occurrences < 2) return "";
-  return ` [Repeat-failure guard: you have now requested this exact range set ${occurrences} time(s) in this session and it keeps failing with the same error. Repeating it deterministically fails the same way until the visible context changes \u2014 do NOT re-issue it. Call acp_status first and pick from its CURRENT compressible ranges, or extend your range(s) to cover more adjacent messages.]`;
+  const advice = repeatAdvice ?? "Call acp_status first and pick from its CURRENT compressible ranges, or extend your range(s) to cover more adjacent messages.";
+  return ` [Repeat-failure guard: you have now requested this exact spec ${occurrences} time(s) in this session and it keeps failing with the same error. Repeating it deterministically fails the same way until the visible context changes \u2014 do NOT re-issue it. ${advice}]`;
 }
 function clearCompressFailures(session) {
   if (session.metadata[FAIL_STREAK_KEY] !== void 0) {
@@ -61210,15 +61227,42 @@ function clearCompressFailures(session) {
     markDirty(session);
   }
 }
+function currentRefsSnapshot(ctx) {
+  let loId = "";
+  let hiId = "";
+  let loN = Infinity;
+  let hiN = -1;
+  for (const ref of Object.keys(ctx.session.state.messageRefs?.byRef ?? {})) {
+    const m2 = M_REF_NUM_RE.exec(ref);
+    if (!m2) continue;
+    const n = Number(m2[1]);
+    if (n < loN) {
+      loN = n;
+      loId = ref;
+    }
+    if (n > hiN) {
+      hiN = n;
+      hiId = ref;
+    }
+  }
+  if (!loId || !hiId) return "";
+  const activeBlocks2 = ctx.session.state.blocks.filter((b2) => b2.active).length;
+  return ` [Current context: ${ctx.messages.length} visible message(s), refs ${loId}\u2013${hiId}, ${activeBlocks2} active block(s). Refs restart at m00001 after a session-generation change \u2014 request only refs inside this span, or call acp_status for exact compressible ranges.]`;
+}
 function applyRanges(parsed, ctx) {
   const { ranges, diagnostics } = parsed;
   if (ranges.length === 0) {
     ctx.log("[acp-proxy: compress call had no valid ranges; nothing compressed.]");
-    const reasons = diagnostics.invalidReasons?.slice(0, 8).map((r) => r.length > 200 ? r.slice(0, 200) + "..." : r) ?? [];
-    const why = reasons.length > 0 ? ` Rejected entries:
-${reasons.map((r) => `- ${r}`).join("\n")}` : "";
-    return `[Compression FAILED: no valid ranges parsed (kind=${diagnostics.kind}, dropped=${diagnostics.invalidItems}).${why}
- compress requires a non-empty 'content' array where each element is EITHER an object {startId, endId, summary} OR one line-form string whose first line is 'mNNNNN\u2013mNNNNN optional topic' with the summary markdown on the following lines (a separate summary-only element right after a bare header line is also accepted). startId/endId are mNNNNN message refs from the conversation. Re-issue the compress call with a valid content array.]`;
+    const rawReasons = diagnostics.invalidReasons ?? [];
+    for (const reason of rawReasons) ctx.log(`[acp-proxy: rejected entry: ${reason}]`);
+    const reasons = rawReasons.slice(0, 3).map((r) => r.length > 160 ? r.slice(0, 160) + "..." : r);
+    const why = reasons.length > 0 ? ` Rejected entries: ${reasons.join(" | ")}.` : "";
+    const guard = recordCompressFailure(
+      ctx.session,
+      `parse:${diagnostics.kind}:${diagnostics.invalidItems}:${rawReasons.slice(0, 3).join("|")}`,
+      "Fix the argument shape against the format below instead of re-issuing the same malformed call."
+    );
+    return `[Compression FAILED: no valid ranges parsed (kind=${diagnostics.kind}, dropped=${diagnostics.invalidItems}).${why} compress requires a non-empty 'content' array where each element is EITHER an object {startId, endId, summary} OR one line-form string whose first line is 'mNNNNN\u2013mNNNNN optional topic' with the summary markdown on the following lines (a separate summary-only element right after a bare header line is also accepted). startId/endId are mNNNNN message refs from the conversation (call acp_status to see current refs).${compressibleSpanHint(ctx.session.state)} Re-issue the compress call with a valid content array.${guard}]`;
   }
   const revs = reversedRanges(ranges);
   const swappedRanges = normalizeRangeOrder(ranges);
@@ -61261,7 +61305,8 @@ ${reasons.map((r) => `- ${r}`).join("\n")}` : "";
       const errs = r.errors.join("; ") || "no blocks created";
       const revNote = revs.length > 0 ? ` Note: startId > endId in range(s) ${revs.map((rg) => `${rg.startRef}\u2192${rg.endRef}`).join(", ")} \u2014 your refs were reversed; they were normalized to ascending order before evaluation, so check your ref order.` : "";
       ctx.log(`[acp-proxy: compress FAILED ${detail} \u2192 0 blocks. ${errs}${revs.length > 0 ? " [reversed refs]" : ""}]`);
-      return `[Compression FAILED: ${errs}${revNote}${recordCompressFailure(ctx.session, normalizedSpecKey(ranges))}]`;
+      const spanHint = ctx.session.state.blocks.some((b2) => b2.active) ? compressibleSpanHint(ctx.session.state) : "";
+      return `[Compression FAILED: ${errs}${revNote}${currentRefsSnapshot(ctx)}${recordCompressFailure(ctx.session, normalizedSpecKey(ranges))}${spanHint}]`;
     }
     clearCompressFailures(ctx.session);
     const preContext = ctx.session.stats.lastInputTokens;
@@ -63060,9 +63105,14 @@ function dumpRejectedBody(status, sessionId, body) {
 }
 
 // src/strict-echo.ts
-function isStrictReasoningEcho(session, upstreamOrigin) {
+function isStrictReasoningEcho(session, upstreamOrigin, model) {
   if (session.metadata.strictReasoningEcho === true) return true;
-  return upstreamOrigin !== void 0 && /deepseek/i.test(upstreamOrigin);
+  if (upstreamOrigin !== void 0 && /deepseek/i.test(upstreamOrigin)) return true;
+  return typeof model === "string" && model.length > 0 && /deepseek/i.test(model);
+}
+function modelIdOf(body) {
+  const m2 = body?.model;
+  return typeof m2 === "string" ? m2 : void 0;
 }
 function normalizeStrictEchoReasoning(messages, enabled, log2, sessionId) {
   if (!enabled) return messages;
@@ -63283,6 +63333,7 @@ async function* runCompressLoop(upstream, ctx, requestBody, requestOptions, adap
   let truncationRetried = false;
   let continuationRetried = false;
   let degenerateRetried = false;
+  const seenMarkers = /* @__PURE__ */ new Set();
   const fetchUpstream = (body) => fetchWithRetry(
     requestOptions.url,
     {
@@ -63510,7 +63561,15 @@ async function* runCompressLoop(upstream, ctx, requestBody, requestOptions, adap
           }
           const result = await withSessionLock(ctx.session, () => executeProxyTool(call.name, parsedArgs, ctx, call.callId));
           proxyResults.push({ name: call.name, callId: call.callId, result, arguments: call.arguments, signature: call.signature });
-          if (ctx.visibilityMarkers !== false) yield adapter.emitMarker(call.name, result);
+          if (ctx.visibilityMarkers !== false) {
+            const markerKey = `${call.name}\0${result}`;
+            if (seenMarkers.has(markerKey)) {
+              ctx.log(`[acp-loop] suppressed duplicate ${call.name} status marker (identical failure repeated this request)`);
+            } else {
+              seenMarkers.add(markerKey);
+              yield adapter.emitMarker(call.name, result);
+            }
+          }
         } else {
           realToolCalls.push(call);
           realCalls += 1;
@@ -63653,7 +63712,7 @@ async function* runCompressLoop(upstream, ctx, requestBody, requestOptions, adap
       ctx.log(`[acp-loop] round ${round}: proxy tool executed; re-requesting so the model sees the result`);
       if (signal?.aborted) break;
       let newBody = adapter.buildRequest(coreMessages, systemPrompt, requestBody);
-      newBody = normalizeStrictEchoBody(newBody, isStrictReasoningEcho(ctx.session, strictEchoOrigin), (level, msg) => log(level, `[acp-loop] ${msg}`), ctx.session.id ?? "unknown");
+      newBody = normalizeStrictEchoBody(newBody, isStrictReasoningEcho(ctx.session, strictEchoOrigin, modelIdOf(requestBody)), (level, msg) => log(level, `[acp-loop] ${msg}`), ctx.session.id ?? "unknown");
       if (process.env.ACP_DUMP_BODY === "1") {
         try {
           const fs16 = await import("fs");
@@ -66370,6 +66429,41 @@ function stripAcpPanelMessages(messages) {
   }
   return stripped;
 }
+function stripAcpStatusMarkers(messages) {
+  if (!Array.isArray(messages)) return 0;
+  let stripped = 0;
+  for (const rec of messages) {
+    if (rec === null || typeof rec !== "object") continue;
+    const m2 = rec;
+    if (m2.role !== "user" && m2.role !== "assistant") continue;
+    const content = m2.content;
+    if (typeof content === "string") {
+      const out = stripMarkerLines(content);
+      if (out !== content) {
+        m2.content = out.trim().length > 0 ? out : " ";
+        stripped += countMarkerLines(content);
+      }
+    } else if (Array.isArray(content)) {
+      for (const part of content) {
+        if (part === null || typeof part !== "object") continue;
+        const p2 = part;
+        const text = p2.text;
+        if (typeof text !== "string") continue;
+        const out = stripMarkerLines(text);
+        if (out !== text) {
+          p2.text = out.trim().length > 0 ? out : " ";
+          stripped += countMarkerLines(text);
+        }
+      }
+    }
+  }
+  return stripped;
+}
+function countMarkerLines(text) {
+  let n = 0;
+  for (const _2 of text.matchAll(MARKER_LINE)) n++;
+  return n;
+}
 function stripAcpPanelResponsesInput(input) {
   if (!Array.isArray(input)) return 0;
   let stripped = 0;
@@ -71126,7 +71220,7 @@ function prepareAnthropic(parsed, req, opts, core, config, prompts, surface, log
   const stream2 = parsed.stream === true;
   ++session.stats.requests;
   const injectTools = opts.compress.injectTool && !pluginMode;
-  const stripReasoning = (msgs) => withReasoningDrop(msgs, reasoning, log2, sessionId, isStrictReasoningEcho(session, upstreamOrigin));
+  const stripReasoning = (msgs) => withReasoningDrop(msgs, reasoning, log2, sessionId, isStrictReasoningEcho(session, upstreamOrigin, modelIdOf(parsed)));
   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, surface };
@@ -71141,6 +71235,10 @@ function prepareAnthropic(parsed, req, opts, core, config, prompts, surface, log
   if (strippedPanels > 0) {
     log2("info", `[${sessionId}] stripped ${strippedPanels} ACP panel message(s) before projection (UI-only, issue #359)`);
   }
+  const strippedMarkerLines = stripAcpStatusMarkers(parsed.messages);
+  if (strippedMarkerLines > 0) {
+    log2("info", `[${sessionId}] stripped ${strippedMarkerLines} ACP status marker line(s) from incoming history (ephemeral proxy status, issue #1029)`);
+  }
   try {
     const { msgs, cacheControls } = anthropicToCore(parsed);
     originalMessages = msgs;
@@ -71209,7 +71307,7 @@ function prepareOpenai(parsed, req, opts, core, config, prompts, surface, log2,
   const stream2 = parsed.stream === true;
   ++session.stats.requests;
   let openaiSystemText = "";
-  const stripReasoning = (msgs) => withReasoningDrop(msgs, reasoning, log2, sessionId, isStrictReasoningEcho(session, upstreamOrigin));
+  const stripReasoning = (msgs) => withReasoningDrop(msgs, reasoning, log2, sessionId, isStrictReasoningEcho(session, upstreamOrigin, modelIdOf(parsed)));
   let openaiOutboundSystem;
   let processedMessages = [];
   let originalMessages = [];
@@ -71224,6 +71322,10 @@ function prepareOpenai(parsed, req, opts, core, config, prompts, surface, log2,
   if (strippedPanels > 0) {
     log2("info", `[${sessionId}] stripped ${strippedPanels} ACP panel message(s) before projection (UI-only, issue #359)`);
   }
+  const strippedMarkerLines = stripAcpStatusMarkers(parsed.messages);
+  if (strippedMarkerLines > 0) {
+    log2("info", `[${sessionId}] stripped ${strippedMarkerLines} ACP status marker line(s) from incoming history (ephemeral proxy status, issue #1029)`);
+  }
   try {
     const { msgs, systemText } = openaiToCore(parsed);
     openaiSystemText = systemText;
@@ -71283,7 +71385,7 @@ function prepareOpenai(parsed, req, opts, core, config, prompts, surface, log2,
     log2("warn", `[${sessionId}] kernel transform failed, forwarding unchanged: ${String(err2)}`);
     processedMessages = [];
   }
-  rebuiltMessages = normalizeStrictEchoReasoning(rebuiltMessages, isStrictReasoningEcho(session, upstreamOrigin), log2, sessionId);
+  rebuiltMessages = normalizeStrictEchoReasoning(rebuiltMessages, isStrictReasoningEcho(session, upstreamOrigin, modelIdOf(parsed)), log2, sessionId);
   const rebuilt = { ...parsed, messages: rebuiltMessages, tools: toolsOut };
   warnReasoningPairs(rebuiltMessages, log2, sessionId);
   clampOutgoingOutput(rebuilt, typeof parsed.max_completion_tokens === "number" ? "max_completion_tokens" : "max_tokens", { systemText: openaiSystemText, tools: toolsOut, processedMessages, lastInputTokens: session.stats.lastInputTokens, nativeWindow, imageTokens: imageTokensInParsedBody("openai", rebuilt, imageBillingFor(opts, billingUpstream ?? upstreamOrigin)) }, sessionId, log2);
@@ -71420,7 +71522,7 @@ function prepareResponses(parsed, req, opts, core, config, prompts, surface, log
   const sessionId = session.id;
   const stream2 = parsed.stream === true;
   ++session.stats.requests;
-  const stripReasoning = (msgs) => withReasoningDrop(msgs, reasoning, log2, sessionId, isStrictReasoningEcho(session, upstreamOrigin));
+  const stripReasoning = (msgs) => withReasoningDrop(msgs, reasoning, log2, sessionId, isStrictReasoningEcho(session, upstreamOrigin, modelIdOf(parsed)));
   if (reconcileNativeCompactionBoundary(session)) {
     log2("info", `[${sessionId}] reconciled ACP state after native Responses compact boundary`);
   }
@@ -71454,6 +71556,10 @@ function prepareResponses(parsed, req, opts, core, config, prompts, surface, log
   if (strippedPanels > 0) {
     log2("info", `[${sessionId}] stripped ${strippedPanels} ACP panel message(s) before projection (UI-only, issue #359)`);
   }
+  const strippedMarkerLines = stripAcpStatusMarkers(parsed.input);
+  if (strippedMarkerLines > 0) {
+    log2("info", `[${sessionId}] stripped ${strippedMarkerLines} ACP status marker line(s) from incoming history (ephemeral proxy status, issue #1029)`);
+  }
   const shouldInject = opts.compress.injectTool;
   const injectTools = shouldInject && !pluginMode;
   const isCompactionTrigger = hasCompactionTrigger(parsed.input);