billion-context-omp 0.1.5 → 0.1.7

package/README.md

@@ -90,20 +90,23 @@ Rich status display for the user:
 ╭─────────────────────────────────────────────╮
 │           ACP Context Analysis              │
 ╰─────────────────────────────────────────────╯
- billion-context-omp@0.1.1
+ billion-context-omp@0.2.0
 
- Context: 6% (57k / 1.0M)
+ Context (session accounting, host footer scale): 9% (93k / 1.0M) — never shrinks; includes compressed originals
 
- Token Breakdown:
-   SysPrompt  ██░░░░░░░░░░░░░░░░░░  10%  5.9k
-   Framework  ██████████████████░░  90%  51k
+ Sent to LLM (after compression, est.): 63k (6% of limit)
+ Session-only (compressed originals, est.): 110k — pruned from every request; the footer/nudge still count them
 
- Nudge: idle — max compressible 0 < threshold 20000; growth 0 < floor 20000
+ Token Breakdown (sent view):
+   Tool       ██████████████████░░  88%  55k
+   SysPrompt  ██░░░░░░░░░░░░░░░░░   9%  5.9k
+   Text       ░░░░░░░░░░░░░░░░░░░░   1%  553
+   Summaries  ░░░░░░░░░░░░░░░░░░░░   2%  1.5k
 
- Blocks: 2 active / 2 total (21k tokens compressed)
-   [b1] T1 5.1k→1.0k: Fold architecture port
-   [b2] T1 15k→799: Replay guard hardening
+ Nudge: idle — growth 0 < floor 20000, ready: T1 50394
 
+ Blocks: 1 active / 1 total (112k tokens compressed)
+   [b1] T1 112k→1.5k: PR141 review + follow-up fixes
  Tag visibility: tags injected to LLM only (deep copy), not persisted in session, not shown in terminal.
 ```
 

package/README.zh-CN.md

@@ -90,20 +90,23 @@ omp 内置的 `/compact` 被拦截，替换为 ACP 模型摘要式 compaction，
 ╭─────────────────────────────────────────────╮
 │           ACP Context Analysis              │
 ╰─────────────────────────────────────────────╯
- billion-context-omp@0.1.1
+ billion-context-omp@0.2.0
 
- Context: 6% (57k / 1.0M)
+ Context (session accounting, host footer scale): 9% (93k / 1.0M) — never shrinks; includes compressed originals
 
- Token Breakdown:
-   SysPrompt  ██░░░░░░░░░░░░░░░░░░  10%  5.9k
-   Framework  ██████████████████░░  90%  51k
+ Sent to LLM (after compression, est.): 63k (6% of limit)
+ Session-only (compressed originals, est.): 110k — pruned from every request; the footer/nudge still count them
 
- Nudge: idle — max compressible 0 < threshold 20000; growth 0 < floor 20000
+ Token Breakdown (sent view):
+   Tool       ██████████████████░░  88%  55k
+   SysPrompt  ██░░░░░░░░░░░░░░░░░   9%  5.9k
+   Text       ░░░░░░░░░░░░░░░░░░░░   1%  553
+   Summaries  ░░░░░░░░░░░░░░░░░░░░   2%  1.5k
 
- Blocks: 2 active / 2 total (21k tokens compressed)
-   [b1] T1 5.1k→1.0k: Fold architecture port
-   [b2] T1 15k→799: Replay guard hardening
+ Nudge: idle — growth 0 < floor 20000, ready: T1 50394
 
+ Blocks: 1 active / 1 total (112k tokens compressed)
+   [b1] T1 112k→1.5k: PR141 review + follow-up fixes
  Tag visibility: tags injected to LLM only (deep copy), not persisted in session, not shown in terminal.
 ```
 

package/dist/compress-tool.d.ts

@@ -3,7 +3,6 @@ import type { AcpRuntime } from "./runtime.js";
 /** Label shown for a block when the model did not pass a topic: first
  *  sentence-ish slice of the summary (≤30 chars). Decorative only — never
  *  blocks compression. */
-export declare function topicFallback(summary: string): string;
 declare const CompressParams: import("@oh-my-pi/omptype").FluentType<{
     content: {
         endId: string;

package/dist/index.js

@@ -126,9 +126,11 @@ function prune(messages, state, options = {}) {
   const indexById = /* @__PURE__ */ new Map();
   messages.forEach((message, index) => indexById.set(message.id, index));
   const anchors = inject ? collectSummaryAnchors(state, indexById) : [];
-  return stripOrphanedToolResults(
-    stripOrphanedToolCalls(
-      rebuildMessages(messages, covered, firstUserIndex, anchors)
+  return stripOrphanedReasoning(
+    stripOrphanedToolResults(
+      stripOrphanedToolCalls(
+        rebuildMessages(messages, covered, firstUserIndex, anchors)
+      )
     )
   );
 }
@@ -205,6 +207,24 @@ function stripOrphanedToolCalls(messages) {
     (m) => m.contentType !== "tool-call" || !m.toolCallId || m.toolName === "compress" || knownResultIds.has(m.toolCallId)
   );
 }
+function stripOrphanedReasoning(messages) {
+  const drop = /* @__PURE__ */ new Set();
+  for (let i = 0; i < messages.length; i++) {
+    if (drop.has(i)) continue;
+    if (messages[i].contentType !== "reasoning") continue;
+    let j = i;
+    while (j + 1 < messages.length && messages[j + 1].contentType === "reasoning") {
+      j++;
+    }
+    const companion = messages[j + 1];
+    const hasCompanion = companion !== void 0 && companion.role === "assistant" && (companion.contentType === "text" || companion.contentType === "tool-call");
+    if (!hasCompanion) {
+      for (let k = i; k <= j; k++) drop.add(k);
+    }
+  }
+  if (drop.size === 0) return messages;
+  return messages.filter((_, i) => !drop.has(i));
+}
 function syncBlocks(messages, state) {
   const presentIds = new Set(messages.map((message) => message.id));
   const deactivated = [];
@@ -814,6 +834,38 @@ function adjustBoundariesForToolPairs(startIndex, endIndex, messages, maxScan =
   }
   return { startIndex: newStartIndex, endIndex: newEndIndex };
 }
+function adjustBoundariesForReasoningPairs(startIndex, endIndex, messages) {
+  if (startIndex > endIndex) {
+    return { startIndex, endIndex };
+  }
+  let newStartIndex = startIndex;
+  let newEndIndex = endIndex;
+  for (let i = startIndex; i <= endIndex && i < messages.length; i++) {
+    const msg = messages[i];
+    if (!msg) continue;
+    if (msg.contentType === "reasoning") {
+      let j = i;
+      while (j + 1 < messages.length && messages[j + 1].contentType === "reasoning") {
+        j++;
+      }
+      const companion = messages[j + 1];
+      if (companion !== void 0 && companion.role === "assistant" && (companion.contentType === "text" || companion.contentType === "tool-call") && j + 1 > newEndIndex) {
+        newEndIndex = j + 1;
+      }
+    }
+    if (msg.role === "assistant" && (msg.contentType === "text" || msg.contentType === "tool-call")) {
+      let k = i - 1;
+      while (k >= 0 && messages[k].contentType === "reasoning") {
+        k--;
+      }
+      const runStart = k + 1;
+      if (runStart < i && runStart >= 0 && messages[runStart].contentType === "reasoning" && runStart < newStartIndex) {
+        newStartIndex = runStart;
+      }
+    }
+  }
+  return { startIndex: newStartIndex, endIndex: newEndIndex };
+}
 function refNum(ref) {
   const n = parseInt(ref.slice(1), 10);
   return Number.isNaN(n) ? -1 : n;
@@ -1309,7 +1361,7 @@ function applySingleRange(input) {
     messages: input.messages,
     state: input.state
   });
-  const rangeMessageIds = applyToolPairAdjustment(
+  const rangeMessageIds = applyPairBoundaryAdjustments(
     resolved,
     input.messages
   );
@@ -1423,20 +1475,33 @@ function applySingleRange(input) {
   }
   return { tokens: compressedTokens, warnings };
 }
-function applyToolPairAdjustment(resolved, messages) {
+function applyPairBoundaryAdjustments(resolved, messages) {
   if (resolved.boundaryKind === "block") {
     return resolved.messageIds;
   }
-  const adjusted = adjustBoundariesForToolPairs(
-    resolved.startIndex,
-    resolved.endIndex,
-    messages
-  );
-  if (adjusted.startIndex === resolved.startIndex && adjusted.endIndex === resolved.endIndex) {
+  let startIndex = resolved.startIndex;
+  let endIndex = resolved.endIndex;
+  for (let pass = 0; pass < 2; pass++) {
+    const reasoningAdjusted = adjustBoundariesForReasoningPairs(
+      startIndex,
+      endIndex,
+      messages
+    );
+    const toolAdjusted = adjustBoundariesForToolPairs(
+      reasoningAdjusted.startIndex,
+      reasoningAdjusted.endIndex,
+      messages
+    );
+    const changed = toolAdjusted.startIndex !== startIndex || toolAdjusted.endIndex !== endIndex;
+    startIndex = toolAdjusted.startIndex;
+    endIndex = toolAdjusted.endIndex;
+    if (!changed) break;
+  }
+  if (startIndex === resolved.startIndex && endIndex === resolved.endIndex) {
     return resolved.messageIds;
   }
   const ids = [];
-  for (let i = adjusted.startIndex; i <= adjusted.endIndex; i++) {
+  for (let i = startIndex; i <= endIndex; i++) {
     const msg = messages[i];
     if (msg) ids.push(msg.id);
   }
@@ -3003,10 +3068,6 @@ function boundaryRaw(ref, byRef, blocks, pick) {
   const pos = (id) => rawPos(byRef[id] ?? id);
   return pick === "min" ? ids.reduce((a, b) => pos(a) <= pos(b) ? a : b) : ids.reduce((a, b) => pos(a) >= pos(b) ? a : b);
 }
-var VIABLE_RANGE_MIN_TOKENS = 200;
-function viableRanges(ranges) {
-  return ranges.filter((r) => r.tokens >= VIABLE_RANGE_MIN_TOKENS);
-}
 function rangeFingerprints(ranges, coreMessages, byRef, blocks) {
   return ranges.map((r) => {
     const start = boundaryRaw(r.startRef, byRef, blocks, "min");
@@ -3240,11 +3301,6 @@ var RangeSpec = type({
   summary: type("string").describe("Complete technical summary replacing all content in range. Keep only essential details (conclusions, file paths, decisions, exact values, etc.)."),
   "topic?": type("string").describe("Short label (3-5 words) for THIS range, e.g. 'Auth System Exploration'. Recommended for every range; omit to use top-level topic.")
 });
-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`;
-}
 var CompressParams = type({
   "topic?": type("string").describe("Fallback topic for entries without their own. Omit when each content entry specifies its own topic."),
   content: RangeSpec.array().describe("One or more ranges to compress, each with start/end boundaries and a summary. When compressing multiple unrelated ranges in one call, give each its own topic."),
@@ -3623,6 +3679,130 @@ function truncate(s, n) {
 
 // src/status-tool.ts
 import { type as type4 } from "@oh-my-pi/omptype";
+
+// src/compat.ts
+function normalizeSystemPrompt(input) {
+  if (input === void 0) return "";
+  if (Array.isArray(input)) return input.join("\n");
+  return input;
+}
+function formatSystemPromptForEvent(base, append) {
+  const normalized = normalizeSystemPrompt(base);
+  return [`${normalized}
+
+${append}`];
+}
+function getSystemPromptText(ctx) {
+  const result = ctx.getSystemPrompt?.();
+  return normalizeSystemPrompt(result);
+}
+
+// node_modules/billion-context-kit/dist/index.js
+var VIABLE_RANGE_MIN_TOKENS = 200;
+function viableRanges(ranges) {
+  return ranges.filter((r) => r.tokens >= VIABLE_RANGE_MIN_TOKENS);
+}
+function topicFallback(summary) {
+  const first = summary.split(/[.\n]/)[0] ?? "";
+  const t = first.trim().replace(/^["'`]+/, "").trim();
+  return t.length <= 30 ? t : `${t.slice(0, 30).trimEnd()}\u2026`;
+}
+function formatCompactTokens(count) {
+  if (count < 1e3) return count.toString();
+  if (count < 1e4) return `${(count / 1e3).toFixed(1)}k`;
+  if (count < 1e6) return `${Math.round(count / 1e3)}k`;
+  if (count < 1e7) return `${(count / 1e6).toFixed(1)}M`;
+  return `${Math.round(count / 1e6)}M`;
+}
+function bar(value, total, width = 20) {
+  if (total === 0) return "";
+  const filled = Math.max(0, Math.min(width, Math.round(value / total * width)));
+  return "\u2588".repeat(filled) + "\u2591".repeat(width - filled);
+}
+function buildStatusPanel(input) {
+  const { tokenCount, state, nudge, modelContextLimit } = input;
+  const fmt2 = input.fmtTokens ?? formatCompactTokens;
+  const bd = nudge?.contextBreakdown;
+  const limit = modelContextLimit;
+  const classified = bd ? bd.system + bd.tool + bd.summaries + bd.code + bd.text : 0;
+  const systemPromptTokens = input.systemPromptTokens;
+  const sentTotal = classified + systemPromptTokens;
+  const sessionOnly = input.unprunedTokens !== void 0 ? Math.max(0, input.unprunedTokens - sentTotal) : 0;
+  const displayTotal = tokenCount;
+  const displayPct = limit > 0 ? Math.round(displayTotal / limit * 100) : 0;
+  const sentPct = limit > 0 ? Math.round(sentTotal / limit * 100) : 0;
+  const activeBlocksList = state.blocks.filter((b) => b.active);
+  const totalBlocksList = state.blocks;
+  const lines = [];
+  lines.push("\u256D\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u256E");
+  lines.push("\u2502           ACP Context Analysis              \u2502");
+  lines.push("\u2570\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u256F");
+  if (input.version) lines.push(input.version);
+  lines.push("");
+  lines.push(`Context (session accounting, host footer scale): ${displayPct}% (${fmt2(displayTotal)} / ${fmt2(limit)}) \u2014 never shrinks; includes compressed originals`);
+  if (nudge && bd) {
+    const growth = bd.growth;
+    if (growth > 0 && displayTotal > 0) {
+      lines.push(`Growth: +${fmt2(growth)} since last nudge`);
+    }
+    lines.push("");
+    lines.push(`Sent to LLM (after compression, est.): ${fmt2(sentTotal)}${limit > 0 ? ` (${sentPct}% of limit)` : ""}`);
+    if (input.unprunedTokens !== void 0 && sessionOnly > 0) {
+      lines.push(`Session-only (compressed originals, est.): ${fmt2(sessionOnly)} \u2014 pruned from every request; the footer/nudge still count them`);
+    }
+    lines.push("");
+    lines.push("Token Breakdown (sent view):");
+    const categories = [
+      { label: "Tool", value: bd.tool },
+      { label: "SysPrompt", value: systemPromptTokens },
+      { label: "Text", value: bd.text },
+      { label: "Code", value: bd.code },
+      { label: "Summaries", value: bd.summaries }
+    ];
+    for (const cat of categories) {
+      if (cat.value <= 0) continue;
+      const pct2 = sentTotal > 0 ? Math.round(cat.value / sentTotal * 100) : 0;
+      const b = bar(cat.value, sentTotal);
+      lines.push(`  ${cat.label.padEnd(10)} ${b} ${String(pct2).padStart(3)}%  ${fmt2(cat.value)}`);
+    }
+  }
+  lines.push("");
+  if (nudge) {
+    if (nudge.shouldInject) {
+      const tierInfo = nudge.tier ? ` [T${nudge.tier} distillation]` : "";
+      lines.push(`Nudge: ACTIVE${tierInfo} \u2014 ${nudge.reason}`);
+    } else {
+      lines.push(`Nudge: idle \u2014 ${nudge.reason}`);
+    }
+  }
+  const ranges = viableRanges(nudge?.compressibleRanges ?? []);
+  const protectedRanges = nudge?.protectedRanges ?? [];
+  if (ranges.length > 0 || protectedRanges.length > 0) {
+    lines.push("");
+    lines.push(formatRanges(ranges, protectedRanges));
+  }
+  if (activeBlocksList.length > 0) {
+    lines.push("");
+    lines.push(`Blocks: ${activeBlocksList.length} active / ${totalBlocksList.length} total (${fmt2(state.stats.tokensCompressed)} tokens compressed)`);
+    for (const b of activeBlocksList) {
+      const topic = b.topic ? `: ${b.topic}` : `: ${topicFallback(b.summary || "")}`;
+      const summaryTok = defaultCountTokens(b.summary || "");
+      const origTok = b.compressedTokens > 0 ? b.compressedTokens : summaryTok;
+      lines.push(`  [${b.blockId}] T${b.tier} ${fmt2(origTok)}\u2192${fmt2(summaryTok)}${topic}`);
+    }
+  } else if (totalBlocksList.length > 0) {
+    lines.push("");
+    lines.push(`Blocks: 0 active / ${totalBlocksList.length} total (${fmt2(state.stats.tokensCompressed)} tokens compressed)`);
+  } else {
+    lines.push("");
+    lines.push("Blocks: none (nothing compressed yet)");
+  }
+  lines.push("");
+  lines.push("Tag visibility: tags injected to LLM only (deep copy), not persisted in session, not shown in terminal.");
+  return lines.join("\n");
+}
+
+// src/status-tool.ts
 var StatusParams = type4({
   "scope?": type4('"compressed" | "uncompressed"').describe('"compressed" = drill into blocks; "uncompressed" = show visible messages/ranges. Default: overview.'),
   "view?": type4('"ranges" | "messages"').describe('For uncompressed scope: "ranges" (default) or "messages" (per-message listing).'),
@@ -3651,13 +3831,16 @@ function makeStatusTool(runtime) {
 async function handleStatus(args, runtime, ctx) {
   const { state, coreMessages } = await runtime.stateFor(ctx);
   const config = runtime.configFor(ctx);
-  const tokenCount = estimateTokens(coreMessages, collectCoveredMessageIds(state));
-  const realUsage = ctx.getContextUsage?.();
+  const tokenCount = estimateTokens(coreMessages, collectCoveredMessageIds(state)) + estimateTextTokens2(getSystemPromptText(ctx) ?? "");
   const turn = runtime.core.processTurn({
     messages: coreMessages,
     state,
     config,
-    tokenCount: realUsage?.tokens && realUsage.tokens > 0 ? realUsage.tokens : tokenCount
+    // Sent-view scale — see src/index.ts context handler. The session-tree
+    // number (ctx.getContextUsage) must never arbitrate emergencies for the
+    // sent view: a tree that outgrew the model window reads as a permanent
+    // 200%+ emergency while the real sent view is a few percent.
+    tokenCount
   });
   const processed = turn.messages;
   const base = buildStatusReport(turn.state, processed, defaultCountTokens, {
@@ -3686,32 +3869,6 @@ async function handleStatus(args, runtime, ctx) {
 ${extra.join("\n")}` : base;
 }
 
-// src/compat.ts
-function normalizeSystemPrompt(input) {
-  if (input === void 0) return "";
-  if (Array.isArray(input)) return input.join("\n");
-  return input;
-}
-function formatSystemPromptForEvent(base, append) {
-  const normalized = normalizeSystemPrompt(base);
-  return [`${normalized}
-
-${append}`];
-}
-function getSystemPromptText(ctx) {
-  const result = ctx.getSystemPrompt?.();
-  return normalizeSystemPrompt(result);
-}
-
-// src/footer-status.ts
-function formatCompactTokens(count) {
-  if (count < 1e3) return count.toString();
-  if (count < 1e4) return `${(count / 1e3).toFixed(1)}k`;
-  if (count < 1e6) return `${Math.round(count / 1e3)}k`;
-  if (count < 1e7) return `${(count / 1e6).toFixed(1)}M`;
-  return `${Math.round(count / 1e6)}M`;
-}
-
 // src/commands.ts
 function safeHandler(handler) {
   return async (args, ctx) => {
@@ -3789,100 +3946,26 @@ ${text}`);
     }
   ];
 }
-function fmtTokens(n) {
-  return formatCompactTokens(n);
-}
-function bar(value, total, width = 20) {
-  if (total === 0) return "";
-  const filled = Math.max(0, Math.min(width, Math.round(value / total * width)));
-  return "\u2588".repeat(filled) + "\u2591".repeat(width - filled);
-}
 async function statusReport(runtime, ctx) {
   const { state, coreMessages } = await runtime.stateFor(ctx);
   const config = runtime.configFor(ctx);
   const realUsage = ctx.getContextUsage?.();
-  const tokenCount = realUsage?.tokens && realUsage.tokens > 0 ? realUsage.tokens : defaultCountTokens(coreMessages.map((m) => m.text ?? "").join("\n"));
-  const turn = runtime.core.processTurn({ messages: coreMessages, state, config, tokenCount });
-  const nudge = turn.nudge;
-  const bd = nudge?.contextBreakdown;
-  const limit = config.modelContextLimit;
-  const classified = bd ? bd.system + bd.tool + bd.summaries + bd.code + bd.text : 0;
+  const sessionTokens = realUsage?.tokens && realUsage.tokens > 0 ? realUsage.tokens : defaultCountTokens(coreMessages.map((m) => m.text ?? "").join("\n"));
   const systemPromptText = getSystemPromptText(ctx);
   const systemPromptTokens = systemPromptText ? defaultCountTokens(systemPromptText) : 0;
-  const sentTotal = classified + systemPromptTokens;
-  const sessionOnly = Math.max(0, tokenCount - sentTotal);
-  const displayTotal = tokenCount;
-  const displayPct = limit > 0 ? Math.round(displayTotal / limit * 100) : 0;
-  const activeBlocksList = state.blocks.filter((b) => b.active);
-  const totalBlocksList = state.blocks;
-  const lines = [];
-  const versionStr = "0.1.5" ? `billion-context-omp@${"0.1.5"}` : "";
-  lines.push("\u256D\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u256E");
-  lines.push("\u2502           ACP Context Analysis              \u2502");
-  lines.push("\u2570\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u256F");
-  if (versionStr) lines.push(versionStr);
-  lines.push("");
-  lines.push(`Context (session accounting): ${displayPct}% (${fmtTokens(displayTotal)} / ${fmtTokens(limit)})`);
-  if (nudge && bd) {
-    const growth = bd.growth;
-    if (growth > 0 && displayTotal > 0) {
-      lines.push(`Growth: +${fmtTokens(growth)} since last nudge`);
-    }
-    lines.push("");
-    lines.push(`Sent to LLM (after compression): ${fmtTokens(sentTotal)}`);
-    if (sessionOnly > 0) {
-      lines.push(`Session-only (compressed originals + host overhead): ${fmtTokens(sessionOnly)} \u2014 pruned from every request; the footer counts it`);
-    }
-    lines.push("");
-    lines.push("Token Breakdown (sent view):");
-    const categories = [
-      { label: "Tool", value: bd.tool },
-      { label: "SysPrompt", value: systemPromptTokens },
-      { label: "Text", value: bd.text },
-      { label: "Code", value: bd.code },
-      { label: "Summaries", value: bd.summaries }
-    ];
-    for (const cat of categories) {
-      if (cat.value <= 0) continue;
-      const pct2 = sentTotal > 0 ? Math.round(cat.value / sentTotal * 100) : 0;
-      const b = bar(cat.value, sentTotal);
-      lines.push(`  ${cat.label.padEnd(10)} ${b} ${String(pct2).padStart(3)}%  ${fmtTokens(cat.value)}`);
-    }
-  }
-  lines.push("");
-  if (nudge) {
-    if (nudge.shouldInject) {
-      const tierInfo = nudge.tier ? ` [T${nudge.tier} distillation]` : "";
-      lines.push(`Nudge: ACTIVE${tierInfo} \u2014 ${nudge.reason}`);
-    } else {
-      lines.push(`Nudge: idle \u2014 ${nudge.reason}`);
-    }
-  }
-  const ranges = viableRanges(nudge?.compressibleRanges ?? []);
-  const protectedRanges = nudge?.protectedRanges ?? [];
-  if (ranges.length > 0 || protectedRanges.length > 0) {
-    lines.push("");
-    lines.push(formatRanges(ranges, protectedRanges));
-  }
-  if (activeBlocksList.length > 0) {
-    lines.push("");
-    lines.push(`Blocks: ${activeBlocksList.length} active / ${totalBlocksList.length} total (${fmtTokens(state.stats.tokensCompressed)} tokens compressed)`);
-    for (const b of activeBlocksList) {
-      const topic = b.topic ? `: ${b.topic}` : `: ${topicFallback(b.summary || "")}`;
-      const summaryTok = defaultCountTokens(b.summary || "");
-      const origTok = b.compressedTokens > 0 ? b.compressedTokens : summaryTok;
-      lines.push(`  [${b.blockId}] T${b.tier} ${fmtTokens(origTok)}\u2192${fmtTokens(summaryTok)}${topic}`);
-    }
-  } else if (totalBlocksList.length > 0) {
-    lines.push("");
-    lines.push(`Blocks: 0 active / ${totalBlocksList.length} total (${fmtTokens(state.stats.tokensCompressed)} tokens compressed)`);
-  } else {
-    lines.push("");
-    lines.push("Blocks: none (nothing compressed yet)");
-  }
-  lines.push("");
-  lines.push("Tag visibility: tags injected to LLM only (deep copy), not persisted in session, not shown in terminal.");
-  return lines.join("\n");
+  const coveredIds = collectCoveredMessageIds(state);
+  const sentTokens = estimateTokens(coreMessages, coveredIds) + systemPromptTokens;
+  const turn = runtime.core.processTurn({ messages: coreMessages, state, config, tokenCount: sentTokens });
+  const versionStr = "0.1.7" ? `billion-context-omp@${"0.1.7"}` : void 0;
+  return buildStatusPanel({
+    version: versionStr,
+    tokenCount: sessionTokens,
+    systemPromptTokens,
+    state: turn.state,
+    nudge: turn.nudge,
+    modelContextLimit: config.modelContextLimit,
+    unprunedTokens: coreMessages.reduce((sum, m) => sum + defaultCountTokens(m.text ?? ""), 0)
+  });
 }
 
 // src/auto-compress.ts
@@ -4313,7 +4396,7 @@ async function checkForUpdate(autoUpdate, notify) {
     const data = await res.json();
     const latest = data.version;
     if (!latest) return;
-    const current = runtimeVersion ?? "0.1.5";
+    const current = runtimeVersion ?? "0.1.7";
     const hasUpdate = isNewer(latest, current);
     debug.event("update-check", {
       current,
@@ -4601,7 +4684,7 @@ function wireCompactionDisable(pi, runtime) {
 function wireSessionLifecycle(pi, runtime) {
   pi.on("session_start", async (_event, ctx) => {
     const sid = ctx.sessionManager.getSessionId();
-    logInfo("session", { event: "start", sid, cwd: ctx.cwd, debug: runtime.adapter.debug ?? null, version: true ? "0.1.5" : null });
+    logInfo("session", { event: "start", sid, cwd: ctx.cwd, debug: runtime.adapter.debug ?? null, version: true ? "0.1.7" : null });
     try {
       const user = await loadUserConfig(ctx.cwd);
       runtime.setAdapter(applyUserConfig(runtime.adapter, user));
@@ -4642,18 +4725,17 @@ function wireContextTransform(pi, runtime) {
       const { state, coreMessages, originalById, streamLen } = runtime.foldStream(ctx, input);
       const config = runtime.configFor(ctx);
       const coveredIds = collectCoveredMessageIds(state);
-      const realUsage = ctx.getContextUsage?.();
-      const estimated = estimateTokens(coreMessages, coveredIds);
-      const tokenCount = realUsage?.tokens && realUsage.tokens > 0 ? realUsage.tokens : estimated;
+      const systemPromptTokens = estimateTextTokens2(getSystemPromptText(ctx) ?? "");
+      const sentTokens = estimateTokens(coreMessages, coveredIds) + systemPromptTokens;
+      const sessionTokens = ctx.getContextUsage?.()?.tokens ?? null;
+      const tokenCount = sentTokens;
       debug.event("context-in", {
         sid,
         eventMsgs: event.messages?.length ?? 0,
         streamLen,
         coreMsgs: coreMessages.length,
         tokenCount,
-        estimatedTokens: estimated,
-        realTokens: realUsage?.tokens ?? null,
-        realPercent: realUsage?.percent ?? null,
+        sessionTokens,
         limit: config.modelContextLimit,
         blocksBefore: state.blocks.length,
         activeBefore: state.blocks.filter((b) => b.active).length
@@ -4665,7 +4747,8 @@ function wireContextTransform(pi, runtime) {
         inMsgs: coreMessages.length,
         outMsgs: turn.messages.length,
         tokens: tokenCount,
-        pct: realUsage?.percent ?? (config.modelContextLimit > 0 ? Math.round(tokenCount / config.modelContextLimit * 100) : null),
+        sessionTokens,
+        pct: config.modelContextLimit > 0 ? Math.round(tokenCount / config.modelContextLimit * 100) : null,
         limit: config.modelContextLimit,
         nudge: turn.nudge?.shouldInject ? turn.nudge.breakdown?.emergencyOverride === 1 ? "emergency" : "active" : "idle",
         nudgeReason: turn.nudge?.reason ?? null,