billion-context-omp 0.1.6 → 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/index.js

@@ -3680,6 +3680,23 @@ 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) {
@@ -3710,9 +3727,10 @@ function buildStatusPanel(input) {
   const classified = bd ? bd.system + bd.tool + bd.summaries + bd.code + bd.text : 0;
   const systemPromptTokens = input.systemPromptTokens;
   const sentTotal = classified + systemPromptTokens;
-  const sessionOnly = Math.max(0, tokenCount - sentTotal);
+  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 = [];
@@ -3721,16 +3739,16 @@ function buildStatusPanel(input) {
   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): ${displayPct}% (${fmt2(displayTotal)} / ${fmt2(limit)})`);
+  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): ${fmt2(sentTotal)}`);
-    if (sessionOnly > 0) {
-      lines.push(`Session-only (compressed originals + host overhead): ${fmt2(sessionOnly)} \u2014 pruned from every request; the footer counts it`);
+    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):");
@@ -3813,13 +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, {
@@ -3848,23 +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/commands.ts
 function safeHandler(handler) {
   return async (args, ctx) => {
@@ -3946,17 +3950,21 @@ 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 sessionTokens = realUsage?.tokens && realUsage.tokens > 0 ? realUsage.tokens : defaultCountTokens(coreMessages.map((m) => m.text ?? "").join("\n"));
   const systemPromptText = getSystemPromptText(ctx);
-  const versionStr = "0.1.6" ? `billion-context-omp@${"0.1.6"}` : void 0;
+  const systemPromptTokens = systemPromptText ? defaultCountTokens(systemPromptText) : 0;
+  const coveredIds = collectCoveredMessageIds(state);
+  const sentTokens = estimateTokens(coreMessages, coveredIds) + systemPromptTokens;
+  const turn = runtime.core.processTurn({ messages: coreMessages, state, config, tokenCount: sentTokens });
+  const versionStr = "0.1.7" ? `billion-context-omp@${"0.1.7"}` : void 0;
   return buildStatusPanel({
     version: versionStr,
-    tokenCount,
-    systemPromptTokens: systemPromptText ? defaultCountTokens(systemPromptText) : 0,
+    tokenCount: sessionTokens,
+    systemPromptTokens,
     state: turn.state,
     nudge: turn.nudge,
-    modelContextLimit: config.modelContextLimit
+    modelContextLimit: config.modelContextLimit,
+    unprunedTokens: coreMessages.reduce((sum, m) => sum + defaultCountTokens(m.text ?? ""), 0)
   });
 }
 
@@ -4388,7 +4396,7 @@ async function checkForUpdate(autoUpdate, notify) {
     const data = await res.json();
     const latest = data.version;
     if (!latest) return;
-    const current = runtimeVersion ?? "0.1.6";
+    const current = runtimeVersion ?? "0.1.7";
     const hasUpdate = isNewer(latest, current);
     debug.event("update-check", {
       current,
@@ -4676,7 +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 ? "0.1.6" : 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));
@@ -4717,18 +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
@@ -4740,7 +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,