billion-context 0.1.86 → 0.1.87-pr.397.392

package/dist/index.js

@@ -43476,10 +43476,6 @@ function defaultCountTokens(text) {
   const cjkCount = cjk?.length ?? 0;
   return cjkCount + Math.ceil((text.length - cjkCount) / 4);
 }
-function estimateTokensFast(text) {
-  if (!text) return 0;
-  return Math.ceil(text.length / 4);
-}
 var COMPRESS_PHILOSOPHY = `Compression Philosophy:
 - All compression serves the primary task, but be frugal.
 - Context capacity is precious. Save context by compressing consumed outputs, not by avoiding tools.
@@ -46339,7 +46335,7 @@ function formatTokens2(n) {
 }
 function pct(n, total) {
   if (n <= 0 || total <= 0) return 0;
-  return Math.max(1, Math.round(n / total * 100));
+  return Math.round(n / total * 100);
 }
 function numericPart2(blockId) {
   const match = /^b(\d+)$/.exec(blockId);
@@ -46378,12 +46374,18 @@ function collectVisible(messages, state, countTokens) {
     if (block.active) summaryTokens += summaryTokensOf(block, countTokens);
   }
   const visible = [];
+  const toolCallNames = /* @__PURE__ */ new Map();
+  for (const message of messages) {
+    if (message.contentType === "tool-call" && message.toolCallId && message.toolName) {
+      toolCallNames.set(message.toolCallId, message.toolName);
+    }
+  }
   messages.forEach((message, index) => {
     if (coveredIds.has(message.id)) return;
     const ref = refForRaw(state.messageRefs, message.id);
     if (!ref) return;
     const tokens = countTokens(message.text ?? "");
-    const tool = message.toolName ?? "text";
+    const tool = isToolMessage(message) ? message.toolName ?? (message.toolCallId ? toolCallNames.get(message.toolCallId) : void 0) ?? "tool" : "text";
     if (tokens > 0) visible.push({ ref, tokens, tool, index });
   });
   return { visible, summaryTokens };
@@ -46471,6 +46473,17 @@ function renderUncompressedRanges(visible) {
     const m2 = ref.match(/\d+/);
     return m2 ? parseInt(m2[0], 10) : 0;
   };
+  const dominantTool = (toolTokens) => {
+    let best = "text";
+    let bestN = -1;
+    for (const [tool, n] of toolTokens) {
+      if (n > bestN) {
+        best = tool;
+        bestN = n;
+      }
+    }
+    return best;
+  };
   const merged = [];
   for (const m2 of visible) {
     const num3 = refNum22(m2.ref);
@@ -46479,13 +46492,15 @@ function renderUncompressedRanges(visible) {
       last.endRef = m2.ref;
       last.count += 1;
       last.tokens += m2.tokens;
+      last.toolTokens.set(m2.tool, (last.toolTokens.get(m2.tool) ?? 0) + m2.tokens);
     } else {
-      merged.push({ startRef: m2.ref, endRef: m2.ref, startNum: num3, count: 1, tokens: m2.tokens, tool: m2.tool });
+      const toolTokens = /* @__PURE__ */ new Map([[m2.tool, m2.tokens]]);
+      merged.push({ startRef: m2.ref, endRef: m2.ref, startNum: num3, count: 1, tokens: m2.tokens, toolTokens });
     }
   }
   for (const r of merged.slice(0, 30)) {
     const range = r.count === 1 ? r.startRef : `${r.startRef}\u2013${r.endRef}`;
-    lines.push(`  ${range}  (${r.count} msgs, ${formatTokens2(r.tokens)}${r.count > 1 ? ` (${Math.round(r.tokens / r.count)}/msg)` : ""}) ${r.tool}`);
+    lines.push(`  ${range}  (${r.count} msgs, ${formatTokens2(r.tokens)}${r.count > 1 ? ` (${Math.round(r.tokens / r.count)}/msg)` : ""}) ${dominantTool(r.toolTokens)}`);
   }
   if (merged.length > 30) {
     lines.push(`  ... and ${merged.length - 30} more ranges`);
@@ -50433,6 +50448,50 @@ function withStagedCompressGuidance(text) {
   return text + STAGED_COMPRESS_GUIDANCE;
 }
 
+// src/acp-status.ts
+function handleAcpStatus(args, ctx) {
+  const scope = typeof args.scope === "string" ? args.scope : void 0;
+  const view = typeof args.view === "string" ? args.view : void 0;
+  const tool = typeof args.tool === "string" ? args.tool : void 0;
+  const sort = typeof args.sort === "string" ? args.sort : void 0;
+  const limit = typeof args.limit === "number" ? args.limit : void 0;
+  const base = buildStatusReport(ctx.session.state, ctx.messages, defaultCountTokens, { scope, view, tool, sort, limit });
+  if (scope) return base;
+  const extra = [];
+  try {
+    const turn = ctx.core.processTurn({
+      messages: ctx.messages,
+      state: ctx.session.state,
+      config: ctx.config,
+      tokenCount: ctx.session.stats.lastInputTokens,
+      renderTags: "none"
+    });
+    const nudge = turn.nudge;
+    if (nudge) {
+      extra.push("");
+      extra.push(nudge.shouldInject ? `Nudge: ACTIVE \u2014 ${nudge.reason}` : `Nudge: idle \u2014 ${nudge.reason}`);
+      const ranges = viableRanges(nudge.compressibleRanges);
+      const protectedRanges = nudge.protectedRanges ?? [];
+      if (ranges.length > 0 || protectedRanges.length > 0) {
+        extra.push("");
+        extra.push(formatRanges(ranges, protectedRanges));
+      }
+    }
+  } catch {
+  }
+  const archive = preCompactionArchiveOf(ctx.session);
+  const archivedIds = Object.keys(archive);
+  if (archivedIds.length > 0) {
+    extra.push("");
+    extra.push(`PRE-COMPACTION ARCHIVE \u2014 ${archivedIds.length} block(s): content was replaced by the client's native compaction summary, so it is no longer in the session history and decompress is unavailable.`);
+    for (const id of archivedIds) {
+      extra.push(`  ${id} \u2014 ${archive[id].reason}`);
+    }
+  }
+  return extra.length > 0 ? `${base}
+${extra.join("\n")}` : base;
+}
+
 // src/decompress-shared.ts
 import { mkdirSync as mkdirSync4, unlinkSync as unlinkSync2, writeFileSync as writeFileSync3 } from "fs";
 import { dirname as dirname2, join as join2 } from "path";
@@ -50754,7 +50813,7 @@ function executeAnthropicProxyTool(toolName, args, ctx) {
 ${lines.join("\n\n")}`;
   }
   if (toolName === "acp_status") {
-    return buildStatusReport(ctx.session.state, ctx.messages, estimateTokensFast);
+    return handleAcpStatus(args, ctx);
   }
   return `[Unknown proxy tool: ${toolName}]`;
 }
@@ -50988,6 +51047,12 @@ TASK: The conversation segment below (messages ${startRef}\u2013${endRef}) must
   }
 }
 var ABORTED_FAILURE = { kind: "aborted", detail: "the client disconnected during preflight compression" };
+function relaxedConfig(config) {
+  return { ...config, preserveRecentMessages: 0, preserveRecentTokens: 0 };
+}
+function noEmergencyTruncate(config) {
+  return { ...config, modelContextLimit: config.modelContextLimit * 100 };
+}
 async function preflightCompress(deps, messages) {
   const limit = deps.config.modelContextLimit;
   const result = { compressedRanges: 0, savedTokens: 0, payloadEstimate: estimateCoreMessages(messages) + (deps.imageFloor ?? 0) };
@@ -50997,6 +51062,9 @@ async function preflightCompress(deps, messages) {
   let currentTokens = deps.session.stats.lastInputTokens;
   let startTokens = -1;
   let failure;
+  let activeConfig = deps.config;
+  let relaxed = false;
+  const relaxedExhaustedDetail = `the payload still exceeds the window after folding everything compressible, including the soft-protected recent zone (last ${deps.config.preserveRecentMessages} messages + most recent user message), which was relaxed under overflow; hard protectedTools remain excluded. Raise the model context window or restart the session to recover.`;
   for (let round = 0; round < MAX_PREFLIGHT_ROUNDS; round++) {
     if (deps.signal?.aborted) {
       failure = ABORTED_FAILURE;
@@ -51005,7 +51073,7 @@ async function preflightCompress(deps, messages) {
     const turn = deps.core.processTurn({
       messages,
       state: deps.session.state,
-      config: deps.config,
+      config: noEmergencyTruncate(activeConfig),
       tokenCount: currentTokens,
       renderTags: "text-only"
     });
@@ -51016,7 +51084,13 @@ async function preflightCompress(deps, messages) {
     if (currentTokens < limit) break;
     const ranges = viableRanges(turn.nudge?.compressibleRanges ?? []);
     if (ranges.length === 0) {
-      failure = { kind: "exhausted", detail: "no compressible ranges remain in the conversation" };
+      if (!relaxed && result.payloadEstimate >= limit) {
+        activeConfig = relaxedConfig(deps.config);
+        relaxed = true;
+        deps.log("warn", "[preflight] no compressible ranges outside the protected recent zone; relaxing soft protection (preserveRecentMessages/Tokens -> 0) and retrying");
+        continue;
+      }
+      failure = { kind: "exhausted", detail: relaxed ? relaxedExhaustedDetail : "no compressible ranges remain in the conversation" };
       break;
     }
     const range = [...ranges].sort((a, b2) => refNum3(a.startRef) - refNum3(b2.startRef))[0];
@@ -51064,7 +51138,7 @@ async function preflightCompress(deps, messages) {
       if (!summary) continue;
       const ctx = {
         core: deps.core,
-        config: deps.config,
+        config: activeConfig,
         messages,
         session: deps.session,
         log: (msg) => deps.log("info", msg)
@@ -51089,7 +51163,10 @@ async function preflightCompress(deps, messages) {
     }
   }
   if (!failure && currentTokens >= limit) {
-    failure = { kind: "exhausted", detail: `the compress budget was exhausted after ${MAX_PREFLIGHT_ROUNDS} rounds` };
+    failure = {
+      kind: "exhausted",
+      detail: relaxed ? relaxedExhaustedDetail : `the compress budget was exhausted after ${MAX_PREFLIGHT_ROUNDS} rounds`
+    };
   }
   if (result.compressedRanges > 0) deps.session.stats.lastInputTokens = currentTokens;
   result.savedTokens = Math.max(0, startTokens - currentTokens);
@@ -51872,6 +51949,23 @@ function isLoopThinking(m2) {
 function stripLoopThinking(messages) {
   return messages.filter((m2) => !isLoopThinking(m2));
 }
+function canonicalArgs(args) {
+  try {
+    return JSON.stringify(sortKeys(JSON.parse(args)));
+  } catch {
+    return args;
+  }
+}
+function sortKeys(v2) {
+  if (Array.isArray(v2)) return v2.map(sortKeys);
+  if (v2 !== null && typeof v2 === "object") {
+    const src = v2;
+    const out = {};
+    for (const k2 of Object.keys(src).sort()) out[k2] = sortKeys(src[k2]);
+    return out;
+  }
+  return v2;
+}
 function executeProxyTool(toolName, args, ctx, callId) {
   if (toolName === "compress") {
     return applyRanges(parseCompressInput(args, callId), ctx);
@@ -51900,38 +51994,6 @@ ${lines.join("\n\n")}`;
   }
   return `[Unknown proxy tool: ${toolName}]`;
 }
-function handleAcpStatus(args, ctx) {
-  const scope = typeof args.scope === "string" ? args.scope : void 0;
-  const view = typeof args.view === "string" ? args.view : void 0;
-  const tool = typeof args.tool === "string" ? args.tool : void 0;
-  const sort = typeof args.sort === "string" ? args.sort : void 0;
-  const limit = typeof args.limit === "number" ? args.limit : void 0;
-  const base = buildStatusReport(ctx.session.state, ctx.messages, estimateTokensFast, { scope, view, tool, sort, limit });
-  if (scope) return base;
-  const nudge = ctx.nudge;
-  const ranges = nudge?.compressibleRanges ?? [];
-  const protectedRanges = nudge?.protectedRanges ?? [];
-  const archive = preCompactionArchiveOf(ctx.session);
-  const archivedIds = Object.keys(archive);
-  const extra = [];
-  if (nudge) {
-    extra.push("");
-    extra.push(nudge.shouldInject ? `Nudge: ACTIVE \u2014 ${nudge.reason}` : `Nudge: idle \u2014 ${nudge.reason}`);
-  }
-  if (ranges.length > 0 || protectedRanges.length > 0) {
-    extra.push("");
-    extra.push(formatRanges(ranges, protectedRanges));
-  }
-  if (archivedIds.length > 0) {
-    extra.push("");
-    extra.push(`PRE-COMPACTION ARCHIVE \u2014 ${archivedIds.length} block(s): content was replaced by the client's native compaction summary, so it is no longer in the session history and decompress is unavailable.`);
-    for (const id of archivedIds) {
-      extra.push(`  ${id} \u2014 ${archive[id].reason}`);
-    }
-  }
-  return extra.length > 0 ? `${base}
-${extra.join("\n")}` : base;
-}
 function recordUsage(ctx, usage, round) {
   const prompt = usage.inputTokens;
   const cached = usage.cachedTokens;
@@ -51974,6 +52036,7 @@ async function* runCompressLoop(upstream, ctx, requestBody, requestOptions, adap
       log("warn", `[acp-loop] upstream rejected replay (HTTP ${info.status}); retrying in ${info.delayMs}ms (attempt ${info.attempt}/${info.maxAttempts})${lc}`);
     }
   );
+  const failedSignatures = /* @__PURE__ */ new Set();
   try {
     for (let round = 1; round <= MAX_LOOP_ROUNDS; round++) {
       if (signal?.aborted) break;
@@ -52229,6 +52292,19 @@ async function* runCompressLoop(upstream, ctx, requestBody, requestOptions, adap
         yield adapter.emitCompletion({ finishReason: "length", usage });
         return;
       }
+      const failedMutating = proxyResults.filter(
+        (pr2) => (pr2.name === "compress" || pr2.name === "decompress") && pr2.result.includes("FAILED")
+      );
+      if (reRequest && failedMutating.length > 0 && failedMutating.length === proxyResults.length) {
+        const repeated = failedMutating.some((pr2) => failedSignatures.has(`${pr2.name}\0${canonicalArgs(pr2.arguments)}`));
+        if (repeated) {
+          ctx.log(`[acp-loop] round ${round}: model re-submitted an already-failed compress with identical arguments; stopping after ${round} round(s) instead of ${MAX_LOOP_ROUNDS}`);
+          log("warn", `[acp-loop] repeated identical compress failure at round ${round}; breaking early (#156)`);
+          yield adapter.emitCompletion({ finishReason: "length", usage });
+          return;
+        }
+      }
+      for (const pr2 of failedMutating) failedSignatures.add(`${pr2.name}\0${canonicalArgs(pr2.arguments)}`);
       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);
@@ -53755,7 +53831,7 @@ function executeProxyTool2(toolName, args, ctx) {
 ${lines.join("\n\n")}`;
   }
   if (toolName === "acp_status") {
-    return buildStatusReport(ctx.session.state, ctx.messages, estimateTokensFast);
+    return handleAcpStatus(args, ctx);
   }
   return `[Unknown proxy tool: ${toolName}]`;
 }
@@ -54334,14 +54410,14 @@ var MIN_TAIL_MATCH = 3;
 var MAX_STORED_ITEMS = 128;
 var MIN_FORK_PREFIX = 3;
 function stableStringify(value) {
-  return JSON.stringify(sortKeys(value));
+  return JSON.stringify(sortKeys2(value));
 }
-function sortKeys(value) {
-  if (Array.isArray(value)) return value.map(sortKeys);
+function sortKeys2(value) {
+  if (Array.isArray(value)) return value.map(sortKeys2);
   if (value && typeof value === "object") {
     const record = value;
     const out = {};
-    for (const key of Object.keys(record).sort()) out[key] = sortKeys(record[key]);
+    for (const key of Object.keys(record).sort()) out[key] = sortKeys2(record[key]);
     return out;
   }
   return value;
@@ -55058,8 +55134,7 @@ async function handlePluginTool(payload, res, deps) {
         config: deps.config,
         messages,
         session,
-        log: (m2) => deps.log("info", `[${session.id}] [plugin] ${m2}`),
-        nudge: mem?.nudge
+        log: (m2) => deps.log("info", `[${session.id}] [plugin] ${m2}`)
       }, callId);
     });
   } catch (err2) {
@@ -58421,12 +58496,9 @@ async function preflightCompressIfNeeded(prepared, runPrepare, req, res, opts, c
     log2("error", `[${session.id}] preflight fail-fast ${status2} (retryable=${retryable}): ${message}`);
     return { failFast: true, status: status2, message, retryable, respond: !res.writableEnded };
   };
-  if ((prepared.nudge?.compressibleRanges ?? []).length === 0) {
-    if (payloadEstimate < limit) {
-      log2("warn", `[${session.id}] preflight trigger fired on a stale baseline (~${tokenCount}) but the payload fits (~${payloadEstimate}/${limit}); forwarding as-is`);
-      return prepared;
-    }
-    return failFast(502, "no part of the conversation is compressible (nothing left to fold)", false);
+  if ((prepared.nudge?.compressibleRanges ?? []).length === 0 && payloadEstimate < limit) {
+    log2("warn", `[${session.id}] preflight trigger fired on a stale baseline (~${tokenCount}) but the payload fits (~${payloadEstimate}/${limit}); forwarding as-is`);
+    return prepared;
   }
   log2("warn", `[${session.id}] context ${tokenCount} tokens exceeds model window ${limit} (model=${model}); preflight compressing before forward`);
   const { upstreamUrl, headers, proxyUrl } = buildForwardTarget(req, opts, route, affinity, instanceId);
@@ -58452,14 +58524,13 @@ async function preflightCompressIfNeeded(prepared, runPrepare, req, res, opts, c
     },
     prepared.originalMessages
   );
-  if (result.payloadEstimate < limit) {
-    if (result.compressedRanges > 0) {
-      log2("info", `[${session.id}] preflight compressed ${result.compressedRanges} range(s), ~${result.savedTokens} tokens saved (${tokenCount} \u2192 ${session.stats.lastInputTokens}) in ${Date.now() - started}ms; rebuilding payload`);
-      const rebuilt = runPrepare();
-      session.stats.requests -= 1;
-      return rebuilt;
-    }
-    log2("warn", `[${session.id}] preflight made no progress but the payload fits (~${result.payloadEstimate}/${limit}); forwarding as-is`);
+  if (result.compressedRanges > 0) {
+    log2("info", `[${session.id}] preflight compressed ${result.compressedRanges} range(s), ~${result.savedTokens} tokens saved (${tokenCount} \u2192 ${session.stats.lastInputTokens}) in ${Date.now() - started}ms; rebuilding payload`);
+    const rebuilt = runPrepare();
+    session.stats.requests -= 1;
+    if (estimateCoreMessages(rebuilt.processedMessages) + imageTokens < limit) return rebuilt;
+  } else if (estimateCoreMessages(prepared.processedMessages) + imageTokens < limit) {
+    log2("warn", `[${session.id}] preflight made no progress but the payload fits; forwarding as-is`);
     return prepared;
   }
   const f2 = result.failure;
@@ -58468,7 +58539,7 @@ async function preflightCompressIfNeeded(prepared, runPrepare, req, res, opts, c
     return { failFast: true, status: 0, message: f2.detail, retryable: false, respond: false };
   }
   const status = f2?.kind === "upstream" && f2.status === 429 ? 503 : 502;
-  return failFast(status, f2?.detail ?? "unknown reason", status === 503);
+  return failFast(status, f2?.detail ?? "the payload still exceeds the window after preflight compression", status === 503);
 }
 async function forward(req, res, opts, body, prepared, core, config, log2, route, instanceId, affinity) {
   if (prepared?.codexForge) {
@@ -58789,7 +58860,7 @@ ${hdrText}
         };
         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 },
+          { 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, refreshFolded },
           parsedReq,
           { url: upstreamUrl, headers: reqHeaders },
           adapter,