billion-context-pi 0.1.71 → 0.1.72

package/dist/index.js

@@ -14,7 +14,7 @@ import { readFileSync as readFileSync6 } from "fs";
 import { homedir as homedir10 } from "os";
 import { join as join16 } from "path";
 
-// node_modules/acp-kernel/dist/chunk-37CVFHFQ.js
+// node_modules/acp-kernel/dist/chunk-QPVKTDYF.js
 import { createRequire } from "module";
 var require2 = createRequire(import.meta.url);
 function defaultCountTokens(text) {
@@ -216,7 +216,7 @@ function formatRanges(compressible, protectedRanges) {
   if (compressible.length === 0 && protectedRanges.length === 0) {
     return "[No specific ranges detected \u2014 compress any consumed content.]";
   }
-  const refNum2 = (ref) => {
+  const refNum3 = (ref) => {
     const m2 = ref.match(/\d+/);
     return m2 ? parseInt(m2[0], 10) : 0;
   };
@@ -225,8 +225,8 @@ function formatRanges(compressible, protectedRanges) {
     entries.push({
       startRef: r.startRef,
       endRef: r.endRef,
-      startNum: refNum2(r.startRef),
-      endNum: refNum2(r.endRef),
+      startNum: refNum3(r.startRef),
+      endNum: refNum3(r.endRef),
       count: r.count,
       tokens: r.tokens,
       userMsgs: r.userMsgs ?? 0,
@@ -244,8 +244,8 @@ function formatRanges(compressible, protectedRanges) {
     entries.push({
       startRef: r.startRef,
       endRef: r.endRef,
-      startNum: refNum2(r.startRef),
-      endNum: refNum2(r.endRef),
+      startNum: refNum3(r.startRef),
+      endNum: refNum3(r.endRef),
       count: r.count,
       tokens: r.tokens,
       userMsgs: 0,
@@ -379,6 +379,24 @@ var VIABLE_RANGE_MIN_TOKENS = 200;
 function viableRanges(ranges) {
   return ranges.filter((r) => r.tokens >= VIABLE_RANGE_MIN_TOKENS);
 }
+function isHighSurrogate(c) {
+  return c >= 55296 && c <= 56319;
+}
+function isLowSurrogate(c) {
+  return c >= 56320 && c <= 57343;
+}
+function clampPrefix(text, maxUnits) {
+  const cut = Math.min(maxUnits, text.length);
+  if (cut > 0 && isHighSurrogate(text.charCodeAt(cut - 1))) return text.slice(0, cut - 1);
+  return text.slice(0, cut);
+}
+function clampWindow(text, start, end) {
+  let s = Math.max(0, Math.min(start, text.length));
+  let e = Math.min(text.length, Math.max(s, end));
+  if (s > 0 && isLowSurrogate(text.charCodeAt(s))) s += 1;
+  if (e > s && isHighSurrogate(text.charCodeAt(e - 1))) e -= 1;
+  return text.slice(s, Math.max(s, e));
+}
 
 // node_modules/acp-kernel/dist/chunk-UX4LINT7.js
 function createInitialState() {
@@ -566,9 +584,43 @@ function collectSummaryAnchors(state, indexById, summaryIndexById) {
   anchors.sort((left, right) => left.insertAt - right.insertAt);
   return anchors;
 }
+function pairSafeAnchorIndex(messages, index) {
+  let safe = index;
+  while (safe > 0 && safe < messages.length && messages[safe - 1].role === "assistant" && messages[safe].role === "assistant") {
+    safe--;
+  }
+  const resultIndexByCallId2 = /* @__PURE__ */ new Map();
+  messages.forEach((message, at) => {
+    if (message.contentType !== "tool-result") return;
+    if (typeof message.toolCallId !== "string") return;
+    if (!resultIndexByCallId2.has(message.toolCallId)) {
+      resultIndexByCallId2.set(message.toolCallId, at);
+    }
+  });
+  for (let guard = 0; guard < messages.length; guard++) {
+    let moved = safe;
+    for (let at = 0; at < safe && at < messages.length; at++) {
+      const message = messages[at];
+      if (message.role !== "assistant") continue;
+      if (message.contentType !== "tool-call") continue;
+      if (typeof message.toolCallId !== "string") continue;
+      const resultIndex = resultIndexByCallId2.get(message.toolCallId);
+      if (resultIndex !== void 0 && resultIndex >= safe && resultIndex + 1 > moved) {
+        moved = resultIndex + 1;
+      }
+    }
+    if (moved === safe) break;
+    safe = moved;
+  }
+  return safe;
+}
 function rebuildMessages(messages, covered, firstUserIndex, anchors) {
+  const safeAnchors = anchors.map((anchor) => ({
+    ...anchor,
+    insertAt: pairSafeAnchorIndex(messages, anchor.insertAt)
+  })).sort((left, right) => left.insertAt - right.insertAt);
   const result = [];
-  const pending = [...anchors];
+  const pending = [...safeAnchors];
   const anchoredSummaryIds = new Set(
     anchors.map((anchor) => summaryMessageId(anchor.blockId))
   );
@@ -661,6 +713,7 @@ function syncBlocks(messages, state) {
     nudge: { ...state.nudge, anchors: { ...state.nudge.anchors } },
     stats: { ...state.stats },
     absorbed: (state.absorbed ?? []).map((record) => ({ ...record })),
+    terminalStreak: state.terminalStreak,
     nextBlockId: state.nextBlockId,
     nextRunId: state.nextRunId
   };
@@ -716,7 +769,7 @@ function defaultConfig(modelContextLimit, overrides = {}) {
       tier2GrowthMultiplier: 1.5
     },
     promotionThreshold: 5,
-    truncate: { threshold: 0.95 },
+    truncate: { threshold: 0.95, terminalEscapeAfter: 3 },
     compress: {
       minCompressRange: 5e3,
       maxSummaryLength: 2e4,
@@ -768,6 +821,9 @@ function validateConfig(config) {
   if (config.truncate.threshold <= 0 || config.truncate.threshold > 1) {
     errors.push("truncate.threshold must be in (0, 1]");
   }
+  if (config.truncate.terminalEscapeAfter !== void 0 && (!Number.isInteger(config.truncate.terminalEscapeAfter) || config.truncate.terminalEscapeAfter < 0)) {
+    errors.push("truncate.terminalEscapeAfter must be an integer >= 0");
+  }
   for (const tier of [config.tiers.tier2Trigger, config.tiers.tier3Trigger]) {
     if (tier < 1) errors.push("tier triggers must be >= 1");
   }
@@ -1013,47 +1069,69 @@ var DEFAULTS = {
 };
 function truncateLargeToolOutputs(messages, tokenCount, config, countTokens, options = {}) {
   const opts = { ...DEFAULTS, ...options };
-  if (config.modelContextLimit <= 0) return { messages, truncatedCount: 0, savedTokens: 0 };
-  const threshold = config.truncate.threshold * config.modelContextLimit;
-  if (tokenCount < threshold) return { messages, truncatedCount: 0, savedTokens: 0 };
-  const protectedIndex = messages.length - opts.protectRecentMessages;
-  const candidates = [];
-  for (let index = 0; index < messages.length; index++) {
-    if (index >= protectedIndex) break;
-    const message = messages[index];
-    if (message.contentType !== "tool-result") continue;
-    const text = message.text ?? "";
-    if (text.length === 0 || text.includes(TRUNCATION_MARKER)) continue;
-    const tokens = countTokens(text);
-    if (tokens < opts.minOutputTokens) continue;
-    candidates.push({ index, tokens });
+  const limit = config.modelContextLimit;
+  if (limit <= 0 || tokenCount < config.truncate.threshold * limit) {
+    return { messages, truncatedCount: 0, savedTokens: 0, candidatesFound: 0 };
   }
-  if (candidates.length === 0) return { messages, truncatedCount: 0, savedTokens: 0 };
-  candidates.sort((left, right) => right.tokens - left.tokens);
-  const targetTokens = threshold * 0.9;
-  let savedTokens = 0;
-  const edits = /* @__PURE__ */ new Map();
+  const protectedIndex = messages.length - opts.protectRecentMessages;
+  const findCandidates = (predicate) => {
+    const found = [];
+    for (let i = 0; i < protectedIndex; i++) {
+      const message = messages[i];
+      if (!predicate(message)) continue;
+      const text = message.text ?? "";
+      if (text.length === 0 || text.includes(TRUNCATION_MARKER)) continue;
+      if (countTokens(text) < opts.minOutputTokens) continue;
+      found.push(message);
+    }
+    return found.sort(
+      (a, b2) => countTokens(b2.text ?? "") - countTokens(a.text ?? "")
+    );
+  };
+  const toolResults = findCandidates((m2) => m2.contentType === "tool-result");
+  const textMessages = options.includeTextMessages ? findCandidates(
+    (m2) => m2.contentType === "text" && (m2.role === "user" || m2.role === "assistant") && !isRenderedSummaryMessage(m2) && !m2.text?.startsWith(SUMMARY_HEADER)
+  ) : [];
+  const candidatesFound = toolResults.length + textMessages.length;
   let truncatedCount = 0;
-  for (const candidate of candidates) {
-    if (tokenCount - savedTokens <= targetTokens) break;
-    const original = messages[candidate.index].text ?? "";
-    if (original.length <= opts.keepPrefixChars + opts.keepSuffixChars) continue;
-    const prefix = original.slice(0, opts.keepPrefixChars);
-    const suffix = original.slice(-opts.keepSuffixChars);
-    const replacement = prefix + `
-
-...${TRUNCATION_MARKER} \u2014 original ~${candidate.tokens} tokens]...
-
-` + suffix;
-    edits.set(candidate.index, replacement);
-    savedTokens += candidate.tokens - countTokens(replacement);
-    truncatedCount++;
-  }
-  if (truncatedCount === 0) return { messages, truncatedCount: 0, savedTokens: 0 };
-  const updated = messages.map(
-    (message, index) => edits.has(index) ? { ...message, text: edits.get(index) } : message
-  );
-  return { messages: updated, truncatedCount, savedTokens };
+  let savedTokens = 0;
+  let remaining = tokenCount;
+  const targetTokens = config.truncate.threshold * limit * 0.9;
+  const replacements = /* @__PURE__ */ new Map();
+  const applyTo = (candidates) => {
+    for (const candidate of candidates) {
+      if (remaining <= targetTokens) break;
+      const original = candidate.text ?? "";
+      const tokens = countTokens(original);
+      if (original.length <= opts.keepPrefixChars + opts.keepSuffixChars) {
+        continue;
+      }
+      const prefix = clampPrefix(original, opts.keepPrefixChars);
+      const suffix = clampWindow(original, original.length - opts.keepSuffixChars, original.length);
+      const replacement = `${prefix}
+
+...${TRUNCATION_MARKER} \u2014 original ~${tokens} tokens]...
+
+${suffix}`;
+      replacements.set(candidate.id, replacement);
+      truncatedCount++;
+      remaining -= tokens - countTokens(replacement);
+      savedTokens += tokens - countTokens(replacement);
+    }
+  };
+  applyTo(toolResults);
+  applyTo(textMessages);
+  if (replacements.size === 0) {
+    return { messages, truncatedCount: 0, savedTokens: 0, candidatesFound };
+  }
+  return {
+    messages: messages.map(
+      (m2) => replacements.has(m2.id) ? { ...m2, text: replacements.get(m2.id) } : m2
+    ),
+    truncatedCount,
+    savedTokens,
+    candidatesFound
+  };
 }
 var KEEP_LAST_ORPHANED = 2;
 function rangeKey(startRef, endRef) {
@@ -1106,7 +1184,7 @@ function compactEntry(entry) {
   if (!entry || typeof entry !== "object") return entry;
   const e = entry;
   if (typeof e.summary !== "string" || e.summary.length <= SUMMARY_STUB_CHARS) return entry;
-  return { ...e, summary: `${e.summary.slice(0, SUMMARY_STUB_CHARS - 1)}\u2026` };
+  return { ...e, summary: `${clampPrefix(e.summary, SUMMARY_STUB_CHARS - 1)}\u2026` };
 }
 function serializeCompacted(obj, content, contentWasString) {
   let changed = false;
@@ -1205,6 +1283,7 @@ var COMPRESS_TOOL_NAME = "compress";
 var DECOMPRESS_TOOL_NAME = "decompress";
 var SEARCH_CONTEXT_TOOL_NAME = "search_context";
 var ACP_STATUS_TOOL_NAME = "acp_status";
+var ACP_CACHE_TOOL_NAME = "acp_cache";
 var ABSORB_TOOL_NAME = "absorb";
 var ACP_TEXT_OPEN = "<acp_compress>";
 var ACP_TEXT_CLOSE = "</acp_compress>";
@@ -1308,10 +1387,15 @@ var COMPRESS_TOOL_OPENAI = {
   }
 };
 var TEXT_PROMPT_SECTIONS = [
-  ["acpTags", `ACP TAGS
+  [
+    "acpTags",
+    `ACP TAGS
 
-Each message in the conversation is annotated with a <acp tokens="2.1K" type="tool:bash">m00175</acp> tag showing its reference ID, approximate token size, and content type. These tags are system metadata. NEVER echo these history tags. Use only the ref ID (e.g. m00005), never the XML wrapper.`],
-  ["textProtocol", `COMPRESSION PROTOCOL (TEXT)
+Each message in the conversation is annotated with a <acp tokens="2.1K" type="tool:bash">m00175</acp> tag showing its reference ID, approximate token size, and content type. These tags are system metadata. NEVER echo these history tags. Use only the ref ID (e.g. m00005), never the XML wrapper.`
+  ],
+  [
+    "textProtocol",
+    `COMPRESSION PROTOCOL (TEXT)
 
 You manage context by emitting a special trigger in your text output. When you decide a range of conversation is genuinely consumed and should be compressed into a summary, output EXACTLY this marker (the proxy intercepts and executes it; the marker is stripped from what the user sees):
 
@@ -1322,8 +1406,11 @@ Rules for the trigger:
 - JSON shape matches the compress tool: {"content":[{startId,endId,summary,topic?}]}. Batch multiple ranges in one trigger.
 - After emitting the marker, STOP your turn. Do not continue with other text \u2014 the proxy will execute the compression and return the result, then you continue fresh.
 - Do NOT wrap the marker in code fences, quotes, or commentary.
-- NEVER compress on short conversations or when context is small (well below the window limit). Only compress when context is genuinely large.`],
-  ["textTools", `ACP TOOLS (TEXT TRIGGERS)
+- NEVER compress on short conversations or when context is small (well below the window limit). Only compress when context is genuinely large.`
+  ],
+  [
+    "textTools",
+    `ACP TOOLS (TEXT TRIGGERS)
 
 Since host tools cannot coexist with a declared tools field, ALL ACP tools use text triggers. Emit the marker; the proxy intercepts and executes it; the marker is stripped from what the user sees.
 
@@ -1343,13 +1430,19 @@ Since host tools cannot coexist with a declared tools field, ALL ACP tools use t
 Rules for ALL triggers:
 - Output on its own, NO surrounding prose. Just the raw marker.
 - After emitting, STOP your turn. The proxy executes and returns the result.
-- Do NOT wrap in code fences, quotes, or commentary.`]
+- Do NOT wrap in code fences, quotes, or commentary.`
+  ]
 ];
 var HYBRID_PROMPT_SECTIONS = [
-  ["acpTags", `ACP TAGS
+  [
+    "acpTags",
+    `ACP TAGS
 
-Each message in the conversation is annotated with a <acp> tag showing its reference ID, approximate token size, and content type. These tags are system metadata. NEVER echo these history tags. Use only the ref ID (e.g. m00005), never the XML wrapper.`],
-  ["textProtocol", `COMPRESSION PROTOCOL (TEXT)
+Each message in the conversation is annotated with a <acp> tag showing its reference ID, approximate token size, and content type. These tags are system metadata. NEVER echo these history tags. Use only the ref ID (e.g. m00005), never the XML wrapper.`
+  ],
+  [
+    "textProtocol",
+    `COMPRESSION PROTOCOL (TEXT)
 
 You manage context by emitting a special trigger in your text output. When you decide a range of conversation is genuinely consumed and should be compressed into a summary, output EXACTLY this marker (the proxy intercepts and executes it; the marker is stripped from what the user sees):
 
@@ -1360,8 +1453,11 @@ Rules for the trigger:
 - JSON shape: {"content":[{startId,endId,summary,topic?}]}. Batch multiple ranges in one trigger.
 - After emitting the marker, STOP your turn. Do not continue with other text \u2014 the proxy will execute the compression and return the result, then you continue fresh.
 - Do NOT wrap the marker in code fences, quotes, or commentary.
-- NEVER compress on short conversations or when context is small (well below the window limit). Only compress when context is genuinely large.`],
-  ["functionTools", `ACP TOOLS (FUNCTION CALLS)
+- NEVER compress on short conversations or when context is small (well below the window limit). Only compress when context is genuinely large.`
+  ],
+  [
+    "functionTools",
+    `ACP TOOLS (FUNCTION CALLS)
 
 The proxy also provides these as real function tools you can call directly (they appear in your tool list). Call them like any other function; the proxy executes them and returns the result, then you continue.
 
@@ -1369,7 +1465,8 @@ The proxy also provides these as real function tools you can call directly (they
 - search_context \u2014 search compressed block summaries by keyword. Arguments: {"query":"...","limit":5}.
 - decompress \u2014 restore compressed content for exact details. Arguments: {"blockId":"b5"} (optional "toFile":"/tmp/x.txt", "full":true).
 
-Note: compress is ONLY available via the text marker above (it needs batch ranges + an immediate stop), NOT as a function tool.`]
+Note: compress is ONLY available via the text marker above (it needs batch ranges + an immediate stop), NOT as a function tool.`
+  ]
 ];
 var DECOMPRESS_TOOL_OPENAI = {
   type: "function",
@@ -1422,6 +1519,18 @@ var ACP_STATUS_TOOL_OPENAI = {
     }
   }
 };
+var ACP_CACHE_TOOL_DESCRIPTION = "Prompt-cache reconciliation: grand ledger (total input/cached/output, overall hit rate) with every request's miss split into new content / compression re-pay / TTL expiry, plus per-fold economics (breakeven turns vs measured cadence). Read-only. Call when asked about cache hits, cache invalidation, or what compression costs.";
+var ACP_CACHE_TOOL_OPENAI = {
+  type: "function",
+  function: {
+    name: ACP_CACHE_TOOL_NAME,
+    description: ACP_CACHE_TOOL_DESCRIPTION,
+    parameters: {
+      type: "object",
+      properties: {}
+    }
+  }
+};
 var DECOMPRESS_TOOL = {
   name: DECOMPRESS_TOOL_NAME,
   description: DECOMPRESS_TOOL_OPENAI.function.description,
@@ -1437,6 +1546,11 @@ var ACP_STATUS_TOOL = {
   description: ACP_STATUS_TOOL_OPENAI.function.description,
   input_schema: ACP_STATUS_TOOL_OPENAI.function.parameters
 };
+var ACP_CACHE_TOOL = {
+  name: ACP_CACHE_TOOL_NAME,
+  description: ACP_CACHE_TOOL_DESCRIPTION,
+  input_schema: ACP_CACHE_TOOL_OPENAI.function.parameters
+};
 var COMPRESS_TOOL_RESPONSES = {
   type: "function",
   name: COMPRESS_TOOL_NAME,
@@ -1461,11 +1575,18 @@ var ACP_STATUS_TOOL_RESPONSES = {
   description: ACP_STATUS_TOOL_OPENAI.function.description,
   parameters: ACP_STATUS_TOOL_OPENAI.function.parameters
 };
+var ACP_CACHE_TOOL_RESPONSES = {
+  type: "function",
+  name: ACP_CACHE_TOOL_NAME,
+  description: ACP_CACHE_TOOL_DESCRIPTION,
+  parameters: ACP_CACHE_TOOL_OPENAI.function.parameters
+};
 var ACP_TOOL_NAMES = /* @__PURE__ */ new Set([
   COMPRESS_TOOL_NAME,
   DECOMPRESS_TOOL_NAME,
   SEARCH_CONTEXT_TOOL_NAME,
-  ACP_STATUS_TOOL_NAME
+  ACP_STATUS_TOOL_NAME,
+  ACP_CACHE_TOOL_NAME
 ]);
 var ALWAYS_PROTECTED_TOOLS = ["compress"];
 var NEVER_PRESERVE_RECENT_TOOLS = [
@@ -2084,8 +2205,9 @@ function mergeRangesToThreshold(ranges, minChars) {
       batchChars = 0;
     }
   }
-  if (batch.length > 0) {
-    result.push(mergeBatch(batch));
+  if (batch.length > 0 && result.length > 0) {
+    const prev = result[result.length - 1];
+    result[result.length - 1] = mergeBatch([prev, ...batch]);
   }
   return result;
 }
@@ -2288,6 +2410,7 @@ function createCore(ports = {}) {
       state.nudge.lastPerMessageNudgeTokens = 0;
       state.nudge.lastNudgeShownTokens = 0;
       state.nudge.lastShownByTier = {};
+      state.terminalStreak = 0;
     }
     return {
       state,
@@ -2317,7 +2440,9 @@ function createCore(ports = {}) {
     return {
       messages: result.messages,
       state: result.state,
-      nudge: result.effects.nudge
+      nudge: result.effects.nudge,
+      terminalEscape: result.effects.terminalEscape,
+      truncationSkipped: result.effects.truncationSkipped
     };
   }
   function decompress(blockId, state) {
@@ -2510,18 +2635,49 @@ var emergencyTruncateNode = {
   name: "emergency-truncate",
   run(io, ctx) {
     const usage = ctx.config.modelContextLimit > 0 ? ctx.tokenCount / ctx.config.modelContextLimit : 0;
-    if (usage < ctx.config.truncate.threshold) return io;
+    const prevStreak = io.state.terminalStreak ?? 0;
+    if (usage < ctx.config.truncate.threshold) {
+      return prevStreak > 0 ? { ...io, state: { ...io.state, terminalStreak: 0 } } : io;
+    }
     const trunc = truncateLargeToolOutputs(
       io.messages,
       ctx.tokenCount,
       ctx.config,
       ctx.countTokens,
-      { protectRecentMessages: ctx.config.preserveRecentMessages }
+      {
+        protectRecentMessages: ctx.config.preserveRecentMessages,
+        includeTextMessages: true
+      }
     );
+    const nudge = io.effects.nudge;
+    const minBenefit = nudge?.breakdown.minPressureBenefit ?? 0;
+    const maxPending = nudge?.breakdown.maxPending ?? 0;
+    const noViableCompression = nudge !== void 0 && (minBenefit > 0 ? maxPending < minBenefit : maxPending <= 0);
+    const stuck = noViableCompression && trunc.savedTokens <= 0;
+    const streak = stuck ? prevStreak + 1 : 0;
+    const escapeAfter = ctx.config.truncate.terminalEscapeAfter ?? 3;
+    const triggered = stuck && escapeAfter > 0 && streak >= escapeAfter;
+    const effects = {
+      ...io.effects,
+      truncatedCount: trunc.truncatedCount
+    };
+    if (trunc.savedTokens <= 0) {
+      effects.truncationSkipped = trunc.candidatesFound === 0 ? `emergency-truncate ran at ${Math.round(usage * 100)}% usage but found no truncatable content (no oversized tool-result or text message outside the last ${ctx.config.preserveRecentMessages} messages)` : `emergency-truncate found ${trunc.candidatesFound} candidate(s) at ${Math.round(usage * 100)}% usage but none were large enough to save tokens`;
+    }
+    if (triggered) {
+      effects.terminalEscape = {
+        message: `Usage at ${Math.round(usage * 100)}% (${ctx.tokenCount}/${ctx.config.modelContextLimit} tokens) persists with nothing compressible above the benefit floor and no truncatable content: compression cannot reduce this context below the limit. Start a new session or use native compaction.`,
+        usage,
+        tokenCount: ctx.tokenCount,
+        modelContextLimit: ctx.config.modelContextLimit,
+        stuckEvents: streak
+      };
+    }
     return {
       ...io,
       messages: trunc.messages,
-      effects: { ...io.effects, truncatedCount: trunc.truncatedCount }
+      state: { ...io.state, terminalStreak: streak },
+      effects
     };
   }
 };
@@ -2614,7 +2770,8 @@ function applySingleRange(input) {
     for (const m2 of input.messages) {
       if (!m2.id) continue;
       if (m2.contentType === "reasoning") reasoningIds.add(m2.id);
-      if (m2.role === "assistant" && m2.contentType === "tool-call") callIds.add(m2.id);
+      if (m2.role === "assistant" && m2.contentType === "tool-call")
+        callIds.add(m2.id);
     }
     const withdrawIds = /* @__PURE__ */ new Set();
     let splitTurnCount = 0;
@@ -2630,17 +2787,42 @@ function applySingleRange(input) {
       splitTurnCount++;
       for (const id of group) withdrawIds.add(id);
     }
+    const resultIdByCallId = /* @__PURE__ */ new Map();
+    for (const m2 of input.messages) {
+      if (!m2.id || m2.contentType !== "tool-result") continue;
+      if (typeof m2.toolCallId !== "string") continue;
+      if (!resultIdByCallId.has(m2.toolCallId)) {
+        resultIdByCallId.set(m2.toolCallId, m2.id);
+      }
+    }
+    let splitPairCount = 0;
+    for (const m2 of input.messages) {
+      if (!m2.id || m2.contentType !== "tool-call") continue;
+      if (typeof m2.toolCallId !== "string") continue;
+      const resultId = resultIdByCallId.get(m2.toolCallId);
+      if (resultId === void 0) continue;
+      if (effectiveMessageIds.has(m2.id) === effectiveMessageIds.has(resultId)) {
+        continue;
+      }
+      withdrawIds.add(m2.id);
+      withdrawIds.add(resultId);
+      splitPairCount++;
+    }
     if (withdrawIds.size > 0) {
       for (const id of withdrawIds) effectiveMessageIds.delete(id);
       const beforeWithdraw = filteredIds.length;
       filteredIds = filteredIds.filter((id) => !withdrawIds.has(id));
+      const splitDesc = [
+        splitTurnCount > 0 ? `${splitTurnCount} turn(s)` : null,
+        splitPairCount > 0 ? `${splitPairCount} tool call/result pair(s)` : null
+      ].filter((part) => part !== null).join(" and ");
       if (filteredIds.length === 0 && consumedBlockIds.length === 0) {
         throw new Error(
-          `Range would split ${splitTurnCount} turn(s) at the protected-zone boundary: a visible tool-call must keep its reasoning run (strict-echo providers reject a rebuilt request that lost it). Shrink the range to end before the turn starts, or wait until the whole turn ages out of the protected zone.`
+          `Range would split ${splitDesc} at the protected-zone boundary: a visible tool-call must keep its reasoning run and its results (strict providers reject a rebuilt request that lost either). Shrink the range to end before the turn starts, or wait until the whole turn ages out of the protected zone.`
         );
       }
       warnings.push(
-        `Withdrawn ${beforeWithdraw - filteredIds.length} message(s) from compression range to keep ${splitTurnCount} turn(s) intact (visible tool-call would lose its reasoning run).`
+        `Withdrawn ${beforeWithdraw - filteredIds.length} message(s) from compression range to keep ${splitDesc} intact (visible tool-call would lose its reasoning run or its results).`
       );
     }
   }
@@ -2860,6 +3042,7 @@ function decideNudge(input) {
   const t1Eff = tiers[1]?.pending ?? 0;
   const t2Pen = tiers[2]?.pending ?? 0;
   const t3Pen = tiers[3]?.pending ?? 0;
+  const maxPending = Math.max(0, t1Eff, t2Pen, t3Pen);
   const firstSightMassReady = state.nudge.lastNudgeShownTokens === 0 && baseline === 0 && usage >= config.nudge.minContextLimitPct && Math.max(t1Eff, t2Pen, t3Pen) >= nudgeGrowthTokens;
   const growthReady = firstSightMassReady || growthSinceReference >= growthFloor;
   const t2Count = tiers[2]?.targetBlocks.length ?? 0;
@@ -2932,10 +3115,6 @@ function decideNudge(input) {
       (t) => ((tiers[t]?.pending ?? 0) >= nudgeGrowthTokens || countReady(t)) && (state.nudge.lastShownByTier[t] ?? 0) > 0 && tokenCount - (state.nudge.lastShownByTier[t] ?? 0) < growthFloor
     ).map((t) => `T${t} (cadence)`);
     const blockedHint = blocked.length > 0 ? `, blocked: ${blocked.join(", ")}` : "";
-    const maxPending = Math.max(
-      0,
-      ...Object.values(tiers).map((t) => t.pending)
-    );
     const pendingShort = maxPending < nudgeGrowthTokens;
     const growthShort = growthSinceReference < growthFloor;
     const parts = [];
@@ -2979,7 +3158,8 @@ function decideNudge(input) {
       minPressureBenefit,
       pendingT1: tiers[1].pending,
       pendingT2: tiers[2].pending,
-      pendingT3: tiers[3].pending
+      pendingT3: tiers[3].pending,
+      maxPending
     },
     contextBreakdown: ctxBreakdown
   };
@@ -3019,6 +3199,7 @@ function cloneState(state) {
     nudge: { ...state.nudge, anchors: { ...state.nudge.anchors } },
     stats: { ...state.stats },
     absorbed: (state.absorbed ?? []).map((record) => ({ ...record })),
+    terminalStreak: state.terminalStreak,
     nextBlockId: state.nextBlockId,
     nextRunId: state.nextRunId
   };
@@ -3488,7 +3669,7 @@ function renderUncompressedRanges(visible) {
     lines.push("  (no uncompressed messages)");
     return lines.join("\n");
   }
-  const refNum2 = (ref) => {
+  const refNum22 = (ref) => {
     const m2 = ref.match(/\d+/);
     return m2 ? parseInt(m2[0], 10) : 0;
   };
@@ -3505,16 +3686,16 @@ function renderUncompressedRanges(visible) {
   };
   const merged = [];
   for (const m2 of visible) {
-    const num2 = refNum2(m2.ref);
+    const num3 = refNum22(m2.ref);
     const last = merged[merged.length - 1];
-    if (last && num2 === last.startNum + last.count) {
+    if (last && num3 === last.startNum + last.count) {
       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 {
       const toolTokens = /* @__PURE__ */ new Map([[m2.tool, m2.tokens]]);
-      merged.push({ startRef: m2.ref, endRef: m2.ref, startNum: num2, count: 1, tokens: m2.tokens, toolTokens });
+      merged.push({ startRef: m2.ref, endRef: m2.ref, startNum: num3, count: 1, tokens: m2.tokens, toolTokens });
     }
   }
   for (const r of merged.slice(0, 30)) {
@@ -3587,6 +3768,276 @@ function renderCompressedDrilldown(blocks, state, sort, limit, countTokens, meta
   }
   return lines.join("\n");
 }
+function decomposeSample(prev, cur, pendingFolds) {
+  const missed = Math.max(0, cur.input - cur.cached);
+  const growth = prev ? Math.max(0, cur.input - prev.input) : 0;
+  const newContent = Math.min(missed, growth);
+  const remainder = missed - newContent;
+  let compRepay = 0;
+  let foldIndex = null;
+  if (remainder > 0 && pendingFolds.length > 0) {
+    let bestX = Infinity;
+    let owner = -1;
+    for (let i = 0; i < pendingFolds.length; i++) {
+      const f = pendingFolds[i];
+      const x2 = f.firstFoldStartTokens ?? f.viewAfter ?? 0;
+      if (x2 < bestX) {
+        bestX = x2;
+        owner = i;
+      }
+    }
+    const structuralExcess = Math.max(0, cur.input - bestX - newContent);
+    compRepay = Math.min(remainder, structuralExcess);
+    if (compRepay > 0) foldIndex = owner;
+  }
+  return {
+    missed,
+    newContent,
+    compRepay,
+    ttlRepay: remainder - compRepay,
+    foldIndex
+  };
+}
+function hitPct(input, cached) {
+  return input > 0 ? cached / input * 100 : 0;
+}
+function computeFoldEconomics(f, price) {
+  const w2 = price?.w ?? 1;
+  const r = price?.r ?? 0.1;
+  const q2 = price?.q ?? 4;
+  const netTokenDelta = f.T + f.sigma - f.S;
+  const savingPerTurn = f.S - f.sigma;
+  const oneTimeCostUnits = (w2 - r) * f.T + q2 * f.sigma - r * f.S;
+  const perTurnSavingUnits = savingPerTurn * r;
+  const breakevenTurns = perTurnSavingUnits > 0 ? Math.max(0, oneTimeCostUnits) / perTurnSavingUnits : null;
+  const paidBack = f.turnsToNextFold !== null && breakevenTurns !== null ? f.turnsToNextFold >= breakevenTurns : null;
+  return {
+    seq: f.seq,
+    at: f.at,
+    S: f.S,
+    sigma: f.sigma,
+    Vprime: f.Vprime ?? null,
+    hPct: f.hPct,
+    T: f.T,
+    requestsAfter: f.requestsAfter,
+    savedSoFar: Math.max(0, savingPerTurn) * f.requestsAfter,
+    turnsToNextFold: f.turnsToNextFold,
+    netTokenDelta,
+    oneTimeCostUnits: round1(oneTimeCostUnits),
+    perTurnSavingUnits: round1(perTurnSavingUnits),
+    breakevenTurns,
+    paidBack
+  };
+}
+function summarizeFoldEconomics(foldEcon) {
+  const econ = {
+    folds: foldEcon.length,
+    grossSaved: foldEcon.reduce((n, e) => n + e.savedSoFar, 0),
+    repayCost: foldEcon.reduce((n, e) => n + e.T, 0),
+    summaryCost: foldEcon.reduce((n, e) => n + e.sigma, 0),
+    netTokens: 0,
+    paidBackCount: 0,
+    notPaidBackCount: 0,
+    unobservedCount: 0
+  };
+  econ.netTokens = econ.grossSaved - econ.repayCost - econ.summaryCost;
+  for (const e of foldEcon) {
+    if (e.paidBack === true) econ.paidBackCount++;
+    else if (e.paidBack === false) econ.notPaidBackCount++;
+    else econ.unobservedCount++;
+  }
+  return econ;
+}
+function buildCacheReport(samplesIn, foldsIn, opts = {}) {
+  const samples = [...samplesIn].sort((a, b2) => a.at - b2.at);
+  const folds = [...foldsIn].sort(
+    (a, b2) => a.at - b2.at || a.tokensCompressed - b2.tokensCompressed
+  );
+  const maxLines = opts.maxLines && opts.maxLines > 0 ? opts.maxLines : 512;
+  const price = opts.priceProfile;
+  const w2 = price?.w ?? 1;
+  const r = price?.r ?? 0.1;
+  const q2 = price?.q ?? 4;
+  const decs = [];
+  for (let i = 0; i < samples.length; i++) {
+    const s = samples[i];
+    const prev = i > 0 ? samples[i - 1] : null;
+    const prevAt = prev ? prev.at : Number.NEGATIVE_INFINITY;
+    const pending = [];
+    let pendingBase = 0;
+    for (let j2 = 0; j2 < folds.length; j2++) {
+      const fj = folds[j2];
+      if (fj.at <= prevAt) continue;
+      if (fj.at > s.at) break;
+      if (pending.length === 0) pendingBase = j2;
+      pending.push(fj);
+    }
+    const d2 = decomposeSample(prev, s, pending);
+    decs.push({
+      ...d2,
+      foldSeq: d2.foldIndex !== null ? pendingBase + d2.foldIndex + 1 : null
+    });
+  }
+  const totals = {
+    requests: samples.length,
+    input: 0,
+    cached: 0,
+    output: 0,
+    hitPct: 0,
+    newContent: 0,
+    compRepay: 0,
+    ttlRepay: 0,
+    residual: 0,
+    balanced: true
+  };
+  for (let i = 0; i < samples.length; i++) {
+    const s = samples[i];
+    const d2 = decs[i];
+    totals.input += s.input;
+    totals.cached += s.cached;
+    totals.output += s.output ?? 0;
+    totals.newContent += d2.newContent;
+    totals.compRepay += d2.compRepay;
+    totals.ttlRepay += d2.ttlRepay;
+  }
+  totals.hitPct = round1(hitPct(totals.input, totals.cached));
+  totals.residual = totals.input - totals.cached - (totals.newContent + totals.compRepay + totals.ttlRepay);
+  totals.balanced = totals.residual === 0;
+  const lines = samples.map((s, i) => {
+    const { foldIndex, ...rest } = decs[i];
+    return {
+      seq: i + 1,
+      at: s.at,
+      input: s.input,
+      cached: s.cached,
+      output: s.output ?? 0,
+      hitPct: round1(hitPct(s.input, s.cached)),
+      ...rest
+    };
+  });
+  const linesOmitted = Math.max(0, lines.length - maxLines);
+  const visibleLines = lines.slice(linesOmitted);
+  const foldEcon = folds.map((f, j2) => {
+    const seq = j2 + 1;
+    const S = f.tokensCompressed;
+    const sigma = f.summaryTokens ?? 0;
+    const T2 = decs.reduce(
+      (n, d2) => d2.foldSeq === seq ? n + d2.compRepay : n,
+      0
+    );
+    let hPct = null;
+    let requestsAfter = 0;
+    for (const s of samples) {
+      if (s.at <= f.at) continue;
+      requestsAfter++;
+      if (hPct === null) hPct = round1(hitPct(s.input, s.cached));
+    }
+    const nextAt = j2 + 1 < folds.length ? folds[j2 + 1]?.at : void 0;
+    const turnsToNextFold = nextAt !== void 0 ? samples.filter((s) => s.at > f.at && s.at <= nextAt).length : null;
+    return computeFoldEconomics(
+      {
+        seq,
+        at: f.at,
+        S,
+        sigma,
+        Vprime: f.viewAfter,
+        hPct,
+        T: T2,
+        requestsAfter,
+        turnsToNextFold
+      },
+      { w: w2, r, q: q2 }
+    );
+  });
+  const econ = summarizeFoldEconomics(foldEcon);
+  return {
+    generatedAt: Date.now(),
+    profile: { w: w2, r, q: q2 },
+    totals,
+    economics: econ,
+    folds: foldEcon,
+    lines: visibleLines,
+    linesOmitted
+  };
+}
+function round1(n) {
+  return Math.round(n * 10) / 10;
+}
+function fmtTok(n) {
+  const v2 = Math.round(n);
+  return v2 >= 1e6 ? `${(v2 / 1e6).toFixed(2)}M` : v2 >= 1e4 ? `${(v2 / 1e3).toFixed(1)}K` : String(v2);
+}
+function fmtTime(at) {
+  const d2 = new Date(at);
+  const p = (x2) => String(x2).padStart(2, "0");
+  return `${p(d2.getHours())}:${p(d2.getMinutes())}:${p(d2.getSeconds())}`;
+}
+function formatCacheReport(report, sessionLabel) {
+  const t = report.totals;
+  const out = [];
+  out.push(
+    `ACP CACHE REPORT${sessionLabel ? ` (${sessionLabel})` : ""} \u2014 ${t.requests} requests`
+  );
+  out.push("");
+  out.push("GRAND LEDGER");
+  out.push(`  total input    ${fmtTok(t.input)} tok`);
+  out.push(
+    `  total cached   ${fmtTok(t.cached)} tok  (hit ${t.hitPct.toFixed(1)}%)`
+  );
+  out.push(`  total output   ${fmtTok(t.output)} tok`);
+  out.push(
+    `  miss breakdown (input \u2212 cached = ${fmtTok(Math.max(0, t.input - t.cached))} tok):`
+  );
+  out.push(
+    `    new content     ${fmtTok(t.newContent)} tok  (fresh append \u2014 not an invalidation)`
+  );
+  out.push(`    compress re-pay ${fmtTok(t.compRepay)} tok  (caused by folds)`);
+  out.push(
+    `    ttl/other       ${fmtTok(t.ttlRepay)} tok  (stable-prefix misses: TTL expiry / eviction)`
+  );
+  out.push(
+    `  identity check   ${t.balanced ? "OK" : "BROKEN"} \u2014 ${fmtTok(t.input)} = ${fmtTok(t.cached)} + ${fmtTok(t.newContent)} + ${fmtTok(t.compRepay)} + ${fmtTok(t.ttlRepay)} (residual ${t.residual})`
+  );
+  const e = report.economics;
+  if (e.folds > 0) {
+    const p = report.profile;
+    out.push("");
+    out.push(`FOLD ECONOMICS (${e.folds} folds @ w=${p.w} r=${p.r} q=${p.q})`);
+    out.push(
+      `  gross saved ${fmtTok(e.grossSaved)} tok \xB7 repay cost ${fmtTok(e.repayCost)} tok \xB7 summary cost ${fmtTok(e.summaryCost)} tok \u2192 net ${e.netTokens >= 0 ? "+" : ""}${fmtTok(e.netTokens)} tok`
+    );
+    out.push(
+      `  verdict: ${e.paidBackCount} paid back, ${e.notPaidBackCount} not paid back, ${e.unobservedCount} unobserved`
+    );
+    for (const f of report.folds) {
+      const nstar = f.breakevenTurns === null ? "n/a" : f.breakevenTurns.toFixed(1);
+      const k = f.turnsToNextFold === null ? "\u2014" : String(f.turnsToNextFold);
+      const verdict = f.paidBack === null ? "?" : f.paidBack ? "PAID BACK" : "NOT PAID BACK";
+      const h = f.hPct === null ? "n/a" : `${f.hPct.toFixed(1)}%`;
+      out.push(
+        `  #${f.seq} ${fmtTime(f.at)} S=${fmtTok(f.S)} \u03C3=${fmtTok(f.sigma)} h=${h} T=${fmtTok(f.T)} \u0394C\u2081=${fmtTok(f.oneTimeCostUnits)} \u0394s=${fmtTok(f.perTurnSavingUnits)}/turn n*=${nstar} k=${k} \u2192 ${verdict}`
+      );
+    }
+  }
+  if (report.lines.length > 0) {
+    out.push("");
+    if (report.linesOmitted > 0)
+      out.push(
+        `LINE ITEMS (last ${report.lines.length} of ${report.lines.length + report.linesOmitted}):`
+      );
+    else out.push("LINE ITEMS:");
+    out.push(
+      "  #     time      input  cached    hit%      new    comp     ttl  fold"
+    );
+    for (const l3 of report.lines) {
+      const fold = l3.foldSeq !== null ? `#${l3.foldSeq}` : "";
+      out.push(
+        `  ${String(l3.seq).padStart(4)}  ${fmtTime(l3.at)}  ${fmtTok(l3.input).padStart(7)}  ${fmtTok(l3.cached).padStart(7)}  ${l3.hitPct.toFixed(1).padStart(5)}%  ${fmtTok(l3.newContent).padStart(6)}  ${fmtTok(l3.compRepay).padStart(6)}  ${fmtTok(l3.ttlRepay).padStart(6)}  ${fold}`
+      );
+    }
+  }
+  return out.join("\n");
+}
 function renderMessage2(m2) {
   const parts = [];
   switch (m2.contentType) {
@@ -3672,7 +4123,7 @@ function parseCompressArgs(input, opts) {
   return parseObjectValue(input, callId, diag);
 }
 function parseStringInput(raw, callId, diag) {
-  diag.rawPrefix = raw.slice(0, 800);
+  diag.rawPrefix = clampPrefix(raw, 800);
   diag.length = raw.length;
   const cleaned = stripFence(raw.trim());
   const first = parseStringCore(cleaned, callId, diag);
@@ -3856,7 +4307,7 @@ function deriveTopicFromSummary(summary) {
   const source = heading !== null ? heading[1] : summary.split("\n").find((l3) => l3.trim().length > 0) ?? "";
   const text = source.trim().replace(/^[\u2022\-*]\s+/, "");
   if (text.length === 0) return void 0;
-  return text.length > 60 ? text.slice(0, 60).trimEnd() : text;
+  return text.length > 60 ? clampPrefix(text, 60).trimEnd() : text;
 }
 function splitLineEntries(content) {
   const parts = content.split(/\n(?=\s*(?:["']\s*,\s*)?["']?\s*[mb]\d{1,7}\s*(?:[-\u2013\u2014\u2026~]|\.\.\.|to)\s*[mb]\d{1,7}\b)/i).map((p) => p.trim().replace(/^(?:["']\s*,?\s*)+/, "").replace(/["']\s*$/, "")).filter((p) => p.length > 0);
@@ -4369,7 +4820,7 @@ function messageDocs(msgs) {
     kind: "message",
     ref: m2.ref,
     text: m2.text,
-    title: `${m2.role}: ${m2.text.slice(0, 60)}`,
+    title: `${m2.role}: ${clampPrefix(m2.text, 60)}`,
     role: m2.role,
     blockId: m2.blockId,
     tier: m2.tier,
@@ -4431,7 +4882,7 @@ function searchBlocks(docs, query, options = {}) {
 function makePreview(text, query, len) {
   if (!text) return "";
   const terms = query.toLowerCase().trim().split(/\s+/).filter((t) => t.length > 1);
-  if (terms.length === 0) return text.slice(0, len);
+  if (terms.length === 0) return clampPrefix(text, len);
   const lower = text.toLowerCase();
   let hitIdx = -1;
   for (const term of terms) {
@@ -4441,13 +4892,13 @@ function makePreview(text, query, len) {
       break;
     }
   }
-  if (hitIdx < 0) return text.slice(0, len);
+  if (hitIdx < 0) return clampPrefix(text, len);
   const half = Math.max(0, Math.floor(len / 2) - 10);
   const start = Math.max(0, hitIdx - half);
   const end = Math.min(text.length, start + len);
   const prefix = start > 0 ? "\u2026" : "";
   const suffix = end < text.length ? "\u2026" : "";
-  return prefix + text.slice(start, end).trim() + suffix;
+  return prefix + clampWindow(text, start, end).trim() + suffix;
 }
 
 // src/log.ts
@@ -5695,7 +6146,7 @@ When a background delegate finishes, an automated completion notification is inj
 `;
 
 // src/surface.ts
-var TOOL_NAMES = /* @__PURE__ */ new Set(["compress", "decompress", "search_context", "acp_status"]);
+var TOOL_NAMES = /* @__PURE__ */ new Set(["compress", "decompress", "search_context", "acp_status", "acp_cache"]);
 var NUDGE_KEYS = /* @__PURE__ */ new Set(["efficiencyNote", "emergencyHeader", "t2Guidance", "t3Guidance"]);
 function normalizeGuidelines(g2) {
   return Array.isArray(g2) ? g2 : [g2];
@@ -11672,7 +12123,7 @@ function resolveActivePack(adapter, cwd, provider, modelId, resolver) {
 }
 var HOST_VERSION = (() => {
   try {
-    return "0.1.71";
+    return "0.1.72";
   } catch {
     return "dev";
   }
@@ -13720,6 +14171,71 @@ async function handleStatus(args, runtime, ctx) {
 ${extra.join("\n")}` : base;
 }
 
+// src/cache-tool.ts
+var CacheParams = typebox_exports.Object({});
+var num = (v2) => typeof v2 === "number" && Number.isFinite(v2) && v2 > 0 ? v2 : 0;
+var refNum2 = (ref) => Number(ref.replace(/\D/g, "")) || 0;
+function cacheSamples(entries) {
+  const out = [];
+  for (const e of entries) {
+    if (e.type !== "message") continue;
+    const m2 = e.message;
+    if (m2?.role !== "assistant" || !m2.usage) continue;
+    const read = num(m2.usage.cacheRead);
+    const write = num(m2.usage.cacheWrite);
+    const fresh = num(m2.usage.input);
+    if (fresh + read + write <= 0) continue;
+    out.push({ at: Date.parse(e.timestamp) || 0, input: fresh + read + write, cached: read, output: num(m2.usage.output) });
+  }
+  return out;
+}
+function foldEvents(state) {
+  const byRef = state.messageRefs?.byRef ?? {};
+  const snap = state.tokenSnapshot ?? {};
+  const prefixBefore = (ref) => {
+    if (!ref || refNum2(ref) === 0) return void 0;
+    let n = 0;
+    for (const key of Object.keys(byRef)) {
+      if (refNum2(key) < refNum2(ref)) n += snap[key] ?? 0;
+    }
+    return n;
+  };
+  return [...state.blocks].sort((a, b2) => a.createdAt - b2.createdAt).map((b2) => ({
+    at: b2.createdAt,
+    tokensCompressed: b2.compressedTokens,
+    summaryTokens: defaultCountTokens(b2.summary || ""),
+    firstFoldStartTokens: prefixBefore(b2.startRef)
+  }));
+}
+async function cacheReportText(runtime, ctx) {
+  const { state, entries } = await runtime.stateFor(ctx);
+  const report = buildCacheReport(cacheSamples(entries), foldEvents(state));
+  return formatCacheReport(report, ctx.sessionManager.getSessionId());
+}
+function makeCacheTool(runtime, overrides) {
+  return applyToolPromptOverrides({
+    name: "acp_cache",
+    label: "ACP Cache Report",
+    description: "Prompt-cache reconciliation: grand ledger (total input/cached/output, session hit rate) with every request's miss split into new content / compression re-pay / TTL expiry, plus per-fold economics (one-time cost, breakeven turns vs measured cadence). Read-only.",
+    promptSnippet: "acp_cache({})",
+    promptGuidelines: [
+      "Call when asked about cache hits, cache invalidation, or what compression costs.",
+      "Read-only: reports numbers, never mutates context."
+    ],
+    parameters: CacheParams,
+    async execute(_toolCallId, _params, _signal, _onUpdate, ctx) {
+      if (runtime.refused) return { details: void 0, content: [{ type: "text", text: runtime.refusalMessage ?? UNSUPPORTED_HOST_MESSAGE }] };
+      try {
+        const text = await cacheReportText(runtime, ctx);
+        return { details: void 0, content: [{ type: "text", text }] };
+      } catch (e) {
+        logThrow("cache", e, { sid: ctx.sessionManager.getSessionId() });
+        throw e;
+      }
+    }
+  }, overrides);
+}
+
 // src/commands.ts
 import * as path12 from "path";
 
@@ -13847,7 +14363,7 @@ function parseExportArgs(args) {
 function topicFallback(summary) {
   const first = summary.split(/[.\n]/)[0] ?? "";
   const t = first.trim().replace(/^["'`]+/, "").trim();
-  return t.length <= 30 ? t : `${t.slice(0, 30).trimEnd()}\u2026`;
+  return t.length <= 30 ? t : `${clampPrefix(t, 30).trimEnd()}\u2026`;
 }
 function formatCompactTokens2(count) {
   if (count < 1e3) return count.toString();
@@ -13856,7 +14372,7 @@ function formatCompactTokens2(count) {
   if (count < 1e7) return `${(count / 1e6).toFixed(1)}M`;
   return `${Math.round(count / 1e6)}M`;
 }
-function num(v2) {
+function num2(v2) {
   return typeof v2 === "number" && Number.isFinite(v2) && v2 > 0 ? v2 : 0;
 }
 function cacheHitStats(usages) {
@@ -13865,9 +14381,9 @@ function cacheHitStats(usages) {
   let requests = 0;
   let last;
   for (const u of usages) {
-    const read = num(u?.cacheRead);
-    const write = num(u?.cacheWrite);
-    const input = num(u?.input);
+    const read = num2(u?.cacheRead);
+    const write = num2(u?.cacheWrite);
+    const input = num2(u?.input);
     if (read + write <= 0) continue;
     const total = read + write + input;
     if (total <= 0) continue;
@@ -18195,6 +18711,26 @@ function makeCommands(runtime, pi) {
         handler: statusHandler
       }
     },
+    {
+      name: "acp-cache",
+      options: {
+        description: "Prompt-cache reconciliation: grand ledger (input/cached/hit rate) with every request's miss split into new content / compression re-pay / TTL expiry, plus per-fold economics.",
+        handler: async (_args, ctx) => {
+          let text;
+          try {
+            text = await cacheReportText(runtime, ctx);
+          } catch (e) {
+            ctx.ui.notify(e instanceof Error ? e.message : String(e), "error");
+            return;
+          }
+          if (typeof pi?.sendMessage === "function") {
+            pi.sendMessage({ customType: ACP_STATUS_CUSTOM_TYPE, content: text, display: true });
+            return;
+          }
+          ctx.ui.notify(text);
+        }
+      }
+    },
     {
       name: "acp-export",
       options: {
@@ -18332,7 +18868,7 @@ async function statusReport(runtime, ctx) {
   const sentTokens = estimateTokens(coreMessages, coveredIds, imageTokens) + systemPromptTokens;
   const viewSentTokens = adjustedTokenCount(runtime.core, coreMessages, state, config, sentTokens, imageTokens, systemPromptTokens);
   const turn = runtime.core.processTurn({ messages: coreMessages, state, config, tokenCount: anchorStale ? viewSentTokens : Math.max(viewSentTokens, realUsage?.tokens ?? 0) });
-  const versionStr = "0.1.71" ? `billion-context-pi@${"0.1.71"} \xB7 pack: ${resolveSurfaceMeta(runtime.adapter, ctx?.cwd ?? process.cwd(), ctx?.model?.provider, ctx?.model?.id).pack}` : void 0;
+  const versionStr = "0.1.72" ? `billion-context-pi@${"0.1.72"} \xB7 pack: ${resolveSurfaceMeta(runtime.adapter, ctx?.cwd ?? process.cwd(), ctx?.model?.provider, ctx?.model?.id).pack}` : void 0;
   let text = buildStatusPanel({
     version: versionStr,
     tokenCount: sessionTokens,
@@ -18944,7 +19480,7 @@ async function autoInstallLatest(latest, extDirOverride) {
       "--no-fund",
       "--no-save"
     ];
-    const prevVersion = (await readPackageJson(join15(extDir, "package.json")))?.version ?? "0.1.71";
+    const prevVersion = (await readPackageJson(join15(extDir, "package.json")))?.version ?? "0.1.72";
     const { code, stderr } = await runNpmImpl(installArgs(latest), { cwd: npmDir, timeout: 6e4 });
     if (code !== 0) {
       logWarn("update", {
@@ -18962,7 +19498,7 @@ async function autoInstallLatest(latest, extDirOverride) {
     }
     const verify = await verifyInstall(npmDir, latest);
     if (!verify.ok) {
-      const rollbackTo = SEMVER_RE.test(prevVersion) ? prevVersion : "0.1.71";
+      const rollbackTo = SEMVER_RE.test(prevVersion) ? prevVersion : "0.1.72";
       logWarn("update", { event: "auto-install-verify-failed", latest, reason: verify.reason, rollbackTo });
       const rb = await runNpmImpl(installArgs(rollbackTo), { cwd: npmDir, timeout: 6e4 });
       logInfo("update", { event: "rollback", from: latest, to: rollbackTo, ok: rb.code === 0 });
@@ -19037,7 +19573,7 @@ async function checkForUpdate(autoUpdate, notify) {
     }
     const latest = await fetchLatestVersion(tag);
     if (!latest) return;
-    const current = runtimeVersion ?? "0.1.71";
+    const current = runtimeVersion ?? "0.1.72";
     const hasUpdate = isVersionNewer(latest, current);
     debug.event("update-check", {
       current,
@@ -19115,6 +19651,13 @@ var PROXY_STAND_DOWN_MESSAGE = [
   "If your traffic reaches the proxy via HTTPS_PROXY instead (no /bili/ prefix in the URL), export BILLION_CONTEXT_PROXY=1 before starting pi.",
   "Docs: https://github.com/ranxianglei/billion-context"
 ].join("\n");
+function nativeStandDownMessage(host) {
+  return [
+    `[billion-context-pi] billion-context native host detected (BILLION_CONTEXT_NATIVE=${host}) \u2014 the in-process native plugin proxies model API traffic and owns compression.`,
+    "ACP client-side compression has been disabled for this session to avoid double compression; nothing else to configure.",
+    "Docs: https://github.com/ranxianglei/billion-context"
+  ].join("\n");
+}
 
 // src/index.ts
 function createAcpExtension(adapter = {}) {
@@ -19128,16 +19671,18 @@ function createAcpExtension(adapter = {}) {
       return;
     }
     const runtime = createRuntime(adapter);
-    let proxyWarned = false;
+    let standDownWarned = false;
     const standDownIfProxied = (ctx) => {
-      if (!isBiliProxyBaseUrl(ctx.model?.baseUrl)) return false;
+      const nativeHost = process.env.BILLION_CONTEXT_NATIVE || void 0;
+      if (nativeHost === void 0 && !process.env.BILLION_CONTEXT_PROXY && !isBiliProxyBaseUrl(ctx.model?.baseUrl)) return false;
+      const message = nativeHost !== void 0 ? nativeStandDownMessage(nativeHost) : PROXY_STAND_DOWN_MESSAGE;
       runtime.refused = true;
-      runtime.refusalMessage = PROXY_STAND_DOWN_MESSAGE;
-      if (!proxyWarned) {
-        proxyWarned = true;
-        logWarn("host", { event: "proxy-baseurl-detected", sid: ctx.sessionManager.getSessionId(), action: "refused" });
-        if (ctx.hasUI) ctx.ui.notify(PROXY_STAND_DOWN_MESSAGE, "warning");
-        else console.error(PROXY_STAND_DOWN_MESSAGE);
+      runtime.refusalMessage = message;
+      if (!standDownWarned) {
+        standDownWarned = true;
+        logWarn("host", { event: nativeHost !== void 0 ? "native-host-detected" : "proxy-detected", sid: ctx.sessionManager.getSessionId(), action: "refused", native: nativeHost ?? null });
+        if (ctx.hasUI) ctx.ui.notify(message, "warning");
+        else console.error(message);
       }
       return true;
     };
@@ -19154,6 +19699,7 @@ function createAcpExtension(adapter = {}) {
     pi.registerTool(makeDecompressTool(runtime, toolSurface.decompress));
     pi.registerTool(makeSearchTool(runtime, toolSurface.search_context));
     pi.registerTool(makeStatusTool(runtime, toolSurface.acp_status));
+    pi.registerTool(makeCacheTool(runtime, toolSurface.acp_cache));
     for (const { name, options } of makeCommands(runtime, pi)) {
       pi.registerCommand(name, options);
     }
@@ -19229,7 +19775,7 @@ function wireSessionLifecycle(pi, runtime, standDownIfProxied) {
     setDelegatePolicy(DEFAULT_DELEGATE_POLICY);
     const sid = ctx.sessionManager.getSessionId();
     const modelInfo = ctx.model;
-    logInfo("session", { event: "start", sid, cwd: ctx.cwd, debug: runtime.adapter.debug ?? null, version: true ? "0.1.71" : null, model: modelInfo?.id ?? null, modelApi: modelInfo?.api ?? null, contextWindow: modelInfo?.contextWindow ?? null });
+    logInfo("session", { event: "start", sid, cwd: ctx.cwd, debug: runtime.adapter.debug ?? null, version: true ? "0.1.72" : null, model: modelInfo?.id ?? null, modelApi: modelInfo?.api ?? null, contextWindow: modelInfo?.contextWindow ?? null });
     let delegateStoodDown = false;
     try {
       await runtime.reloadConfig(ctx.cwd);