billion-context-omp 0.1.8 → 0.1.9

package/dist/auto-compress.d.ts

@@ -50,6 +50,10 @@ export declare function summarizeMessages(ctx: ExtensionContext, messages: Agent
     customInstructions?: string;
     signal?: AbortSignal;
     completeFn?: typeof complete;
+    /** Fold-slot messageRefs — when provided, formatSlice renders stable
+     *  mNNNNN refs instead of raw pN position ids (issue #14 Minor1: the
+     *  model was shown p-ids it must never echo back). */
+    messageRefs?: CompressionState["messageRefs"];
 }): Promise<{
     summary: string;
     model: string;

package/dist/index.js

@@ -3158,8 +3158,13 @@ function rangeFingerprints(ranges, coreMessages, byRef, blocks) {
 }
 
 // src/runtime.ts
-function freshSlot() {
-  return { identities: [], foldedLen: 0, preview: false, state: createInitialState(), coreMessages: [], appliedCallIds: /* @__PURE__ */ new Set() };
+function freshSlot(preserveFrom) {
+  const slot = { identities: [], foldedLen: 0, preview: false, state: createInitialState(), coreMessages: [], appliedCallIds: /* @__PURE__ */ new Set(), rejectStreak: 0 };
+  if (preserveFrom) {
+    slot.state = { ...slot.state, nudge: preserveFrom.state.nudge };
+    slot.rejectStreak = preserveFrom.rejectStreak;
+  }
+  return slot;
 }
 function stateHasCompressCall(state, callId) {
   return state.blocks.some((b) => b.compressCallId === callId);
@@ -3205,7 +3210,7 @@ function createRuntime(adapter) {
     let slot = slotFor(sid);
     if (slot.preview) {
       debug.event("fold-refold", { sid, foldedLen: slot.foldedLen, lcp: 0, streamLen: stream.length, reason: "preview" });
-      slot = freshSlot();
+      slot = freshSlot(slot);
       slots.set(sid, slot);
     }
     const ids = stream.map(messageIdentity);
@@ -3213,7 +3218,7 @@ function createRuntime(adapter) {
     while (lcp < Math.min(ids.length, slot.identities.length) && ids[lcp] === slot.identities[lcp]) lcp++;
     if (lcp < slot.foldedLen) {
       debug.event("fold-refold", { sid, foldedLen: slot.foldedLen, lcp, streamLen: ids.length });
-      slot = freshSlot();
+      slot = freshSlot(slot);
       slots.set(sid, slot);
       lcp = 0;
     }
@@ -3295,6 +3300,11 @@ function createRuntime(adapter) {
     slot.state = state;
     if (toolCallId) slot.appliedCallIds.add(toolCallId);
   }
+  function noteCompressOutcome(ctx, ok) {
+    const slot = slotFor(sidOf(ctx));
+    slot.rejectStreak = ok ? 0 : slot.rejectStreak + 1;
+    return slot.rejectStreak;
+  }
   return {
     core,
     get adapter() {
@@ -3314,6 +3324,7 @@ function createRuntime(adapter) {
     foldStream,
     stateFor,
     commitFoldState,
+    noteCompressOutcome,
     forgetSession,
     primeFold,
     acquireLock
@@ -3444,7 +3455,7 @@ async function handleCompress(args, runtime, ctx, toolCallId) {
     const invalidRanges = rangeSpecs.filter((r) => !r.startRef || !r.endRef || typeof r.startRef !== "string" || typeof r.endRef !== "string");
     if (invalidRanges.length > 0) {
       logError("compress", { sid: ctx.sessionManager.getSessionId(), event: "invalid-ranges", count: invalidRanges.length, ranges: invalidRanges.map((r) => `${r.startRef}..${r.endRef}`) });
-      return `Rejected: ${invalidRanges.length} range(s) have invalid startId or endId (missing or non-string). All ranges must have valid message refs (e.g. "m00005") or block IDs (e.g. "b3"). No changes applied \u2014 run acp_status for current refs.`;
+      return rejectionMessage(ctx, runtime, `Rejected: ${invalidRanges.length} range(s) have invalid startId or endId (missing or non-string). All ranges must have valid message refs (e.g. "m00005") or block IDs (e.g. "b3"). No changes applied \u2014 run acp_status for current refs.`);
     }
     let applied;
     try {
@@ -3456,12 +3467,13 @@ async function handleCompress(args, runtime, ctx, toolCallId) {
       });
     } catch (e) {
       logThrow("compress", e, { sid: ctx.sessionManager.getSessionId(), phase: "applyCompression", ranges: rangeSpecs.length });
-      return `Compression failed: ${e instanceof Error ? e.message : String(e)}. No changes applied \u2014 state is unchanged.`;
+      return rejectionMessage(ctx, runtime, `Compression failed: ${e instanceof Error ? e.message : String(e)}. No changes applied \u2014 state is unchanged.`);
     }
     if (applied.result.errors.length > 0) {
       logError("compress", { sid: ctx.sessionManager.getSessionId(), event: "apply-errors", count: applied.result.errors.length, errors: applied.result.errors.slice(0, 5) });
-      return `Compression rejected: ${applied.result.errors.join("; ")}. No changes applied \u2014 run acp_status to verify current state.`;
+      return rejectionMessage(ctx, runtime, `Compression rejected: ${applied.result.errors.join("; ")}. No changes applied \u2014 run acp_status to verify current state.`);
     }
+    runtime.noteCompressOutcome(ctx, true);
     await runtime.commitFoldState(ctx, applied.state, toolCallId);
     const { blocksCreated, tokensCompressed, warnings } = applied.result;
     const afterTokens = Math.max(0, beforeTokens - tokensCompressed);
@@ -3502,6 +3514,22 @@ async function handleCompress(args, runtime, ctx, toolCallId) {
     releaseLock();
   }
 }
+var LOOP_GUARD_STOP = 3;
+var LOOP_GUARD_SUPPRESS = 4;
+function rejectionMessage(ctx, runtime, base) {
+  const streak = runtime.noteCompressOutcome(ctx, false);
+  if (streak >= LOOP_GUARD_SUPPRESS) {
+    logWarn("compress", { sid: ctx.sessionManager.getSessionId(), event: "loop-guard", streak, mode: "suppressed" });
+    return `Compression rejected (again \u2014 ${streak} consecutive rejections). No changes applied. STOP calling compress; it is not converging. Continue the task. Compress stays available: a fresh attempt works when acp_status shows a range that can meet the minimum size.`;
+  }
+  if (streak >= LOOP_GUARD_STOP) {
+    logWarn("compress", { sid: ctx.sessionManager.getSessionId(), event: "loop-guard", streak, mode: "stop-directive" });
+    return `${base}
+
+STOP: ${streak} compress calls rejected in a row. Do NOT retry the same range. Run acp_status to see what is actually compressible now; if no range can meet the minimum size, nothing is left to compress \u2014 stop and continue the actual task.`;
+  }
+  return base;
+}
 
 // src/decompress-tool.ts
 import { type as type2 } from "@oh-my-pi/omptype";
@@ -4464,7 +4492,7 @@ async function statusReport(runtime, ctx) {
   const coveredIds = collectCoveredMessageIds(state);
   const sentTokens = estimateTokens(coreMessages, coveredIds) + systemPromptTokens;
   const turn = runtime.core.processTurn({ messages: coreMessages, state, config, tokenCount: sentTokens });
-  const versionStr = "0.1.8" ? `billion-context-omp@${"0.1.8"}` : void 0;
+  const versionStr = "0.1.9" ? `billion-context-omp@${"0.1.9"}` : void 0;
   return buildStatusPanel({
     version: versionStr,
     tokenCount: sessionTokens,
@@ -4570,7 +4598,7 @@ async function summarizeMessages(ctx, messages, prompts, configuredModel, opts)
 User instructions for this compaction: ${custom}`;
     const userText = `ENTIRE conversation to compress (${slice.length} messages, ~${tokens} tokens). Compress it:
 
-` + formatSlice(slice, createInitialState());
+` + formatSlice(slice, opts?.messageRefs ? { ...createInitialState(), messageRefs: opts.messageRefs } : createInitialState());
     const response = await run(
       model,
       { systemPrompt: [instructions], messages: [{ role: "user", content: [{ type: "text", text: userText }], timestamp: Date.now() }] },
@@ -4638,6 +4666,8 @@ WHEN NOT TO COMPRESS
 - Content the current task step is actively reading or reasoning about.
 - Important user messages \u2014 preserve their exact intent, constraints, and acceptance criteria. If a message in the range must stay verbatim, exclude it from the compress range instead of compressing it.
 - Protected tool outputs \u2014 hard-excluded from compression ranges, survive intact in visible context.
+- Nothing left worth compressing \u2014 if no compressible range can meet the minimum size, do NOT call compress at all. Compression is maintenance, never the task goal; when there is nothing to compress, do the task.
+- A rejected compress call \u2014 never retry a rejected range unchanged. Re-check acp_status first; after 3 rejections, stop compressing and continue the task.
 
 ${prompts.howToCompressRules}
 
@@ -4784,7 +4814,7 @@ var PACKAGE_NAME = "billion-context-omp";
 var REGISTRY_URL = `https://registry.npmjs.org/${PACKAGE_NAME}/latest`;
 var SEMVER_RE = /^\d+\.\d+\.\d+(?:[-+][0-9A-Za-z-.]+)?$/;
 var CHECK_INTERVAL_MS = 3 * 60 * 1e3;
-var THROTTLE_FILE = join4(homeDir(), CONFIG_DIR_NAME3, ".billion-context-omp-update-check");
+var throttleFile = () => join4(homeDir(), CONFIG_DIR_NAME3, ".billion-context-omp-update-check");
 var updateInFlight = false;
 function parseVersion(v) {
   return v.replace(/^v/, "").split(".").map((n) => parseInt(n, 10) || 0);
@@ -4800,7 +4830,7 @@ function isNewer(latest, current) {
 }
 async function readLastCheck() {
   try {
-    const data = await readFile(THROTTLE_FILE, "utf-8");
+    const data = await readFile(throttleFile(), "utf-8");
     return parseInt(data.trim(), 10) || 0;
   } catch {
     return 0;
@@ -4808,8 +4838,8 @@ async function readLastCheck() {
 }
 async function writeLastCheck(timestamp) {
   try {
-    await mkdir2(dirname3(THROTTLE_FILE), { recursive: true });
-    await writeFile2(THROTTLE_FILE, String(timestamp), "utf-8");
+    await mkdir2(dirname3(throttleFile()), { recursive: true });
+    await writeFile2(throttleFile(), String(timestamp), "utf-8");
   } catch {
   }
 }
@@ -4891,12 +4921,12 @@ async function checkForUpdate(autoUpdate, notify) {
     const now = Date.now();
     const lastCheck = await readLastCheck();
     if (now - lastCheck < CHECK_INTERVAL_MS) return;
-    await writeLastCheck(now);
     const runtimeVersion = await getRuntimeVersion();
     const res = await fetch(REGISTRY_URL, {
       signal: AbortSignal.timeout(5e3),
       headers: { Accept: "application/json" }
     });
+    await writeLastCheck(now);
     if (!res.ok) {
       logWarn("update", { event: "check-http", status: res.status });
       return;
@@ -4904,7 +4934,7 @@ async function checkForUpdate(autoUpdate, notify) {
     const data = await res.json();
     const latest = data.version;
     if (!latest) return;
-    const current = runtimeVersion ?? "0.1.8";
+    const current = runtimeVersion ?? "0.1.9";
     const hasUpdate = isNewer(latest, current);
     debug.event("update-check", {
       current,
@@ -5181,11 +5211,13 @@ function wireCompactionDisable(pi, runtime) {
       const prep = event.preparation;
       const toSummarize = [...prep.messagesToSummarize ?? [], ...prep.turnPrefixMessages ?? []];
       if (toSummarize.length === 0) return void 0;
+      const slot = await runtime.stateFor(ctx);
       ctx.ui?.notify?.(`ACP: compacting ${toSummarize.length} messages\u2026`, "info");
       const result = await summarizeMessages(ctx, toSummarize, runtime.prompts, runtime.adapter.compress?.compressModel, {
         previousSummary: prep.previousSummary,
         customInstructions: event.customInstructions,
-        signal: event.signal
+        signal: event.signal,
+        messageRefs: slot.state.messageRefs
       });
       if (!result) {
         ctx.ui?.notify?.("ACP: compression fell back to Pi native compaction", "warning");
@@ -5216,7 +5248,7 @@ 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 ? "0.1.8" : null });
+    logInfo("session", { event: "start", sid, cwd: ctx.cwd, debug: runtime.adapter.debug ?? null, version: true ? "0.1.9" : null });
     try {
       const user = await loadUserConfig(ctx.cwd);
       runtime.setAdapter(applyUserConfig(runtime.adapter, user));
@@ -5255,6 +5287,9 @@ function wireContextTransform(pi, runtime) {
       }
       debug.event("context-in-raw", { sid, msgs: input.length });
       const { state, coreMessages, originalById, streamLen } = runtime.foldStream(ctx, input);
+      const preTurnNudgeBaseline = state.nudge.lastPerMessageNudgeTokens;
+      const preTurnNudgeShownTokens = state.nudge.lastNudgeShownTokens;
+      const preTurnNudgeShownByTier = state.nudge.lastShownByTier;
       const config = runtime.configFor(ctx);
       const coveredIds = collectCoveredMessageIds(state);
       const systemPromptTokens = estimateTextTokens2(getSystemPromptText(ctx) ?? "");
@@ -5310,29 +5345,49 @@ function wireContextTransform(pi, runtime) {
         rebuiltMsgs: rebuilt.length
       });
       const debugOn2 = debug.enabled;
+      let nudgeInjected = false;
       if (turn.nudge?.shouldInject) {
-        const emergency = turn.nudge.breakdown?.emergencyOverride === 1;
-        {
-          turn.nudge.compressibleRanges = viableRanges(turn.nudge.compressibleRanges);
-          const rendered = renderNudgeText(turn.nudge, runtime.prompts);
-          const top = [...turn.nudge.compressibleRanges].sort((a, b) => b.tokens - a.tokens)[0];
-          const example = top ? `
+        const lastUser = [...input].reverse().find((m) => m.role === "user");
+        const tailText = lastUser ? JSON.stringify(lastUser.content ?? "") : "";
+        const isFeedbackView = tailText.includes("efficiency nudge to compress early") || tailText.includes("Context limit reached");
+        if (isFeedbackView) {
+          debug.event("nudge-feedback-skip", { sid: ctx.sessionManager.getSessionId(), msgs: input.length });
+        } else {
+          const emergency = turn.nudge.breakdown?.emergencyOverride === 1;
+          const epochReset = turn.state.nudge.lastPerMessageNudgeTokens !== preTurnNudgeBaseline;
+          const prevShown = epochReset ? 0 : preTurnNudgeShownTokens;
+          const cadenceFloor = turn.nudge.breakdown?.growthFloor ?? 0;
+          const suppressed = !emergency && prevShown > 0 && tokenCount - prevShown < cadenceFloor;
+          if (suppressed) {
+            turn.state.nudge.lastNudgeShownTokens = prevShown;
+            turn.state.nudge.lastShownByTier = preTurnNudgeShownByTier;
+            logInfo("nudge", { sid, event: "cadence-suppressed", growth: tokenCount - prevShown, floor: cadenceFloor, pct: Math.round(turn.nudge.contextUsage * 100), reason: turn.nudge.reason });
+            debug.event("nudge-suppressed", { sid, growth: tokenCount - prevShown, floor: cadenceFloor, pct: Math.round(turn.nudge.contextUsage * 100), reason: turn.nudge.reason });
+          } else {
+            nudgeInjected = true;
+            {
+              turn.nudge.compressibleRanges = viableRanges(turn.nudge.compressibleRanges);
+              const rendered = renderNudgeText(turn.nudge, runtime.prompts);
+              const top = [...turn.nudge.compressibleRanges].sort((a, b) => b.tokens - a.tokens)[0];
+              const example = top ? `
 
 Example: compress({ content: [{ startId: "${top.startRef}", endId: "${top.endRef}", summary: "..." }] })` : "";
-          rebuilt.push(nudgeMessage(turn.nudge, turn.state.blocks.filter((b) => b.active), runtime.prompts, example));
-          if (emergency) {
-            logWarn("nudge", { sid: ctx.sessionManager.getSessionId(), event: "emergency-inject", pct: Math.round(turn.nudge.contextUsage * 100), voice: rendered.voice, compressible: turn.nudge.compressibleRanges.length });
-          }
-          if (debugOn2 && ctx.hasUI) {
-            ctx.ui.notify(`[ACP nudge \u2192 context]${emergency ? " [EMERGENCY]" : ""}
+              rebuilt.push(nudgeMessage(turn.nudge, turn.state.blocks.filter((b) => b.active), runtime.prompts, example));
+              if (emergency) {
+                logWarn("nudge", { sid: ctx.sessionManager.getSessionId(), event: "emergency-inject", pct: Math.round(turn.nudge.contextUsage * 100), voice: rendered.voice, compressible: turn.nudge.compressibleRanges.length });
+              }
+              if (debugOn2 && ctx.hasUI) {
+                ctx.ui.notify(`[ACP nudge \u2192 context]${emergency ? " [EMERGENCY]" : ""}
 ${rendered.text}${example}`);
+              }
+              debug.event("nudge-injected", { sid: ctx.sessionManager.getSessionId(), voice: rendered.voice, channels: ["context", debugOn2 ? "terminal" : null].filter(Boolean), emergency, text: rendered.text + example });
+            }
           }
-          debug.event("nudge-injected", { sid: ctx.sessionManager.getSessionId(), voice: rendered.voice, channels: ["context", debugOn2 ? "terminal" : null].filter(Boolean), emergency, text: rendered.text + example });
         }
       }
       dumpContextMessages(rebuilt, {
         sid,
-        injected: turn.nudge?.shouldInject ?? false,
+        injected: nudgeInjected,
         emergency: turn.nudge?.breakdown?.emergencyOverride === 1
       });
       await checkForUpdate(runtime.adapter.autoUpdate ?? true, (msg) => {