billion-context-pi 0.1.52 → 0.1.53-pr.262.50

package/dist/commands.d.ts

@@ -1,7 +1,7 @@
-import type { RegisteredCommand } from "@earendil-works/pi-coding-agent";
+import type { ExtensionAPI, RegisteredCommand } from "@earendil-works/pi-coding-agent";
 import type { AcpRuntime } from "./runtime.js";
 type CommandOptions = Omit<RegisteredCommand, "name" | "sourceInfo">;
-export declare function makeCommands(runtime: AcpRuntime): Array<{
+export declare function makeCommands(runtime: AcpRuntime, pi?: ExtensionAPI): Array<{
     name: string;
     options: CommandOptions;
 }>;

package/dist/floor-stale.d.ts

@@ -0,0 +1,25 @@
+type UsageLike = {
+    totalTokens?: number;
+    input?: number;
+    output?: number;
+    cacheRead?: number;
+    cacheWrite?: number;
+} | null | undefined;
+type AnchorEntry = {
+    type: string;
+    message?: {
+        role?: string;
+        stopReason?: string;
+        usage?: UsageLike;
+        toolName?: string;
+        toolCallId?: string;
+        isError?: boolean;
+        content?: unknown;
+    };
+};
+/** True when the last valid assistant usage anchor comes strictly BEFORE the
+ *  last successful compress toolResult — the host's provider-usage number
+ *  still reflects the pre-compression request. Failed/no-op compresses don't
+ *  count (nothing was reclaimed). */
+export declare function usageAnchorPredatesCompression(entries: AnchorEntry[]): boolean;
+export {};

package/dist/index.js

@@ -3569,89 +3569,6 @@ function parsePercent(v) {
   return Number(s);
 }
 
-// src/density.ts
-var DENSITY_MIN = 0.5;
-var DENSITY_MAX = 2.5;
-var MIN_DELTA_EST = 50;
-var CONFIRM_RATIO = 0.2;
-var INITIAL_DENSITY = 1;
-var DensityEstimator = class {
-  models = /* @__PURE__ */ new Map();
-  /** 重置指定模型（模型切换/会话开始时调用）。 */
-  resetModel(modelId) {
-    this.models.delete(modelId);
-  }
-  /** 返回当前密度系数（未知模型返回初始 1）。 */
-  densityFor(modelId) {
-    return this.models.get(modelId)?.density ?? INITIAL_DENSITY;
-  }
-  /**
-   * 每轮 context 事件调用。realTotal 为 provider 真实 usage（可空），
-   * estTotal 为本地估算总 token。postCompression 为压缩刚发生标志。
-   */
-  update(modelId, realTotal, estTotal, postCompression = false) {
-    if (realTotal === null) return;
-    let est = this.models.get(modelId);
-    if (!est) {
-      est = {
-        density: INITIAL_DENSITY,
-        anchorReal: null,
-        anchorEst: null,
-        pendingDensity: null,
-        confirmCount: 0,
-        postCompressionSkip: false
-      };
-      this.models.set(modelId, est);
-    }
-    if (postCompression) {
-      est.postCompressionSkip = true;
-      return;
-    }
-    if (est.postCompressionSkip) {
-      est.postCompressionSkip = false;
-      est.anchorReal = realTotal;
-      est.anchorEst = estTotal;
-      est.pendingDensity = null;
-      est.confirmCount = 0;
-      return;
-    }
-    if (est.anchorReal === null || est.anchorEst === null) {
-      est.anchorReal = realTotal;
-      est.anchorEst = estTotal;
-      return;
-    }
-    const dReal = realTotal - est.anchorReal;
-    const dEst = estTotal - est.anchorEst;
-    if (dEst < MIN_DELTA_EST) return;
-    est.anchorReal = realTotal;
-    est.anchorEst = estTotal;
-    const instant = clamp(dReal / dEst, DENSITY_MIN, DENSITY_MAX);
-    if (est.pendingDensity === null) {
-      est.pendingDensity = instant;
-      est.confirmCount = 1;
-    } else if (Math.abs(instant - est.pendingDensity) / est.pendingDensity <= CONFIRM_RATIO) {
-      est.confirmCount += 1;
-    } else {
-      est.pendingDensity = instant;
-      est.confirmCount = 1;
-    }
-    if (est.confirmCount >= 2) {
-      est.density = est.pendingDensity;
-      est.confirmCount = 0;
-      est.pendingDensity = null;
-    }
-  }
-  /** 注入器：估算文本 token = defaultCountTokens × density。 */
-  estimateWithDensity(modelId, text) {
-    const d = this.densityFor(modelId);
-    if (d === 1) return defaultCountTokens(text);
-    return Math.round(defaultCountTokens(text) * d);
-  }
-};
-function clamp(v, lo, hi) {
-  return v < lo ? lo : v > hi ? hi : v;
-}
-
 // src/tag-tokens.ts
 function formatTokens3(tokens) {
   if (tokens < 1e3) return String(tokens);
@@ -3669,11 +3586,12 @@ function rewriteTagTokens(tag, body) {
 var REF_TAG_SOURCE = "(?:<acp\\s[^>]*>m\\d{5}</acp>|\\[m\\d{1,5}\\])";
 var REF_TAG = new RegExp(`^${REF_TAG_SOURCE}\\s?\\n?`);
 var TRAILING_REF_TAG = new RegExp(`\\n*${REF_TAG_SOURCE}\\s*$`);
+var ACP_STATUS_CUSTOM_TYPE = "acp-status";
 function entriesToCoreMessages(entries) {
   const out = [];
   for (const entry of entries) {
     if (entry.type !== "message") {
-      if (entry.type === "custom_message") {
+      if (entry.type === "custom_message" && entry.customType !== ACP_STATUS_CUSTOM_TYPE) {
         const text = extractText(entry.content);
         if (text.length > 0) {
           out.push({ id: entry.id, role: "user", contentType: "text", text });
@@ -4622,14 +4540,12 @@ function pruneOrphanRefs(state, messages) {
 }
 var MAX_COMPRESS_ATTEMPTS = 3;
 function createRuntime(adapter) {
-  const density = new DensityEstimator();
-  let countModelId = "default";
   const core = createCore({
-    // 密度校准版 countTokens（Phase 2）：默认回落 defaultCountTokens（density=1）
-    countTokens: (text) => density.estimateWithDensity(countModelId, text)
+    // CJK-aware base counter (kernel default); the meter's real-scale floor
+    // (src/index.ts) covers provider-anchored calibration.
+    countTokens: defaultCountTokens
   });
   const store = new SessionStateStore();
-  const lastActiveBlockIds = /* @__PURE__ */ new Map();
   const locks = /* @__PURE__ */ new Map();
   const factoryAdapter = adapter;
   let adapterRef = adapter;
@@ -4754,19 +4670,7 @@ function createRuntime(adapter) {
     const sm = ctx.sessionManager;
     await store.save(state, sm.getSessionFile() ?? void 0, sm.getSessionId());
   }
-  function noteActiveBlocks(sid, activeBlockIds) {
-    const current = new Set(activeBlockIds);
-    const prev = lastActiveBlockIds.get(sid);
-    const isNew = prev !== void 0 && activeBlockIds.some((id) => !prev.has(id));
-    lastActiveBlockIds.set(sid, current);
-    return isNew;
-  }
-  function clearSessionTracking(sid) {
-    lastActiveBlockIds.delete(sid);
-  }
-  return { core, store, density, setCountModel: (m) => {
-    countModelId = m;
-  }, noteActiveBlocks, clearSessionTracking, get adapter() {
+  return { core, store, get adapter() {
     return adapterRef;
   }, setAdapter: (a) => {
     adapterRef = a;
@@ -9210,9 +9114,6 @@ function estimateTokens(messages, coveredIds, imageTokensById) {
   }
   return tokens;
 }
-function calibrateTokens(estimate, density) {
-  return density === 1 ? estimate : Math.round(estimate * density);
-}
 function lastUserMessageId(entries) {
   for (let i = entries.length - 1; i >= 0; i--) {
     const e = entries[i];
@@ -9346,18 +9247,16 @@ async function handleCompress(args, runtime, ctx, toolCallId) {
     messages: coreMessages,
     state: initialState,
     config,
-    tokenCount: calibrateTokens(sentTokens, runtime.density.densityFor(modelId))
+    tokenCount: sentTokens
   });
   const state = turn.state;
   const messages = turn.messages;
-  const density = runtime.density.densityFor(modelId);
-  const beforeTokens = calibrateTokens(estimateTokens(messages, collectCoveredMessageIds(state), imageTokens), density);
+  const beforeTokens = estimateTokens(messages, collectCoveredMessageIds(state), imageTokens);
   const summaryMaxChars = args.summaryMaxChars;
   const topLevelTopic = args.topic;
   debug.event("compress-in", {
     sid: ctx.sessionManager.getSessionId(),
     modelId,
-    density,
     ranges: ranges.length,
     spans: ranges.map((r) => ({ span: `${r.startId}..${r.endId}`, summaryLen: r.summary.length, summary: r.summary, topic: r.topic ?? topLevelTopic ?? null })),
     blocksBefore: state.blocks.length,
@@ -9389,9 +9288,9 @@ async function handleCompress(args, runtime, ctx, toolCallId) {
     messages: coreMessages,
     state: applied.state,
     config,
-    tokenCount: calibrateTokens(sentTokens, density)
+    tokenCount: sentTokens
   });
-  const afterTokens = calibrateTokens(estimateTokens(afterTurn.messages, collectCoveredMessageIds(applied.state), imageTokens), density);
+  const afterTokens = estimateTokens(afterTurn.messages, collectCoveredMessageIds(applied.state), imageTokens);
   const reclaimed = Math.max(0, beforeTokens - afterTokens);
   const newBlocks = applied.state.blocks.slice(-blocksCreated);
   debug.event("compress-out", {
@@ -9748,6 +9647,27 @@ function formatSize(tokens) {
   return `${(tokens / 1e6).toFixed(1)}M`;
 }
 
+// src/floor-stale.ts
+function validAnchorUsage(u) {
+  if (!u) return false;
+  const total = (u.totalTokens ?? 0) || (u.input ?? 0) + (u.output ?? 0) + (u.cacheRead ?? 0) + (u.cacheWrite ?? 0);
+  return total > 0;
+}
+function usageAnchorPredatesCompression(entries) {
+  let lastUsageIdx = -1;
+  let lastCompressIdx = -1;
+  for (let i = 0; i < entries.length; i++) {
+    const m = entries[i].message;
+    if (!m) continue;
+    if (m.role === "assistant" && m.stopReason !== "aborted" && m.stopReason !== "error" && validAnchorUsage(m.usage)) {
+      lastUsageIdx = i;
+    } else if (m.role === "toolResult" && m.toolName === "compress" && m.toolCallId !== void 0 && m.isError !== true && isCompressSuccessText(extractText(m.content))) {
+      lastCompressIdx = i;
+    }
+  }
+  return lastCompressIdx > lastUsageIdx;
+}
+
 // src/delegate-tool.ts
 import {
   spawn
@@ -11071,15 +10991,16 @@ async function handleStatus(args, runtime, ctx) {
   const { state, coreMessages, entries } = await runtime.stateFor(ctx);
   const config = runtime.configFor(ctx);
   const coveredIds = collectCoveredMessageIds(state);
-  const modelId = ctx.model?.id ?? "default";
   const systemPromptText = getSystemPromptText(ctx);
   const systemPromptTokens = systemPromptText ? defaultCountTokens(systemPromptText) : 0;
   const sentTokens = estimateTokens(coreMessages, coveredIds, collectImageTokens(entries, modelSupportsImages(ctx.model))) + systemPromptTokens;
+  const providerReal = ctx.getContextUsage?.()?.tokens ?? 0;
+  const anchorStale = usageAnchorPredatesCompression(entries);
   const turn = runtime.core.processTurn({
     messages: coreMessages,
     state,
     config,
-    tokenCount: calibrateTokens(sentTokens, runtime.density.densityFor(modelId))
+    tokenCount: anchorStale ? sentTokens : Math.max(sentTokens, providerReal)
   });
   const processed = turn.messages;
   const base = buildStatusReport(turn.state, processed, defaultCountTokens, {
@@ -11094,6 +11015,15 @@ async function handleStatus(args, runtime, ctx) {
   const ranges = viableRanges(nudge?.compressibleRanges ?? []);
   const protectedRanges = nudge?.protectedRanges ?? [];
   const extra = [];
+  if (providerReal > 0 && config.modelContextLimit > 0) {
+    const fmtK = (n) => n >= 1e3 ? `${(n / 1e3).toFixed(1)}k` : String(n);
+    const estPct = Math.round(sentTokens / config.modelContextLimit * 100);
+    const realPct = Math.round(providerReal / config.modelContextLimit * 100);
+    extra.push("");
+    extra.push(
+      `Estimate: ${fmtK(sentTokens)} (${estPct}%)   |   Provider-reported: ${fmtK(providerReal)} (${realPct}%)`
+    );
+  }
   if (nudge) {
     extra.push("");
     extra.push(
@@ -11467,20 +11397,28 @@ function cacheUsageSamples(entries) {
   }
   return out;
 }
-function makeCommands(runtime) {
+function makeCommands(runtime, pi) {
+  const statusHandler = async (_args, ctx) => {
+    const text = await statusReport(runtime, ctx);
+    if (typeof pi?.sendMessage === "function") {
+      pi.sendMessage({ customType: ACP_STATUS_CUSTOM_TYPE, content: text, display: true });
+      return;
+    }
+    ctx.ui.notify(text);
+  };
   return [
     {
       name: "acp",
       options: {
         description: "Show ACP context usage, token breakdown, and compression status.",
-        handler: async (_args, ctx) => ctx.ui.notify(await statusReport(runtime, ctx))
+        handler: statusHandler
       }
     },
     {
       name: "acp-status",
       options: {
         description: "Detailed ACP status (block tiers, token breakdown, delegate usage).",
-        handler: async (_args, ctx) => ctx.ui.notify(await statusReport(runtime, ctx))
+        handler: statusHandler
       }
     },
     {
@@ -11556,16 +11494,16 @@ async function statusReport(runtime, ctx) {
   const { state, coreMessages, entries } = await runtime.stateFor(ctx);
   const config = runtime.configFor(ctx);
   const realUsage = ctx.getContextUsage?.();
+  const anchorStale = usageAnchorPredatesCompression(entries ?? []);
   const systemPromptText = getSystemPromptText(ctx);
   const systemPromptTokens = systemPromptText ? defaultCountTokens(systemPromptText) : 0;
   const imageTokens = collectImageTokens(entries, modelSupportsImages(ctx.model));
   const imageTokensTotal = [...imageTokens.values()].reduce((a, b) => a + b, 0);
-  const sessionTokens = realUsage?.tokens && realUsage.tokens > 0 ? realUsage.tokens : defaultCountTokens(coreMessages.map((m) => m.text ?? "").join("\n")) + imageTokensTotal;
+  const sessionTokens = !anchorStale && realUsage?.tokens && realUsage.tokens > 0 ? realUsage.tokens : defaultCountTokens(coreMessages.map((m) => m.text ?? "").join("\n")) + imageTokensTotal;
   const coveredIds = collectCoveredMessageIds(state);
-  const modelId = ctx.model?.id ?? "default";
   const sentTokens = estimateTokens(coreMessages, coveredIds, imageTokens) + systemPromptTokens;
-  const turn = runtime.core.processTurn({ messages: coreMessages, state, config, tokenCount: calibrateTokens(sentTokens, runtime.density.densityFor(modelId)) });
-  const versionStr = "0.1.52" ? `billion-context-pi@${"0.1.52"}` : void 0;
+  const turn = runtime.core.processTurn({ messages: coreMessages, state, config, tokenCount: anchorStale ? sentTokens : Math.max(sentTokens, realUsage?.tokens ?? 0) });
+  const versionStr = "0.1.53" ? `billion-context-pi@${"0.1.53"}` : void 0;
   let text = buildStatusPanel({
     version: versionStr,
     tokenCount: sessionTokens,
@@ -11928,7 +11866,7 @@ async function autoInstallLatest(latest, extDirOverride) {
       "--no-fund",
       "--no-save"
     ];
-    const prevVersion = (await readPackageJson(join8(extDir, "package.json")))?.version ?? "0.1.52";
+    const prevVersion = (await readPackageJson(join8(extDir, "package.json")))?.version ?? "0.1.53";
     const { code, stderr } = await runNpmImpl(installArgs(latest), { cwd: npmDir, timeout: 6e4 });
     if (code !== 0) {
       logWarn("update", {
@@ -11941,7 +11879,7 @@ async function autoInstallLatest(latest, extDirOverride) {
     }
     const verify = await verifyInstall(npmDir, latest);
     if (!verify.ok) {
-      const rollbackTo = SEMVER_RE.test(prevVersion) ? prevVersion : "0.1.52";
+      const rollbackTo = SEMVER_RE.test(prevVersion) ? prevVersion : "0.1.53";
       logWarn("update", { event: "auto-install-verify-failed", latest, reason: verify.reason, rollbackTo });
       const rb = await runNpmImpl(installArgs(rollbackTo), { cwd: npmDir, timeout: 6e4 });
       logInfo("update", { event: "rollback", from: latest, to: rollbackTo, ok: rb.code === 0 });
@@ -12002,7 +11940,7 @@ async function checkForUpdate(autoUpdate, notify) {
     const runtimeVersion = await getRuntimeVersion();
     const latest = await fetchLatestVersion();
     if (!latest) return;
-    const current = runtimeVersion ?? "0.1.52";
+    const current = runtimeVersion ?? "0.1.53";
     const hasUpdate = isNewer(latest, current);
     debug.event("update-check", {
       current,
@@ -12059,7 +11997,7 @@ function createAcpExtension(adapter = {}) {
     pi.registerTool(makeDecompressTool(runtime));
     pi.registerTool(makeSearchTool(runtime));
     pi.registerTool(makeStatusTool(runtime));
-    for (const { name, options } of makeCommands(runtime)) {
+    for (const { name, options } of makeCommands(runtime, pi)) {
       pi.registerCommand(name, options);
     }
   };
@@ -12074,14 +12012,11 @@ function wireSessionLifecycle(pi, runtime) {
     runtime.clearNudgeTracking();
     runtime.throttleFor(ctx.sessionManager.getSessionId()).reset();
     runtime.clearCompressRetryTracking();
-    const modelId = ctx.model?.id ?? "default";
-    runtime.density.resetModel(modelId);
     resetDelegateUsage();
     setDelegateDisplayUsage("separate");
     const sid = ctx.sessionManager.getSessionId();
-    runtime.clearSessionTracking(sid);
     const modelInfo = ctx.model;
-    logInfo("session", { event: "start", sid, cwd: ctx.cwd, debug: runtime.adapter.debug ?? null, version: true ? "0.1.52" : null, model: modelInfo?.id ?? null, modelApi: modelInfo?.api ?? null, contextWindow: modelInfo?.contextWindow ?? null });
+    logInfo("session", { event: "start", sid, cwd: ctx.cwd, debug: runtime.adapter.debug ?? null, version: true ? "0.1.53" : null, model: modelInfo?.id ?? null, modelApi: modelInfo?.api ?? null, contextWindow: modelInfo?.contextWindow ?? null });
     try {
       await runtime.reloadConfig(ctx.cwd);
       setDelegateDisplayUsage(resolveDelegate(runtime.adapter).displayUsage);
@@ -12117,7 +12052,6 @@ function wireContextTransform(pi, runtime) {
     try {
       await runtime.reloadConfig(ctx.cwd);
       const modelId = ctx.model?.id ?? "default";
-      runtime.setCountModel(modelId);
       const { state, coreMessages, entries } = await runtime.stateFor(ctx, event.messages);
       const configBase = runtime.configFor(ctx);
       const ov = runtime.overflowFor(sid);
@@ -12141,7 +12075,11 @@ function wireContextTransform(pi, runtime) {
       const systemPromptText = getSystemPromptText(ctx);
       const systemPromptTokens = systemPromptText ? defaultCountTokens(systemPromptText) : 0;
       const sentTokens = estimateTokens(coreMessages, coveredIds, collectImageTokens(entries, modelSupportsImages(ctx.model))) + systemPromptTokens;
-      let tokenCount = calibrateTokens(sentTokens, runtime.density.densityFor(modelId));
+      let tokenCount = sentTokens;
+      const realPromptTokens = realUsage?.tokens ?? 0;
+      if (!usageAnchorPredatesCompression(entries) && realPromptTokens > tokenCount) {
+        tokenCount = realPromptTokens;
+      }
       if (ov.armed && config.modelContextLimit > 0) {
         ov.armed = false;
         const floor = Math.floor(config.modelContextLimit * 0.95);
@@ -12150,14 +12088,9 @@ function wireContextTransform(pi, runtime) {
           logWarn("overflow-selfheal", { sid, event: "armed-emergency", tokenCount, limit: config.modelContextLimit });
         }
       }
-      const postCompression = runtime.noteActiveBlocks(
-        sid,
-        state.blocks.filter((b) => b.active).map((b) => b.blockId)
-      );
       debug.event("context-in", {
         sid,
         modelId,
-        density: runtime.density.densityFor(modelId),
         eventMsgs: event.messages?.length ?? 0,
         entries: entries.length,
         coreMsgs: coreMessages.length,
@@ -12169,7 +12102,6 @@ function wireContextTransform(pi, runtime) {
       });
       const turn = runtime.core.processTurn({ messages: coreMessages, state, config, tokenCount });
       await runtime.save(turn.state, ctx);
-      runtime.density.update(modelId, realUsage?.tokens ?? null, sentTokens, postCompression);
       logInfo("turn", {
         sid,
         model: ctx.model?.id ?? null,
@@ -12185,7 +12117,6 @@ function wireContextTransform(pi, runtime) {
       });
       debug.event("processTurn", {
         modelId,
-        density: runtime.density.densityFor(modelId),
         outMsgs: turn.messages.length,
         summaryMsgs: turn.messages.filter((m) => m.id.startsWith("acp_summary")).length,
         prunedMsgs: coreMessages.length - turn.messages.length + turn.messages.filter((m) => m.id.startsWith("acp_summary")).length,