@cydm/pie 2.7.0 → 2.7.2

package/dist/cli.js

@@ -6,6 +6,8 @@ import {
   createBashTool,
   createCliHostCapabilities,
   createCliWebResearchHealthStore,
+  createTraceTransportSink,
+  createTransportDiagnosticsStreamFn,
   formatSize,
   getAgentDir,
   getBuiltinDir,
@@ -18,9 +20,10 @@ import {
   getSessionsDir,
   getSettingsPath,
   getThemesDir,
+  logTransportLifecycle,
   migrateConfigFromAgentDir,
   truncateHeadTail
-} from "./chunks/chunk-IPNAFMOJ.js";
+} from "./chunks/chunk-GECAIKZL.js";
 import {
   resolveCliProjectRoot
 } from "./chunks/chunk-WLFOBECN.js";
@@ -29,6 +32,7 @@ import {
   AgentSessionController,
   AttentionCoordinator,
   BUILTIN_TOOL_CAPABILITY_METADATA,
+  DEFAULT_COMPACTION_CONFIG,
   ExtensionResourceScope,
   FileSessionStore,
   SessionTraceSnapshotWriter,
@@ -72,6 +76,7 @@ import {
   maybeAdvanceTodoExecutionState,
   modalityForReadFileUnderstandingKind,
   pauseExecutionState,
+  prepareSessionCompaction,
   registerInteractionHandler,
   requestInteraction,
   resolveExtensionIdentity,
@@ -79,9 +84,10 @@ import {
   setEnablementInMap,
   setExtensionEnabledInSettings,
   shouldPreserveExecutionStateForUserText,
-  supersedeExecutionState
-} from "./chunks/chunk-6RBSTII7.js";
-import "./chunks/chunk-TWILEFLD.js";
+  supersedeExecutionState,
+  verifyPreparedCompactionCommit
+} from "./chunks/chunk-65B6P5J5.js";
+import "./chunks/chunk-GZDAIW6F.js";
 import {
   Deref,
   EMPTY_ASSISTANT_CONTENT_ERROR,
@@ -106,7 +112,7 @@ import {
   sortToolModelCandidatesByCapability,
   streamSimple,
   type_exports
-} from "./chunks/chunk-R5I2QLF2.js";
+} from "./chunks/chunk-BRFP5WZ6.js";
 import {
   IsArray,
   IsBigInt,
@@ -52509,6 +52515,9 @@ function validateCompat(value, label) {
     "store",
     "supportsEncryptedReasoning",
     "responsesContinuation",
+    "supportsReasoningSummary",
+    "reasoningSummary",
+    "timeoutReplay",
     "supportsStore",
     "supportsDeveloperRole",
     "supportsReasoningEffort",
@@ -52525,10 +52534,16 @@ function validateCompat(value, label) {
   if (value.forcedReasoningEffort !== void 0 && value.forcedReasoningEffort !== "none") {
     throw new Error(`${label}.forcedReasoningEffort must be "none"`);
   }
+  if (value.reasoningSummary !== void 0 && !["auto", "concise", "detailed", "none"].includes(String(value.reasoningSummary))) {
+    throw new Error(`${label}.reasoningSummary must be auto | concise | detailed | none`);
+  }
   for (const field of [
     "store",
     "supportsEncryptedReasoning",
     "responsesContinuation",
+    "supportsReasoningSummary",
+    "reasoningSummary",
+    "timeoutReplay",
     "supportsStore",
     "supportsDeveloperRole",
     "supportsReasoningEffort",
@@ -52576,8 +52591,8 @@ function validateRemoteModel(profileId, model) {
     throw new Error(`Remote profile "${profileId}" model "${model.id}" has invalid input capabilities`);
   }
   if (!isPlainRecord(model.cost)) throw new Error(`Remote profile "${profileId}" model "${model.id}" has invalid cost values`);
-  assertOnlyKeys(model.cost, ["input", "output", "cacheRead", "cacheWrite"], `Remote profile "${profileId}" model "${model.id}" cost`);
-  if ([model.cost.input, model.cost.output, model.cost.cacheRead, model.cost.cacheWrite].some((value) => typeof value !== "number" || !Number.isFinite(value) || value < 0)) {
+  assertOnlyKeys(model.cost, ["input", "output", "cacheRead", "cacheWrite", "cacheWriteLong"], `Remote profile "${profileId}" model "${model.id}" cost`);
+  if ([model.cost.input, model.cost.output, model.cost.cacheRead, model.cost.cacheWrite, model.cost.cacheWriteLong].some((value) => value !== void 0 && (typeof value !== "number" || !Number.isFinite(value) || value < 0))) {
     throw new Error(`Remote profile "${profileId}" model "${model.id}" has invalid cost values`);
   }
   if (!Number.isSafeInteger(model.contextWindow) || Number(model.contextWindow) <= 0 || !Number.isSafeInteger(model.maxTokens) || Number(model.maxTokens) <= 0 || Number(model.maxTokens) > Number(model.contextWindow)) {
@@ -52689,7 +52704,8 @@ function sanitizeRemotePayload(decoded, remote, catalogUrl, allowLocallyInjected
           input: model.cost.input,
           output: model.cost.output,
           cacheRead: model.cost.cacheRead,
-          cacheWrite: model.cost.cacheWrite
+          cacheWrite: model.cost.cacheWrite,
+          ...model.cost.cacheWriteLong === void 0 ? {} : { cacheWriteLong: model.cost.cacheWriteLong }
         },
         contextWindow: model.contextWindow,
         maxTokens: model.maxTokens,
@@ -52756,8 +52772,8 @@ function validateModelOverrides(value) {
     if (override.maxTokens !== void 0 && (!Number.isInteger(override.maxTokens) || Number(override.maxTokens) < 1)) throw new Error(`models.json modelOverrides["${key}"].maxTokens is invalid`);
     if (override.cost !== void 0) {
       if (!isPlainRecord(override.cost)) throw new Error(`models.json modelOverrides["${key}"].cost must be an object`);
-      for (const field of ["input", "output", "cacheRead", "cacheWrite"]) {
-        if (override.cost[field] !== void 0 && (typeof override.cost[field] !== "number" || Number.isNaN(override.cost[field]))) throw new Error(`models.json modelOverrides["${key}"].cost.${field} must be a number`);
+      for (const field of ["input", "output", "cacheRead", "cacheWrite", "cacheWriteLong"]) {
+        if (override.cost[field] !== void 0 && (typeof override.cost[field] !== "number" || !Number.isFinite(override.cost[field]) || override.cost[field] < 0)) throw new Error(`models.json modelOverrides["${key}"].cost.${field} must be a non-negative finite number`);
       }
     }
     if (override.headers !== void 0) {
@@ -53856,6 +53872,16 @@ var SettingsManager = class {
     this.settings.autoCompactReserveTokens = tokens;
     this.writeSettings();
   }
+  /** Undefined means "no CLI override": the provider default applies. */
+  getPromptCacheRetention() {
+    const retention = this.settings.promptCacheRetention;
+    return retention === "none" || retention === "short" || retention === "long" ? retention : void 0;
+  }
+  setPromptCacheRetention(retention) {
+    if (retention === void 0) delete this.settings.promptCacheRetention;
+    else this.settings.promptCacheRetention = retention;
+    this.writeSettings();
+  }
   getWebSearchMode() {
     const mode = this.settings.tools?.webSearch?.mode;
     if (mode === "disabled" || mode === "cached" || mode === "live") {
@@ -55575,10 +55601,11 @@ async function forkStoredSession(params) {
         timestamp: entry.timestamp
       };
       if (copy.compactionData) {
+        const selfAnchored = copy.compactionData.firstKeptEntryId === entry.id;
         const keptId = idMap.get(copy.compactionData.firstKeptEntryId);
         copy.compactionData = {
           ...copy.compactionData,
-          firstKeptEntryId: keptId ?? firstCopiedId ?? copy.id
+          firstKeptEntryId: selfAnchored ? copy.id : keptId ?? firstCopiedId ?? copy.id
         };
       }
       idMap.set(summary.id, copy.id);
@@ -65073,6 +65100,23 @@ function buildCliFileAccessOptions(params) {
   };
 }
 
+// src/runtime/computer-use-owner.ts
+var COMPUTER_USE_OWNER_ENV = "PIE_COMPUTER_USE_OWNER_ID";
+var COMPUTER_USE_EXECUTION_ENV = "PIE_COMPUTER_USE_EXECUTION_ID";
+var COMPUTER_USE_STANDALONE_OWNER = "standalone";
+var OWNER_PATTERN = /^[A-Za-z0-9][A-Za-z0-9._:@-]{0,255}$/;
+function normalizeComputerUseOwner(value) {
+  if (typeof value !== "string") return COMPUTER_USE_STANDALONE_OWNER;
+  const trimmed = value.trim();
+  if (!trimmed) return COMPUTER_USE_STANDALONE_OWNER;
+  return OWNER_PATTERN.test(trimmed) ? trimmed : COMPUTER_USE_STANDALONE_OWNER;
+}
+function computerUseOwnerEnv(ownerId, executionId) {
+  const env = { [COMPUTER_USE_OWNER_ENV]: normalizeComputerUseOwner(ownerId) };
+  if (executionId) env[COMPUTER_USE_EXECUTION_ENV] = executionId;
+  return env;
+}
+
 // src/theme/theme.ts
 import * as fs13 from "node:fs";
 import * as path12 from "node:path";
@@ -70243,19 +70287,25 @@ var SessionTraceWriter = class extends SessionTraceSnapshotWriter {
 import { createHash as createHash4 } from "crypto";
 function applyContextSignalsToFooter(target) {
   if (!target.agent?.state) return;
+  const state = target.agent.state;
+  const route = {
+    model: state.model ? `${state.model.provider ?? ""}/${state.model.id ?? ""}` : void 0,
+    thinkingLevel: state.thinkingLevel
+  };
   const signals = computeContextSignals(
     target,
-    target.agent.state.systemPrompt ?? "",
-    target.agent.state.tools,
-    target.sessionManager.getMessages()
+    state.systemPrompt ?? "",
+    state.tools,
+    target.sessionManager.getMessages(),
+    route
   );
   target.footer.setState({
     prefixHash: signals.prefixHash,
     prefixHashChanged: signals.prefixHashChanged,
-    cacheHitRate: signals.cacheHitRate
+    cacheHitRate: signals.recentCacheHitRate ?? signals.cacheHitRate
   });
 }
-function computePromptPrefixHash(systemPrompt, tools) {
+function computePromptPrefixHash(systemPrompt, tools, route) {
   return createHash4("sha256").update(
     JSON.stringify({
       systemPrompt,
@@ -70263,21 +70313,31 @@ function computePromptPrefixHash(systemPrompt, tools) {
         name: tool.name,
         description: tool.description ?? "",
         parameters: tool.parameters ?? null
-      }))
+      })),
+      route: { model: route?.model ?? "", thinkingLevel: route?.thinkingLevel ?? "" }
     })
   ).digest("hex").slice(0, 8);
 }
-function calculateCumulativeCacheHitRate(messages) {
-  let input = 0;
-  let cacheRead = 0;
+function calculateCumulativeCacheHitRate(messages, recentTurns) {
+  const usages = [];
   for (const message of messages) {
     if (message.role !== "assistant") continue;
     const usage = message.usage;
     if (!usage) continue;
+    usages.push(usage);
+  }
+  const windowed = recentTurns !== void 0 && usages.length > recentTurns ? usages.slice(-recentTurns) : usages;
+  let input = 0;
+  let cacheRead = 0;
+  let cacheWrite = 0;
+  let cacheWriteLong = 0;
+  for (const usage of windowed) {
     input += usage.input;
     cacheRead += usage.cacheRead;
+    cacheWrite += usage.cacheWrite;
+    cacheWriteLong += usage.cacheWriteLong ?? 0;
   }
-  const base = input + cacheRead;
+  const base = input + cacheRead + cacheWrite + cacheWriteLong;
   return base > 0 ? cacheRead / base : null;
 }
 var signalMemories = /* @__PURE__ */ new WeakMap();
@@ -70285,14 +70345,16 @@ var lastSignals = /* @__PURE__ */ new WeakMap();
 function getLastContextSignals(hostKey) {
   return lastSignals.get(hostKey) ?? null;
 }
-function computeContextSignals(hostKey, systemPrompt, tools, messages) {
-  const prefixHash = computePromptPrefixHash(systemPrompt, tools);
+var RECENT_HIT_RATE_TURNS = 5;
+function computeContextSignals(hostKey, systemPrompt, tools, messages, route) {
+  const routeKey = `${route?.model ?? ""}|${route?.thinkingLevel ?? ""}`;
+  const prefixHash = computePromptPrefixHash(systemPrompt, tools, route);
   const systemHash = computePromptPrefixHash(systemPrompt, []);
   const toolsHash = computePromptPrefixHash("", tools);
   const previous = signalMemories.get(hostKey);
   const prefixHashChanged = previous !== void 0 && previous.hash !== prefixHash;
   if (previous?.hash !== prefixHash) {
-    signalMemories.set(hostKey, { hash: prefixHash, systemHash, toolsHash });
+    signalMemories.set(hostKey, { hash: prefixHash, systemHash, toolsHash, routeKey });
   }
   const signals = {
     prefixHash,
@@ -70300,7 +70362,9 @@ function computeContextSignals(hostKey, systemPrompt, tools, messages) {
     ...prefixHashChanged && previous ? { previousPrefixHash: previous.hash } : {},
     systemPromptChanged: prefixHashChanged && previous !== void 0 && previous.systemHash !== systemHash,
     toolsChanged: prefixHashChanged && previous !== void 0 && previous.toolsHash !== toolsHash,
-    cacheHitRate: calculateCumulativeCacheHitRate(messages)
+    routeChanged: prefixHashChanged && previous !== void 0 && previous.routeKey !== routeKey,
+    cacheHitRate: calculateCumulativeCacheHitRate(messages),
+    recentCacheHitRate: calculateCumulativeCacheHitRate(messages, RECENT_HIT_RATE_TURNS)
   };
   lastSignals.set(hostKey, signals);
   return signals;
@@ -70310,10 +70374,11 @@ function computeContextSignals(hostKey, systemPrompt, tools, messages) {
 function noteCacheInvalidationInChat(host, signals) {
   if (!signals || !signals.prefixHashChanged) return;
   const parts = [];
+  if (signals.routeChanged) parts.push(signals.systemPromptChanged || signals.toolsChanged ? "route + prompt rebuilt" : "model/thinking route changed");
   if (signals.systemPromptChanged && signals.toolsChanged) parts.push("system prompt + tools rebuilt");
   else if (signals.toolsChanged) parts.push("tools rebuilt");
   else if (signals.systemPromptChanged) parts.push("system prompt rebuilt");
-  else parts.push("prefix rebuilt");
+  if (parts.length === 0) parts.push("prefix rebuilt");
   const from = signals.previousPrefixHash ?? "?";
   const line = `\u26A1 Prompt cache invalidated: \u2301${from} \u2192 \u2301${signals.prefixHash} (${parts.join(", ")})`;
   if (host.chatContainer && host.ui) {
@@ -70325,7 +70390,8 @@ function noteCacheInvalidationInChat(host, signals) {
     from,
     to: signals.prefixHash,
     systemPromptChanged: signals.systemPromptChanged,
-    toolsChanged: signals.toolsChanged
+    toolsChanged: signals.toolsChanged,
+    routeChanged: signals.routeChanged
   });
 }
 function applyContextSignalsWithTrace(host) {
@@ -70534,6 +70600,11 @@ function formatReportTokens(count) {
   if (count < 1e7) return `${(count / 1e6).toFixed(1)}M`;
   return `${Math.round(count / 1e6)}M`;
 }
+function formatMeasuredTime(timestamp) {
+  const date = new Date(timestamp);
+  const pad = (value) => String(value).padStart(2, "0");
+  return `${pad(date.getHours())}:${pad(date.getMinutes())}:${pad(date.getSeconds())}`;
+}
 function buildSystemPromptBlocks(systemPrompt, skillsSection, knowledgeSection) {
   const totalTokens = estimateStringTokens(systemPrompt);
   const skillsTokens = skillsSection && systemPrompt.includes(skillsSection.trim()) ? estimateStringTokens(skillsSection) : null;
@@ -70789,8 +70860,12 @@ ${file.content}`).join("\n\n");
     messagesTokens,
     dynamicContextTokens: 0,
     estimatedTotalTokens,
-    actualTokens: inputs.actualTokens,
+    estimatedCurrentTokens: inputs.estimatedCurrentTokens,
+    providerMeasuredTokens: inputs.providerMeasuredTokens,
+    providerMeasuredAt: inputs.providerMeasuredAt ?? null,
+    measurementMatchesCurrentContext: inputs.measurementMatchesCurrentContext,
     contextWindow: inputs.contextWindow,
+    autoCompactEnabled: inputs.autoCompactEnabled,
     autoCompactLimitTokens: inputs.autoCompactLimitTokens,
     reserveTokens: inputs.reserveTokens,
     promptPrefixHash: inputs.promptPrefixHash,
@@ -70847,13 +70922,14 @@ function buildContextOverviewText(data, themeInput) {
       }
     }
   });
-  const autoCompact = data.autoCompactLimitTokens ? `triggers > ~${formatReportTokens(data.autoCompactLimitTokens)} (reserve ${formatReportTokens(data.reserveTokens)})` : "disabled";
+  const autoCompact = !data.autoCompactEnabled ? "Off (disabled in settings)" : data.autoCompactLimitTokens ? `On \xB7 triggers > ~${formatReportTokens(data.autoCompactLimitTokens)} (reserve ${formatReportTokens(data.reserveTokens)})` : "On \xB7 threshold unknown (context window unavailable)";
   lines.push(
     `${padEndVisible(` ${branch(sections.length)} auto-compact`, labelWidth)}${theme2.fg("dim", autoCompact)}`
   );
-  const reconciliation = data.actualTokens !== null ? `actual ${formatReportTokens(data.actualTokens)} (last response) \xB7 est\u2212actual \u2248 ${formatReportTokens(
-    Math.abs(data.estimatedTotalTokens - data.actualTokens)
-  )}${data.estimatedTotalTokens >= data.actualTokens ? " overhead" : " undercount"}` : "actual unknown until the next model response";
+  const measuredAt = data.providerMeasuredAt !== null ? ` \xB7 measured ${formatMeasuredTime(data.providerMeasuredAt)}` : "";
+  const reconciliation = data.providerMeasuredTokens !== null ? data.measurementMatchesCurrentContext ? `est current ~${formatReportTokens(data.estimatedCurrentTokens)} \xB7 provider-measured ${formatReportTokens(data.providerMeasuredTokens)} (last response${measuredAt}) \xB7 est\u2212measured \u2248 ${formatReportTokens(
+    Math.abs(data.estimatedCurrentTokens - data.providerMeasuredTokens)
+  )}${data.estimatedCurrentTokens >= data.providerMeasuredTokens ? " overhead" : " undercount"}` : `est current ~${formatReportTokens(data.estimatedCurrentTokens)} \xB7 provider-measured ${formatReportTokens(data.providerMeasuredTokens)} (last response${measuredAt}) \xB7 not comparable \u2014 context changed since measurement` : `est current ~${formatReportTokens(data.estimatedCurrentTokens)} \xB7 provider usage not reported yet`;
   lines.push(`  ${theme2.fg("dim", reconciliation)}`);
   lines.push(theme2.fg("dim", "  /debugs/context \xB7 Enter to drill into a section, Esc to go back"));
   return lines.join("\n");
@@ -70872,23 +70948,37 @@ function buildWindowRows(data) {
       label: "estimated total (prompt + tools + messages)",
       estTokens: data.estimatedTotalTokens,
       note: `prompt ~${formatReportTokens(data.systemPromptTokens)} \xB7 tools ~${formatReportTokens(data.toolsTokens)} \xB7 messages ~${formatReportTokens(data.messagesTokens)}`
+    },
+    {
+      id: "window:estimated-current",
+      label: "estimated current context (local estimate + tail)",
+      estTokens: data.estimatedCurrentTokens,
+      note: "local estimate \u2014 not a provider measurement"
     }
   ];
-  if (data.actualTokens !== null) {
+  if (data.providerMeasuredTokens !== null) {
+    const measuredAt = data.providerMeasuredAt !== null ? formatMeasuredTime(data.providerMeasuredAt) : "unknown time";
     rows.push({
-      id: "window:actual",
-      label: "actual (last model response)",
-      estTokens: data.actualTokens,
-      valueText: formatReportTokens(data.actualTokens),
-      note: `gap \u2248 ${formatReportTokens(Math.abs(data.estimatedTotalTokens - data.actualTokens))}`
+      id: "window:provider-measured",
+      label: "provider-measured (last response)",
+      estTokens: data.providerMeasuredTokens,
+      valueText: formatReportTokens(data.providerMeasuredTokens),
+      note: data.measurementMatchesCurrentContext ? `est\u2212measured \u2248 ${formatReportTokens(Math.abs(data.estimatedCurrentTokens - data.providerMeasuredTokens))} \xB7 ${measuredAt}` : `not comparable \u2014 context changed since measurement \xB7 ${measuredAt}`
     });
   }
   rows.push({
     id: "window:auto-compact",
+    label: "auto-compact",
+    estTokens: 0,
+    valueText: data.autoCompactEnabled ? "On" : "Off",
+    note: data.autoCompactEnabled ? "enabled in settings" : "disabled in settings"
+  });
+  rows.push({
+    id: "window:auto-compact-threshold",
     label: "auto-compact threshold",
-    estTokens: data.autoCompactLimitTokens ?? 0,
-    valueText: data.autoCompactLimitTokens ? `>${formatReportTokens(data.autoCompactLimitTokens)}` : "off",
-    note: data.autoCompactLimitTokens ? `reserve ${formatReportTokens(data.reserveTokens)}` : "disabled"
+    estTokens: data.autoCompactEnabled ? data.autoCompactLimitTokens ?? 0 : 0,
+    valueText: data.autoCompactEnabled ? data.autoCompactLimitTokens ? `>${formatReportTokens(data.autoCompactLimitTokens)}` : "unknown" : "\u2014",
+    note: data.autoCompactEnabled ? data.autoCompactLimitTokens ? `reserve ${formatReportTokens(data.reserveTokens)}` : "context window unavailable" : "not applied while Off"
   });
   if (data.promptPrefixHash) {
     rows.push({
@@ -71300,11 +71390,20 @@ var TreeSelectorComponent = class {
       this.statusMessage = null;
     }
   }
+  /** Cursor on the resident entry with the highest seq (the newest), not the last rendered row. */
+  selectNewestResidentRow(message) {
+    if (this.filteredRows.length === 0) return;
+    let newestIndex = 0;
+    for (let i = 1; i < this.filteredRows.length; i++) {
+      if (this.filteredRows[i].summary.seq > this.filteredRows[newestIndex].summary.seq) newestIndex = i;
+    }
+    this.selectedIndex = newestIndex;
+    if (message !== null) this.statusMessage = message;
+  }
   /** Jump to the newest page. Seqs are 1..total (append-only journal), so the last window starts at total - pageSize. */
   jumpToLastPage() {
     if (this.pageSummaries.length > 0 && !this.hasNext) {
-      this.selectedIndex = Math.max(0, this.filteredRows.length - 1);
-      this.statusMessage = "latest page - at newest entry";
+      this.selectNewestResidentRow("latest page - at newest entry");
       return;
     }
     let anchor = Math.max(0, this.totalEntries - this.pageSizeValue);
@@ -71314,8 +71413,7 @@ var TreeSelectorComponent = class {
           anchor = this.pageSummaries[this.pageSummaries.length - 1].seq;
           continue;
         }
-        this.selectedIndex = Math.max(0, this.filteredRows.length - 1);
-        this.statusMessage = null;
+        this.selectNewestResidentRow(null);
         return;
       }
       if (anchor === 0 || !this.hasNext) break;
@@ -72781,7 +72879,9 @@ function calculateSessionUsageTotals(messages) {
   let output = 0;
   let cacheRead = 0;
   let cacheWrite = 0;
+  let cacheWriteLong = 0;
   let cost = 0;
+  let costEstimated = false;
   for (const message of messages) {
     if (message.role !== "assistant") continue;
     const usage = message.usage;
@@ -72790,9 +72890,11 @@ function calculateSessionUsageTotals(messages) {
     output += usage.output;
     cacheRead += usage.cacheRead;
     cacheWrite += usage.cacheWrite;
+    cacheWriteLong += usage.cacheWriteLong ?? 0;
     cost += usage.cost?.total ?? 0;
+    if (usage.cost?.cacheWriteLongEstimated === true) costEstimated = true;
   }
-  return { input, output, cacheRead, cacheWrite, cost };
+  return { input, output, cacheRead, cacheWrite, cacheWriteLong, cacheWriteTotal: cacheWrite + cacheWriteLong, cost, costEstimated };
 }
 function formatTokensCompact(count) {
   if (count < 1e3) return String(count);
@@ -72838,16 +72940,18 @@ ${options.theme.bold("Usage (active context)")}
         info += `${options.theme.fg("dim", "Cache Read:")} ${totals.cacheRead.toLocaleString()}
 `;
       }
-      if (totals.cacheWrite > 0) {
-        info += `${options.theme.fg("dim", "Cache Write:")} ${totals.cacheWrite.toLocaleString()}
+      if (totals.cacheWriteTotal > 0) {
+        const split = totals.cacheWriteLong > 0 ? options.theme.fg("dim", ` (short ${totals.cacheWrite.toLocaleString()} \xB7 long ${totals.cacheWriteLong.toLocaleString()})`) : "";
+        info += `${options.theme.fg("dim", "Cache Write:")} ${totals.cacheWriteTotal.toLocaleString()}${split}
 `;
       }
       if (totals.cost > 0) {
-        info += `${options.theme.fg("dim", "Cost:")} $${totals.cost.toFixed(4)}
+        const estimateNote = totals.costEstimated ? " (long-TTL estimate)" : "";
+        info += `${options.theme.fg("dim", "Cost:")} $${totals.cost.toFixed(4)}${estimateNote}
 `;
       }
-      const hitRateBase = totals.input + totals.cacheRead;
-      if (hitRateBase > 0 && totals.cacheRead > 0) {
+      const hitRateBase = totals.input + totals.cacheRead + totals.cacheWriteTotal;
+      if (hitRateBase > 0) {
         const hitRate = (totals.cacheRead / hitRateBase * 100).toFixed(1);
         info += `${options.theme.fg("dim", "Cache Hit Rate:")} ${hitRate}% (~${formatTokensCompact(hitRateBase)} served)
 `;
@@ -72981,7 +73085,9 @@ async function loadSession(host, sessionId) {
       outputTokens: contextUsage.output,
       cacheRead: contextUsage.cacheRead,
       cacheWrite: contextUsage.cacheWrite,
+      cacheWriteLong: contextUsage.cacheWriteLong,
       cost: contextUsage.cost,
+      costEstimated: contextUsage.costEstimated,
       contextPercent,
       contextWindow
     });
@@ -74019,7 +74125,9 @@ function bindInteractiveRuntimeEvents(host) {
           outputTokens: contextUsage.output,
           cacheRead: contextUsage.cacheRead,
           cacheWrite: contextUsage.cacheWrite,
+          cacheWriteLong: contextUsage.cacheWriteLong,
           cost: contextUsage.cost,
+          costEstimated: contextUsage.costEstimated,
           contextPercent,
           contextWindow
         });
@@ -74200,6 +74308,9 @@ function isControllerRecovering(host) {
 function handleMessageStart(host, event) {
   if (event.message?.role === "assistant") {
     host.clearContinuationTimeout("assistant_message_start");
+    for (const [id, component] of host.pendingTools) {
+      if (!component.calm) host.pendingTools.delete(id);
+    }
     host.streamingComponent = new AssistantMessageComponent(void 0, host.thinkingLevel === "off");
     host.streamingComponent.setCalmMode(host.calmMode);
     host.streamingMessageHadToolCalls = false;
@@ -74771,7 +74882,7 @@ function listDiscoveredSkills(host) {
 function applyReloadedSkills(host, result) {
   host.skills = result.skills.filter((skill) => host.options.settingsManager.isSkillEnabled(skill.name));
   host.skillsSection = formatSkillsForPrompt(host.skills);
-  host.applyRuntimeToolMode(host.allTools.length > 0 ? host.allTools : host.agent.state.tools);
+  host.applyRuntimeToolMode();
   host.updateSystemPromptTools();
   for (const skill of host.skills) {
     const skillPath = ["skills", `skill:${skill.name}`];
@@ -75199,6 +75310,23 @@ function handleToolsCommand(host, args) {
 var RESET2 = "\x1B[0m";
 var INVERSE2 = "\x1B[7m";
 var ROLE_FILTER_CYCLE = ["all", "user", "assistant", "toolResult", "compaction"];
+function buildWrappedLines(rawLines, contentWidth) {
+  const width = Math.max(1, contentWidth);
+  const lines = [];
+  const sourceStart = [];
+  for (const rawLine of rawLines) {
+    sourceStart.push(lines.length);
+    lines.push(...wrapTextWithAnsi(rawLine, width));
+  }
+  return { lines: lines.length > 0 ? lines : [""], sourceStart };
+}
+function anchorFor(scroll, sourceStart, sourceCount) {
+  if (sourceStart.length === 0) return { source: 0, offset: 0 };
+  const target = Math.max(0, Math.min(scroll, sourceCount - 1));
+  let source = sourceStart.length - 1;
+  while (source > 0 && sourceStart[source] > target) source -= 1;
+  return { source, offset: target - sourceStart[source] };
+}
 function createLine2(content, width, isSelected) {
   const normalized = content.replace(/[\r\n]+/g, " ");
   const body = isSelected ? INVERSE2 + normalized : normalized;
@@ -75339,12 +75467,33 @@ var ContextReportSelectorComponent = class {
     this.views.push({
       kind: "text",
       title,
-      lines: text.replace(/\r\n/g, "\n").split("\n"),
+      rawLines: text.replace(/\r\n/g, "\n").split("\n"),
+      lines: [],
+      sourceStart: [],
+      wrapWidth: 0,
       scroll: 0
     });
     this.selectedIndex = 0;
     this.statusMessage = null;
   }
+  /** Re-wrap a text view for the current terminal width, preserving the anchor. */
+  ensureTextWrapped(view, width) {
+    const contentWidth = Math.max(1, width - 2);
+    if (view.wrapWidth === contentWidth) return;
+    const previousCount = view.lines.length;
+    const anchor = view.wrapWidth > 0 && previousCount > 0 ? anchorFor(view.scroll, view.sourceStart, previousCount) : { source: 0, offset: 0 };
+    const rebuilt = buildWrappedLines(view.rawLines, contentWidth);
+    view.lines = rebuilt.lines;
+    view.sourceStart = rebuilt.sourceStart;
+    view.wrapWidth = contentWidth;
+    const source = Math.min(anchor.source, Math.max(0, rebuilt.sourceStart.length - 1));
+    const nextSourceStart = rebuilt.sourceStart[source + 1] ?? rebuilt.lines.length;
+    const sourceLineCount = Math.max(1, nextSourceStart - rebuilt.sourceStart[source]);
+    view.scroll = Math.min(
+      Math.max(0, rebuilt.lines.length - 1),
+      rebuilt.sourceStart[source] + Math.min(anchor.offset, sourceLineCount - 1)
+    );
+  }
   drillIntoSelected() {
     const view = this.currentView;
     if (view.kind !== "list") return;
@@ -75526,6 +75675,8 @@ var ContextReportSelectorComponent = class {
         )
       );
     } else {
+      this.ensureTextWrapped(view, width);
+      view.scroll = Math.max(0, Math.min(view.scroll, view.lines.length - 1));
       const total = view.lines.length;
       const start = Math.min(view.scroll, Math.max(0, total - 1));
       const end = Math.min(start + this.maxVisibleLines, total);
@@ -75690,14 +75841,20 @@ function handleContextReportCommand(host) {
   const usage = host.calculateContextUsage();
   const contextWindow = host.options.model?.contextWindow ?? 0;
   const reserveTokens = host.options.settingsManager.getAutoCompactReserveTokens();
+  const autoCompactEnabled = host.options.settingsManager.getAutoCompactEnabled();
   const autoCompactLimit = contextWindow > 0 ? getAutoCompactTokenLimit(contextWindow, reserveTokens) ?? null : null;
+  const state = host.agent.state;
   const prefixHash = computePromptPrefixHash(
     systemPrompt,
-    host.agent.state.tools.map((tool) => ({
+    state.tools.map((tool) => ({
       name: tool.name,
       description: tool.description,
       parameters: tool.parameters
-    }))
+    })),
+    {
+      model: state.model ? `${state.model.provider ?? ""}/${state.model.id ?? ""}` : void 0,
+      thinkingLevel: state.thinkingLevel
+    }
   );
   const lastSignals2 = getLastContextSignals(host);
   const promptPrefixChanged = lastSignals2 !== null && lastSignals2.prefixHash !== prefixHash;
@@ -75721,7 +75878,11 @@ function handleContextReportCommand(host) {
     entries,
     compactionTokensBefore: (entryId) => host.sessionManager.getEntry(entryId)?.compactionData?.tokensBefore,
     contextWindow,
-    actualTokens: usage.tokens,
+    estimatedCurrentTokens: usage.estimatedCurrentTokens,
+    providerMeasuredTokens: usage.providerMeasuredTokens,
+    providerMeasuredAt: usage.providerMeasuredAt,
+    measurementMatchesCurrentContext: usage.measurementMatchesCurrentContext,
+    autoCompactEnabled,
     autoCompactLimitTokens: autoCompactLimit,
     reserveTokens,
     promptPrefixHash: prefixHash,
@@ -76553,7 +76714,9 @@ async function handleClearCommand(host) {
     outputTokens: contextUsage.output,
     cacheRead: contextUsage.cacheRead,
     cacheWrite: contextUsage.cacheWrite,
+    cacheWriteLong: contextUsage.cacheWriteLong,
     cost: contextUsage.cost,
+    costEstimated: contextUsage.costEstimated,
     contextPercent: contextWindow > 0 && contextUsage.tokens !== null ? contextUsage.tokens / contextWindow * 100 : 0,
     contextWindow
   });
@@ -76916,11 +77079,16 @@ async function handleCompactCommand(host) {
       host.showStatus("Nothing to compact yet");
       return;
     }
-    const firstKeptEntryId = findFirstKeptEntrySummaryId(host.sessionManager.getActiveContextSummaries());
-    if (!firstKeptEntryId) {
+    const preparation = prepareSessionCompaction({
+      sessionManager: host.sessionManager,
+      messages: existingMessages,
+      budgetTokens: host.options.model?.contextWindow
+    });
+    if (preparation.entries.length === 0) {
       host.showStatus("Need at least one real user message before compacting");
       return;
     }
+    const tokensBefore = calculateTotalTokens(preparation.messages);
     const compactOptions = {
       model: host.options.model,
       apiKey: host.options.apiKey,
@@ -76939,10 +77107,27 @@ async function handleCompactCommand(host) {
     if (beforeCompact?.summary) {
       summary = beforeCompact.summary;
     } else {
-      summary = await createCompactionSummary(existingMessages, compactOptions);
+      summary = await createCompactionSummary(preparation.messages, compactOptions);
+    }
+    const verified = verifyPreparedCompactionCommit({
+      preparation,
+      sessionManager: host.sessionManager,
+      messages: host.agent.state.messages,
+      result: { summary, retainedTail: preparation.retainedTail.retainedTail, tokensBefore }
+    });
+    if (verified.ok === false) {
+      if (verified.kind === "stale") {
+        host.showStatus("Session changed during compaction; nothing was published");
+        return;
+      }
+      throw new Error(`Compaction abandoned: ${verified.reason}`);
     }
-    const tokensBefore = calculateTotalTokens(existingMessages);
-    host.sessionManager.appendCompaction(summary, firstKeptEntryId, tokensBefore);
+    host.sessionManager.appendCompactionCheckpoint({
+      summary,
+      retainedTail: verified.retainedTail,
+      tokensBefore,
+      excludeFailure: preparation.excludedFailure
+    });
     const compactedMessages = host.sessionManager.getMessages();
     host.replaceAgentMessagesAndMarkSynced(compactedMessages);
     host.resetFileObservationState();
@@ -76967,7 +77152,9 @@ async function handleCompactCommand(host) {
       outputTokens: contextUsage.output,
       cacheRead: contextUsage.cacheRead,
       cacheWrite: contextUsage.cacheWrite,
+      cacheWriteLong: contextUsage.cacheWriteLong,
       cost: contextUsage.cost,
+      costEstimated: contextUsage.costEstimated,
       contextPercent,
       contextWindow
     });
@@ -79350,8 +79537,12 @@ var FooterComponent = class {
     if (state.inputTokens) statsParts.push(`\u2191${formatTokens(state.inputTokens)}`);
     if (state.outputTokens) statsParts.push(`\u2193${formatTokens(state.outputTokens)}`);
     if (state.cacheRead) statsParts.push(`R${formatTokens(state.cacheRead)}`);
-    if (state.cacheWrite) statsParts.push(`W${formatTokens(state.cacheWrite)}`);
-    if (state.cost) statsParts.push(`$${state.cost.toFixed(3)}`);
+    const cacheWriteShort = state.cacheWrite ?? 0;
+    const cacheWriteLong = state.cacheWriteLong ?? 0;
+    const cacheWriteTotal = cacheWriteShort + cacheWriteLong;
+    if (cacheWriteTotal) statsParts.push(`W${formatTokens(cacheWriteTotal)}`);
+    if (cacheWriteLong) statsParts.push(`W1h${formatTokens(cacheWriteLong)}`);
+    if (state.cost) statsParts.push(`${state.costEstimated ? "~" : ""}$${state.cost.toFixed(3)}`);
     if (state.cacheHitRate != null) {
       statsParts.push(theme.fg("dim", `ca ${(state.cacheHitRate * 100).toFixed(0)}%`));
     }
@@ -79501,13 +79692,23 @@ function calculateCliContextUsage(params) {
   const estimate = estimateContextTokens(messages);
   const usageMessage = estimate.lastUsageIndex !== null ? messages[estimate.lastUsageIndex] : void 0;
   const usage = usageMessage?.role === "assistant" ? usageMessage.usage : void 0;
+  const hasProviderMeasurement = usage !== void 0 && estimate.lastUsageIndex !== null;
   return {
     tokens: estimate.tokens,
+    estimatedCurrentTokens: estimate.tokens,
+    providerMeasuredTokens: hasProviderMeasurement ? estimate.providerTokens : null,
+    providerMeasuredAt: hasProviderMeasurement && usageMessage ? usageMessage.timestamp : null,
+    // A provider total describes the context AT that response. Any trailing
+    // tool/user message (or a compaction reset, which zeroes stale usage)
+    // makes the two figures different snapshots, so no error may be shown.
+    measurementMatchesCurrentContext: hasProviderMeasurement && estimate.lastUsageIndex === messages.length - 1,
     input: usage?.input ?? 0,
     output: usage?.output ?? 0,
     cacheRead: usage?.cacheRead ?? 0,
     cacheWrite: usage?.cacheWrite ?? 0,
-    cost: usage?.cost.total ?? 0
+    cacheWriteLong: usage?.cacheWriteLong ?? 0,
+    cost: usage?.cost.total ?? 0,
+    costEstimated: usage?.cost?.cacheWriteLongEstimated === true
   };
 }
 function createCliAutoCompactDecision(params) {
@@ -79538,6 +79739,11 @@ function shouldCliAutoCompactTokens(params) {
   const tokenLimit = getAutoCompactTokenLimit(params.contextWindow, params.reserveTokens);
   return tokenLimit !== void 0 && params.tokens > tokenLimit;
 }
+function createCliCompactionBudget(params) {
+  const contextWindow = params.contextWindow ?? 0;
+  const reserveTokens = params.reserveTokens ?? DEFAULT_COMPACTION_CONFIG.reserveTokens;
+  return () => getAutoCompactTokenLimit(contextWindow, reserveTokens);
+}
 async function createCliAutoCompactionResult(params) {
   if (params.context.signal?.aborted) {
     return { aborted: true };
@@ -79546,8 +79752,9 @@ async function createCliAutoCompactionResult(params) {
   if (!activeEntryId) {
     return { aborted: true };
   }
-  const firstKeptEntryId = findFirstKeptEntrySummaryId(params.context.sessionManager.getActiveContextSummaries());
-  if (!firstKeptEntryId) {
+  const legacyCutId = params.context.preparation ? null : findFirstKeptEntrySummaryId(params.context.sessionManager.getActiveContextSummaries());
+  const retainedTail = params.context.preparation ? params.context.preparation.retainedTail.retainedTail : legacyCutId ? { kind: "from-entry", entryId: legacyCutId } : void 0;
+  if (!retainedTail) {
     return { aborted: true };
   }
   const compactOptions = {
@@ -79562,7 +79769,8 @@ async function createCliAutoCompactionResult(params) {
   }
   return {
     summary,
-    firstKeptEntryId,
+    retainedTail,
+    ...retainedTail.kind === "from-entry" ? { firstKeptEntryId: retainedTail.entryId } : {},
     tokensBefore: calculateTotalTokens(params.context.messages)
   };
 }
@@ -79714,16 +79922,44 @@ function createCliRetryPolicy(isRetryableErrorOverride) {
     maxRetries: CLI_RETRY_MAX_RETRIES,
     baseDelayMs: getCliRetryBaseDelayMs(),
     maxDelayMs: getCliRetryMaxDelayMs(),
-    maxRetriesForFailure: (failure) => failure.kind === "transient_network" ? getCliTransientNetworkMaxRetries() : CLI_RETRY_MAX_RETRIES,
-    maxTotalRetries: getCliRetryTotalMaxRetries()
+    maxRetriesForFailure: (failure) => failure.kind === "transient_network" ? getCliTransientNetworkMaxRetries() : failure.replayExempt === true ? getCliTimeoutReplayMaxRetries() : CLI_RETRY_MAX_RETRIES,
+    maxTotalRetries: getCliRetryTotalMaxRetries(),
+    systemicReplayStallLimit: getCliSystemicReplayStallLimit()
   };
 }
+var CLI_TIMEOUT_REPLAY_MAX_RETRIES = 5;
+function getCliTimeoutReplayMaxRetries() {
+  const override = process.env.PIE_TIMEOUT_REPLAY_MAX_RETRIES;
+  if (!override) return CLI_TIMEOUT_REPLAY_MAX_RETRIES;
+  const parsed = Number(override);
+  return Number.isFinite(parsed) && parsed >= 0 ? Math.floor(parsed) : CLI_TIMEOUT_REPLAY_MAX_RETRIES;
+}
+var CLI_SYSTEMIC_REPLAY_STALL_LIMIT = 3;
+function getCliSystemicReplayStallLimit() {
+  const override = process.env.PIE_SYSTEMIC_REPLAY_STALL_LIMIT;
+  if (!override) return CLI_SYSTEMIC_REPLAY_STALL_LIMIT;
+  const parsed = Number(override);
+  return Number.isFinite(parsed) && parsed >= 1 ? Math.floor(parsed) : CLI_SYSTEMIC_REPLAY_STALL_LIMIT;
+}
 
 // src/runtime/interactive-context-ui.ts
 function calculateContextUsage(host) {
   const messages = host.agent?.state.messages ?? [];
   if (messages.length === 0) {
-    return { tokens: 0, input: 0, output: 0, cacheRead: 0, cacheWrite: 0, cost: 0 };
+    return {
+      tokens: 0,
+      estimatedCurrentTokens: 0,
+      providerMeasuredTokens: null,
+      providerMeasuredAt: null,
+      measurementMatchesCurrentContext: false,
+      input: 0,
+      output: 0,
+      cacheRead: 0,
+      cacheWrite: 0,
+      cacheWriteLong: 0,
+      cost: 0,
+      costEstimated: false
+    };
   }
   return calculateCliContextUsage({
     messages,
@@ -80474,6 +80710,9 @@ function createInteractiveAgentSessionController(host) {
       }
       return void 0;
     },
+    // Same auto-compact limit the threshold decision uses; a checkpoint whose
+    // summary + retained tail would still exceed it is refused.
+    getCompactionBudgetTokens: createCliCompactionBudget({ contextWindow: host.options.model?.contextWindow }),
     getAutoContinueMessage: () => host.getTodoAutoContinueMessage(),
     classifyFailure: (context) => classifyCliFailure(context),
     retry: createCliRetryPolicy(
@@ -80491,9 +80730,16 @@ function createInteractiveAgentSessionController(host) {
           return { aborted: true };
         }
         if (beforeCompact?.summary) {
-          const firstKeptEntryId = findFirstKeptEntrySummaryId(context.sessionManager.getActiveContextSummaries());
-          if (!firstKeptEntryId) return { aborted: true };
-          return { summary: beforeCompact.summary, firstKeptEntryId };
+          const retainedTail = context.preparation?.retainedTail.retainedTail ?? (() => {
+            const legacyCutId = findFirstKeptEntrySummaryId(context.sessionManager.getActiveContextSummaries());
+            return legacyCutId ? { kind: "from-entry", entryId: legacyCutId } : void 0;
+          })();
+          if (!retainedTail) return { aborted: true };
+          return {
+            summary: beforeCompact.summary,
+            retainedTail,
+            ...retainedTail.kind === "from-entry" ? { firstKeptEntryId: retainedTail.entryId } : {}
+          };
         }
         autoCompactBeforeStats = summarizeActiveContext(context.sessionManager.getActiveContextSummaries());
         return createCliAutoCompactionResult({
@@ -80655,6 +80901,13 @@ function createInteractiveAgentSessionController(host) {
         host.isProcessing = true;
         host.footer.setState({ status: "processing", lifecycleEvent: "auto_compact" });
         host.showStatus(event.reason === "overflow" ? "Context overflow detected, compacting..." : "Auto-compacting context...");
+      } else if (event.type === "compaction_committed") {
+        host.resetFileObservationState();
+        host.resetRuntimeModeBlocks?.();
+        host.sessionTrace.noteRuntimeEvent("compaction_committed", {
+          reason: event.reason,
+          willRetry: event.willRetry
+        });
       } else if (event.type === "auto_compaction_end") {
         host.isCompacting = false;
         host.isProcessing = false;
@@ -80667,8 +80920,6 @@ function createInteractiveAgentSessionController(host) {
           autoCompactBeforeStats = null;
           return;
         }
-        host.resetFileObservationState();
-        host.resetRuntimeModeBlocks?.();
         if (!event.result?.aborted) {
           host.streamingComponent = void 0;
           host.pendingTools.clear();
@@ -80687,7 +80938,9 @@ function createInteractiveAgentSessionController(host) {
           outputTokens: contextUsage.output,
           cacheRead: contextUsage.cacheRead,
           cacheWrite: contextUsage.cacheWrite,
+          cacheWriteLong: contextUsage.cacheWriteLong,
           cost: contextUsage.cost,
+          costEstimated: contextUsage.costEstimated,
           contextPercent,
           contextWindow
         });
@@ -80843,20 +81096,80 @@ function currentDateInZone(timeZone) {
   const part = (type) => parts.find((p) => p.type === type)?.value ?? "00";
   return `${part("year")}-${part("month")}-${part("day")}`;
 }
-var RUNTIME_MODE_BLOCK_PATTERN = /^[\t ]*<runtime-mode-context>[\s\S]*?<\/runtime-mode-context>[\t ]*(?:\r?\n){0,2}/;
+var RUNTIME_MODE_BLOCK_START_MARKER = "<runtime-mode-context>";
+function leadingRuntimeBlock(text) {
+  const withoutLeadingWhitespace = text.replace(/^[\t ]+/, "");
+  if (!withoutLeadingWhitespace.startsWith(RUNTIME_MODE_BLOCK_START_MARKER)) return null;
+  const markerStart = text.length - withoutLeadingWhitespace.length;
+  const endIndex = text.indexOf(RUNTIME_MODE_CONTEXT_END_MARKER, markerStart);
+  if (endIndex < 0) return null;
+  const block = text.slice(markerStart, endIndex + RUNTIME_MODE_CONTEXT_END_MARKER.length);
+  const rest = text.slice(endIndex + RUNTIME_MODE_CONTEXT_END_MARKER.length).replace(/^[\t ]*(?:\r?\n){0,2}/, "");
+  return { block, rest };
+}
+function stripLeadingRuntimeBlock(text) {
+  const leading = leadingRuntimeBlock(text);
+  return leading ? { rest: leading.rest } : null;
+}
+function leadingRuntimeBlockOfMessage(message) {
+  if (message.role !== "user") return null;
+  const content = message.content;
+  if (typeof content === "string") return leadingRuntimeBlock(content)?.block ?? null;
+  if (!Array.isArray(content)) return null;
+  const first = content[0];
+  if (first?.type !== "text" || typeof first.text !== "string") return null;
+  return leadingRuntimeBlock(first.text)?.block ?? null;
+}
+function lastUserMessageCarriesRuntimeBlock(messages, block) {
+  for (let index = messages.length - 1; index >= 0; index -= 1) {
+    const message = messages[index];
+    if (message?.role !== "user") continue;
+    return leadingRuntimeBlockOfMessage(message) === block;
+  }
+  return false;
+}
+function applyRuntimeBlockToText(text, block) {
+  const split = stripLeadingRuntimeBlock(text);
+  const rest = split ? split.rest : text;
+  const next = rest.length > 0 ? `${block}
+
+${rest}` : block;
+  return next === text ? text : next;
+}
+function applyRuntimeBlockToContentArray(content, block) {
+  const first = content[0];
+  if (first && typeof first === "object" && first.type === "text" && typeof first.text === "string") {
+    const text = first.text;
+    if (stripLeadingRuntimeBlock(text)) {
+      const nextText = applyRuntimeBlockToText(text, block);
+      if (nextText === text) return content;
+      const copy = [...content];
+      copy[0] = { ...first, text: nextText };
+      return copy;
+    }
+  }
+  return [{ type: "text", text: block }, ...content];
+}
 function refreshRuntimeModeBlockOnLastUserMessage(messages, block) {
   for (let i = messages.length - 1; i >= 0; i--) {
     const message = messages[i];
     if (message?.role !== "user") continue;
-    const content = typeof message.content === "string" ? message.content : "";
-    const stripped = content.replace(RUNTIME_MODE_BLOCK_PATTERN, "");
-    const refreshed = stripped === content && content.includes("<runtime-mode-context>") ? content : `${block}
-
-${stripped}`;
-    if (refreshed === content) return messages;
-    const next = [...messages];
-    next[i] = { ...message, content: refreshed };
-    return next;
+    const content = message.content;
+    if (typeof content === "string") {
+      const next = applyRuntimeBlockToText(content, block);
+      if (next === content) return messages;
+      const copy = [...messages];
+      copy[i] = { ...message, content: next };
+      return copy;
+    }
+    if (Array.isArray(content)) {
+      const next = applyRuntimeBlockToContentArray(content, block);
+      if (next === content) return messages;
+      const copy = [...messages];
+      copy[i] = { ...message, content: next };
+      return copy;
+    }
+    return messages;
   }
   return messages;
 }
@@ -80866,28 +81179,54 @@ var RuntimeModeBlockInjector = class {
   }
   options;
   lastKey = null;
-  pendingRetryRefresh = false;
+  /**
+   * Request-view overlay. While active, the CURRENT block is re-derived on
+   * every request from the live message snapshot and applied to the latest
+   * user message. It is not cached into history and it ends only when a real
+   * user message freezes the current block (see `nextBlock` + the transform),
+   * so consecutive tool rounds after a compaction retry keep the fresh state
+   * instead of losing it after one request.
+   */
+  overlayActive = false;
   reset() {
     this.lastKey = null;
-    this.pendingRetryRefresh = true;
+    this.overlayActive = true;
+  }
+  /**
+   * Arm the request-view overlay without resetting the frozen-block key. Used
+   * when the runtime shape changed while a turn is still running (compaction
+   * checkpoint committed, plan approval, YOLO toggle inside a tool turn): the
+   * next and subsequent model requests must see the current state even though
+   * no new user message has been accepted yet.
+   */
+  armRefresh() {
+    this.overlayActive = true;
   }
-  /** Block to prepend to the next user message, or null when unchanged. */
+  /**
+   * Block to prepend to the next user message, or null when unchanged. Does
+   * NOT end the overlay: the block is only proof once the message carrying it
+   * reaches the request transform.
+   */
   nextBlock() {
-    const block = this.freshBlock();
-    this.pendingRetryRefresh = false;
-    return block;
+    return this.freshBlock();
   }
   /**
-   * Latest block for a request that reuses existing history (overflow retry
-   * after compaction, manual compaction continue): the last user message may
-   * carry a stale or evicted block, so the request-level transform refreshes
-   * it. Consumed once per reset; null when nothing is pending.
+   * Current block for a request that reuses existing history (overflow retry
+   * after compaction, manual-compact continue, or the tool rounds that follow
+   * a mid-turn mode change). Returns the same block for every request until
+   * the latest user message itself carries that block; then the overlay is
+   * retired and null is returned. Derivation is per request from the given
+   * snapshot; no message array is cached.
    */
-  consumeRetryRefreshBlock() {
-    if (!this.pendingRetryRefresh) return null;
-    this.pendingRetryRefresh = false;
+  consumeRetryRefreshBlock(messages) {
+    if (!this.overlayActive) return null;
     const state = this.options.getState();
-    return buildRuntimeModeContext(state);
+    const block = buildRuntimeModeContext(state);
+    if (messages && lastUserMessageCarriesRuntimeBlock(messages, block)) {
+      this.overlayActive = false;
+      return null;
+    }
+    return block;
   }
   freshBlock() {
     const state = this.options.getState();
@@ -80900,14 +81239,17 @@ var RuntimeModeBlockInjector = class {
 };
 
 // src/runtime/cache-retention.ts
-function createCacheRetentionStreamFn(configured) {
+function createCacheRetentionStreamFn(configured, getConfiguredRetention) {
   const base = configured ?? ((model, context, options) => streamSimple(model, context, options));
-  return (model, context, options) => base(model, context, {
-    ...options,
-    // Default only: explicit none/short requests (tests, debug commands) win.
-    // Note: cost estimates still price cache writes at the short-TTL rate.
-    cacheRetention: model.api === "anthropic-messages" && options?.cacheRetention === void 0 ? "long" : options?.cacheRetention
-  });
+  return (model, context, options) => {
+    const requested = options?.cacheRetention;
+    const configuredRetention = model.api === "anthropic-messages" ? getConfiguredRetention?.() : void 0;
+    const cacheRetention = requested ?? configuredRetention;
+    return base(model, context, {
+      ...options,
+      ...cacheRetention === void 0 ? {} : { cacheRetention }
+    });
+  };
 }
 
 // src/runtime/guard-advisory-context.ts
@@ -80931,7 +81273,7 @@ function findLatestUserIndex(messages) {
 function applyRuntimeRequestTransform(messages, options) {
   const advisory = options.takeAdvisory();
   let next = advisory ? injectGuardAdvisory(messages, advisory) : stripGuardAdvisories(messages);
-  const retryBlock = options.takeRetryBlock();
+  const retryBlock = options.takeRetryBlock(next);
   if (retryBlock) next = refreshRuntimeModeBlockOnLastUserMessage(next, retryBlock);
   return next;
 }
@@ -81814,9 +82156,12 @@ async function loadInteractiveExtensions(host) {
   const activeTools = [...host.agent.state.tools, ...extensionTools];
   host.agent.state.tools = wrapToolsWithExtensions(activeTools, host.extensionRunner);
   const availableTools = [
-    ...host.allTools.length > 0 ? host.allTools : host.originalTools.length > 0 ? host.originalTools : host.agent.state.tools,
+    // Prefer the pristine (extension-unwrapped) assembly; re-deriving from
+    // already extension-wrapped `allTools` would nest interceptors.
+    ...host.pristineTools && host.pristineTools.length > 0 ? host.pristineTools : host.allTools.length > 0 ? host.allTools : host.originalTools.length > 0 ? host.originalTools : host.agent.state.tools,
     ...extensionTools
   ];
+  host.pristineTools = [...availableTools];
   host.allTools = wrapToolsWithExtensions(availableTools, host.extensionRunner);
   host.originalTools = [...host.allTools];
   if (host.options.compositionTrace) {
@@ -81899,6 +82244,8 @@ var InteractiveMode = class {
   mentionPickerOptions = [];
   attachmentResolver;
   yoloMode = false;
+  pristineTools = [];
+  // Pristine (extension-unwrapped) tool assembly: the mode-rebuild source.
   /** Calm mode: render only questions and answers (display-only; persisted). */
   calmMode = false;
   originalTools = [];
@@ -82064,7 +82411,7 @@ var InteractiveMode = class {
     const preservedTools = sourceTools.filter(
       (tool) => !CLI_FILESYSTEM_TOOL_NAMES.includes(tool.name) && tool.name !== "bash"
     );
-    const bashTool = createBashTool(this.options.cwd, { isYoloMode: () => this.yoloMode });
+    const bashTool = createBashTool(this.options.cwd, { isYoloMode: () => this.yoloMode, getEnv: () => computerUseOwnerEnv(this.sessionManager.getActiveSessionId() || COMPUTER_USE_STANDALONE_OWNER, String(process.pid)) });
     const fileSystemToolOptions = (allowlistedDirs) => ({
       ...buildCliFileAccessOptions({
         cwd: this.options.cwd,
@@ -82134,7 +82481,7 @@ var InteractiveMode = class {
   }
   applyRuntimeToolMode(sourceTools) {
     if (!this.agent) return;
-    const completeToolSet = sourceTools ?? (this.allTools.length > 0 ? this.allTools : void 0) ?? (this.originalTools.length > 0 ? this.originalTools : void 0) ?? this.agent.state.tools;
+    const completeToolSet = sourceTools ?? ((this.pristineTools?.length ?? 0) > 0 ? this.pristineTools : this.allTools) ?? this.agent.state.tools;
     const rebuilt = this.rebuildSkillAwareTools(completeToolSet);
     const currentTools = this.extensionRunner ? wrapToolsWithExtensions(rebuilt.currentTools, this.extensionRunner) : rebuilt.currentTools;
     const originalTools = this.extensionRunner ? wrapToolsWithExtensions(rebuilt.originalTools, this.extensionRunner) : rebuilt.originalTools;
@@ -82156,6 +82503,7 @@ var InteractiveMode = class {
     });
   }
   noteRuntimeModeChanged(trigger) {
+    this.runtimeModeBlocks?.armRefresh();
     const state = this.getRuntimeModeState();
     this.sessionTrace.noteRuntimeEvent("runtime_mode_changed", {
       trigger,
@@ -82214,7 +82562,7 @@ var InteractiveMode = class {
   async init() {
     const activeSession = this.sessionManager.getActiveSession();
     this.thinkingLevel = resolveInitialThinkingLevel({ session: activeSession, settingsManager: this.options.settingsManager });
-    const { Agent } = await import("./chunks/src-E7NCXMU5.js");
+    const { Agent } = await import("./chunks/src-PRKTGADX.js");
     this.agent = new Agent({
       initialState: {
         systemPrompt: this.buildSystemPrompt(),
@@ -82224,14 +82572,14 @@ var InteractiveMode = class {
       },
       apiKey: this.options.apiKey,
       sessionId: activeSession?.id,
-      streamFn: createCacheRetentionStreamFn(this.options.streamFn),
+      streamFn: createTransportDiagnosticsStreamFn(createCacheRetentionStreamFn(this.options.streamFn, () => this.options.settingsManager.getPromptCacheRetention()), createTraceTransportSink((type, meta) => this.sessionTrace.noteRuntimeEvent(type, meta))),
       transformContext: async (messages) => applyRuntimeRequestTransform(messages, {
         takeAdvisory: () => {
           const advisory = this.pendingRuntimeGuardAdvisory;
           this.pendingRuntimeGuardAdvisory = null;
           return advisory;
         },
-        takeRetryBlock: () => this.runtimeModeBlocks.consumeRetryRefreshBlock()
+        takeRetryBlock: (snapshot) => this.runtimeModeBlocks.consumeRetryRefreshBlock(snapshot)
       })
     });
     this.agentSessionController = createInteractiveAgentSessionController(this.controllerSetupHost);
@@ -82324,6 +82672,7 @@ var InteractiveMode = class {
     }
     this.allTools = [...this.options.allTools];
     this.originalTools = [...this.options.allTools];
+    this.pristineTools = [...this.options.allTools];
     this.agent.state.tools = [...this.options.allTools];
     await this.extensionRunner?.dispose("reload");
     this.extensionWidgets?.clearAll();
@@ -82620,6 +82969,7 @@ var InteractiveMode = class {
 
 // src/app/runtime-composition.ts
 import { existsSync as existsSync18, mkdirSync as mkdirSync13 } from "fs";
+import { randomUUID } from "crypto";
 import * as path31 from "path";
 var CliMissingApiKeyError = class extends Error {
   constructor(message) {
@@ -82749,6 +83099,7 @@ async function composeCliRuntime(params) {
     await sessionManager.save();
   }
   const sessionId = sessionManager.getActiveSession()?.id;
+  const computerUseExecutionId = randomUUID();
   const fileObservationState = createFileObservationState();
   if (sessionId) {
     if (process.env.PIE_DEBUG && !fileLogger.isEnabled()) {
@@ -82777,7 +83128,8 @@ async function composeCliRuntime(params) {
     getWebSearchMode: () => settingsManager.getWebSearchMode(),
     resolveToolModel: (purpose) => modelRegistry.resolveToolModel(purpose, getCurrentModel()),
     resolveToolModelCandidates: (purpose) => modelRegistry.resolveToolModelCandidates(purpose, getCurrentModel()),
-    configDir: paths.configDir
+    configDir: paths.configDir,
+    getBashEnv: () => computerUseOwnerEnv(sessionManager.getActiveSessionId() || COMPUTER_USE_STANDALONE_OWNER, computerUseExecutionId)
   });
   const fileSystemToolOptions = {
     ...buildCliFileAccessOptions({ cwd, configDir: paths.configDir, skills, allowlistedDirs: [paths.configDir] }),
@@ -82905,6 +83257,38 @@ function createPrintModeRetryStatusReporter() {
   };
 }
 
+// src/app/print-mode-compaction.ts
+function createPrintModeCompactionHandler(params) {
+  return {
+    compact: async (context) => {
+      const beforeCompact = await params.runner.emitSessionBeforeCompact({
+        type: "session:before_compact",
+        reason: context.reason === "overflow" ? "overflow" : "threshold",
+        messageCount: context.messages.length
+      });
+      if (beforeCompact.cancel) return { aborted: true };
+      if (beforeCompact.summary) {
+        const retainedTail = context.preparation?.retainedTail.retainedTail ?? (() => {
+          const legacyCutId = findFirstKeptEntrySummaryId(context.sessionManager.getActiveContextSummaries());
+          return legacyCutId ? { kind: "from-entry", entryId: legacyCutId } : void 0;
+        })();
+        if (!retainedTail) return { aborted: true };
+        return {
+          summary: beforeCompact.summary,
+          retainedTail,
+          ...retainedTail.kind === "from-entry" ? { firstKeptEntryId: retainedTail.entryId } : {}
+        };
+      }
+      return createCliAutoCompactionResult({
+        context,
+        model: params.model,
+        apiKey: params.apiKey,
+        systemPrompt: params.systemPrompt
+      });
+    }
+  };
+}
+
 // src/app/print-mode.ts
 function createNoOpExtensionUIContext() {
   return {
@@ -83036,7 +83420,7 @@ async function runPrintMode(params) {
 `);
     console.log("Assistant: ");
   }
-  const { Agent: AgentClass } = await import("./chunks/src-E7NCXMU5.js");
+  const { Agent: AgentClass } = await import("./chunks/src-PRKTGADX.js");
   let agent;
   const printModeExtensions = await loadPrintModeExtensions({
     cwd: runtime.cwd,
@@ -83080,7 +83464,8 @@ async function runPrintMode(params) {
       thinkingLevel: printThinkingLevel
     },
     apiKey: runtime.apiKey,
-    streamFn: params.streamFn
+    // Body-free transport timeline for the main (print-mode) agent.
+    streamFn: createTransportDiagnosticsStreamFn(params.streamFn ?? streamSimple, logTransportLifecycle)
   });
   const sessionTrace = new SessionTraceWriter(runtime.paths.sessionsDir, runtime.sessionManager);
   const reportRetryStatus = createPrintModeRetryStatusReporter();
@@ -83114,28 +83499,16 @@ async function runPrintMode(params) {
       latestCompactionTimestamp: runtime.sessionManager.getLatestCompactionTimestamp(),
       isContextOverflowError: isCliContextOverflowError
     }),
-    compaction: {
-      compact: async (context) => {
-        const beforeCompact = await printModeExtensions.runner.emitSessionBeforeCompact({
-          type: "session:before_compact",
-          reason: context.reason === "overflow" ? "overflow" : "threshold",
-          messageCount: context.messages.length
-        });
-        if (beforeCompact.cancel) return { aborted: true };
-        if (beforeCompact.summary) {
-          const activeEntryId = context.sessionManager.getActiveEntryId();
-          const firstKeptEntryId = activeEntryId ? findFirstKeptEntrySummaryId(context.sessionManager.getActiveContextSummaries()) : void 0;
-          if (!firstKeptEntryId) return { aborted: true };
-          return { summary: beforeCompact.summary, firstKeptEntryId };
-        }
-        return createCliAutoCompactionResult({
-          context,
-          model: runtime.model,
-          apiKey: runtime.apiKey,
-          systemPrompt: agent?.state.systemPrompt ?? baseSystemPrompt
-        });
-      }
-    },
+    getCompactionBudgetTokens: createCliCompactionBudget({
+      contextWindow: runtime.model.contextWindow,
+      reserveTokens: runtime.settingsManager.getAutoCompactReserveTokens()
+    }),
+    compaction: createPrintModeCompactionHandler({
+      runner: printModeExtensions.runner,
+      model: runtime.model,
+      apiKey: runtime.apiKey,
+      systemPrompt: agent?.state.systemPrompt ?? baseSystemPrompt
+    }),
     classifyFailure: (context) => classifyCliFailure(context),
     onEvent: (event) => {
       sessionTrace.noteRuntimeEvent(event.type, event);
@@ -83143,7 +83516,7 @@ async function runPrintMode(params) {
         console.error(event.failure.kind === "context_overflow" ? "Context overflow detected, compacting..." : "Auto-compacting context before retry...");
       } else if (event.type === "auto_compaction_start") {
         console.error(event.reason === "overflow" ? "Context overflow detected, compacting..." : "Auto-compacting context...");
-      } else if (event.type === "auto_compaction_end" && !event.errorMessage) {
+      } else if (event.type === "compaction_committed") {
         runtime.fileObservationState.clear();
       } else {
         reportRetryStatus(event);
@@ -83497,6 +83870,9 @@ async function loadJsonModeExtensions(params) {
     extensionCount: extensions.length
   };
 }
+function createJsonEventsTransportStreamFn(streamFn) {
+  return createTransportDiagnosticsStreamFn(streamFn ?? streamSimple, logTransportLifecycle);
+}
 async function runJsonEventsMode(params) {
   const eventDetail = params.eventDetail === FULL_DETAIL ? FULL_DETAIL : COMPACT_DETAIL;
   const eventStringCap = eventDetail === FULL_DETAIL ? MAX_EVENT_STRING_LENGTH : COMPACT_EVENT_STRING_LENGTH;
@@ -83897,7 +84273,7 @@ async function runJsonEventsMode(params) {
         return;
       }
       const parent = agent;
-      const { Agent: AgentClass } = await import("./chunks/src-E7NCXMU5.js");
+      const { Agent: AgentClass } = await import("./chunks/src-PRKTGADX.js");
       const sideAgent = new AgentClass({
         initialState: {
           systemPrompt: parent.state.systemPrompt,
@@ -83907,7 +84283,11 @@ async function runJsonEventsMode(params) {
           messages: structuredClone(parent.state.messages)
         },
         apiKey: runtime.apiKey,
-        streamFn: params.streamFn
+        // Same body-free transport timeline as the main agent: a sidechat is
+        // an independent Responses request and previously had no record at all
+        // when it stalled. `streamSimple` matches the host default when the
+        // caller injected no stream function.
+        streamFn: createJsonEventsTransportStreamFn(params.streamFn)
       });
       sidechatController.signal.addEventListener("abort", () => sideAgent.abort(), { once: true });
       const timer = setTimeout(() => {
@@ -84065,7 +84445,7 @@ async function runJsonEventsMode(params) {
   }
   function handleControllerEvent(event) {
     trace?.noteRuntimeEvent(event.type, event);
-    if (event.type === "auto_compaction_end" && !event.errorMessage) {
+    if (event.type === "compaction_committed") {
       runtime?.fileObservationState.clear();
     }
     if (event.type === "turn_settled" && activeTurn) {
@@ -84241,7 +84621,7 @@ async function runJsonEventsMode(params) {
     if (runtime.initialModel.unconfigured || controller) {
       return;
     }
-    const { Agent: AgentClass } = await import("./chunks/src-E7NCXMU5.js");
+    const { Agent: AgentClass } = await import("./chunks/src-PRKTGADX.js");
     const jsonModeExtensions = await loadJsonModeExtensions({ runtime, getAgent: () => agent });
     extensionRunner = jsonModeExtensions.runner;
     const tools = [...runtime.tools, ...jsonModeExtensions.extensionTools];
@@ -84273,7 +84653,8 @@ async function runJsonEventsMode(params) {
         thinkingLevel
       },
       apiKey: runtime.apiKey,
-      streamFn: params.streamFn
+      // Body-free transport timeline for the main (JSON-events) agent.
+      streamFn: createJsonEventsTransportStreamFn(params.streamFn)
     });
     controller = new AgentSessionController({
       agent,
@@ -84302,6 +84683,10 @@ async function runJsonEventsMode(params) {
         latestCompactionTimestamp: runtime.sessionManager.getLatestCompactionTimestamp(),
         isContextOverflowError: isCliContextOverflowError
       }),
+      getCompactionBudgetTokens: createCliCompactionBudget({
+        contextWindow: runtime.model.contextWindow,
+        reserveTokens: runtime.settingsManager.getAutoCompactReserveTokens()
+      }),
       compaction: {
         compact: async (context) => createCliAutoCompactionResult({
           context,
@@ -84835,7 +85220,7 @@ async function runWithJsonEventsConsoleIsolation(fn) {
 }
 async function createCliTestStreamFnFromEnv() {
   if (!process.env.PIE_TEST_MOCK_STREAM_SEQUENCE) return void 0;
-  return (await import("./chunks/test-stream-BWBJTFNI.js")).createTestStreamFnFromSequenceEnv();
+  return (await import("./chunks/test-stream-BGAEVPRX.js")).createTestStreamFnFromSequenceEnv();
 }
 function resolveJsonEventsDetailFlag() {
   for (let index = 0; index < process.argv.length; index += 1) {