@wolfx/pi-magic-context 0.25.0 → 0.26.0-patch.1

package/dist/subagent-entry.js

@@ -1,4 +1,3 @@
-import { createRequire } from "node:module";
 var __create = Object.create;
 var __getProtoOf = Object.getPrototypeOf;
 var __defProp = Object.defineProperty;
@@ -44,7 +43,6 @@ var __export = (target, all) => {
     });
 };
 var __esm = (fn, res) => () => (fn && (res = fn(fn = 0)), res);
-var __require = /* @__PURE__ */ createRequire(import.meta.url);
 
 // ../plugin/src/shared/harness.ts
 function setHarness(value) {
@@ -9582,6 +9580,41 @@ var MIGRATIONS = [
                 `);
       }
     }
+  },
+  {
+    version: 37,
+    description: "emergency drain catch-up latch + historian drain failure backoff",
+    up: (db) => {
+      const hasSessionMeta = db.prepare("SELECT 1 FROM sqlite_master WHERE type='table' AND name='session_meta'").get();
+      if (!hasSessionMeta)
+        return;
+      ensureColumn(db, "session_meta", "emergency_drain_active", "INTEGER NOT NULL DEFAULT 0");
+      ensureColumn(db, "session_meta", "historian_drain_failure_at", "INTEGER NOT NULL DEFAULT 0");
+    }
+  },
+  {
+    version: 38,
+    description: "durable transform decisions for cache-event cause attribution",
+    up: (db) => {
+      db.exec(`
+                CREATE TABLE IF NOT EXISTS transform_decisions (
+                    session_id         TEXT    NOT NULL,
+                    harness            TEXT    NOT NULL DEFAULT 'opencode',
+                    message_id         TEXT    NOT NULL,
+                    ts_ms              INTEGER NOT NULL,
+                    decision           TEXT    NOT NULL,
+                    materialized       INTEGER NOT NULL DEFAULT 0,
+                    materialize_reason TEXT,
+                    emergency          INTEGER NOT NULL DEFAULT 0,
+                    dropped_tokens     INTEGER NOT NULL DEFAULT 0,
+                    dropped_count      INTEGER NOT NULL DEFAULT 0,
+                    input_tokens       INTEGER NOT NULL DEFAULT 0,
+                    PRIMARY KEY (session_id, harness, message_id)
+                );
+                CREATE INDEX IF NOT EXISTS idx_transform_decisions_session_harness
+                    ON transform_decisions(session_id, harness);
+            `);
+    }
   }
 ];
 var LATEST_MIGRATION_VERSION = MIGRATIONS.reduce((max, m) => Math.max(max, m.version), 0);
@@ -142558,8 +142591,9 @@ function backfillToolOwnersInChunks(db, result) {
 var databases = new Map;
 var persistenceByDatabase = new WeakMap;
 var persistenceErrorByDatabase = new WeakMap;
+var pathByDatabase = new WeakMap;
 var lastSchemaFenceRejection = null;
-var LATEST_SUPPORTED_VERSION = 36;
+var LATEST_SUPPORTED_VERSION = 38;
 function resolveDatabasePath(dbPathOverride) {
   if (dbPathOverride) {
     return { dbDir: dirname2(dbPathOverride), dbPath: dbPathOverride };
@@ -142572,6 +142606,9 @@ function resolveDatabasePath(dbPathOverride) {
   const dbDir = getMagicContextStorageDir();
   return { dbDir, dbPath: join4(dbDir, "context.db") };
 }
+function getDatabasePath(db) {
+  return pathByDatabase.get(db) ?? null;
+}
 function migrateLegacyStorageIfNeeded(targetDbPath, targetDbDir) {
   if (existsSync3(targetDbPath))
     return;
@@ -143086,6 +143123,8 @@ CREATE INDEX IF NOT EXISTS idx_dream_queue_pending ON dream_queue(started_at, en
       recovery_no_eligible_head_count INTEGER NOT NULL DEFAULT 0,
       force_emergency_bypass_window_start INTEGER NOT NULL DEFAULT 0,
       force_emergency_bypass_used INTEGER NOT NULL DEFAULT 0,
+      emergency_drain_active INTEGER NOT NULL DEFAULT 0,
+      historian_drain_failure_at INTEGER NOT NULL DEFAULT 0,
       cached_m0_materialized_at INTEGER,
       cached_m0_session_facts_version INTEGER,
       cached_m0_upgrade_state TEXT,
@@ -143149,6 +143188,23 @@ CREATE INDEX IF NOT EXISTS idx_dream_queue_pending ON dream_queue(started_at, en
     CREATE INDEX IF NOT EXISTS idx_historian_runs_status
       ON historian_runs(status, created_at DESC);
 
+    CREATE TABLE IF NOT EXISTS transform_decisions (
+      session_id         TEXT    NOT NULL,
+      harness            TEXT    NOT NULL DEFAULT 'opencode',
+      message_id         TEXT    NOT NULL,
+      ts_ms              INTEGER NOT NULL,
+      decision           TEXT    NOT NULL,
+      materialized       INTEGER NOT NULL DEFAULT 0,
+      materialize_reason TEXT,
+      emergency          INTEGER NOT NULL DEFAULT 0,
+      dropped_tokens     INTEGER NOT NULL DEFAULT 0,
+      dropped_count      INTEGER NOT NULL DEFAULT 0,
+      input_tokens       INTEGER NOT NULL DEFAULT 0,
+      PRIMARY KEY (session_id, harness, message_id)
+    );
+    CREATE INDEX IF NOT EXISTS idx_transform_decisions_session_harness
+      ON transform_decisions(session_id, harness);
+
     CREATE INDEX IF NOT EXISTS idx_tags_session_tag_number ON tags(session_id, tag_number);
     CREATE INDEX IF NOT EXISTS idx_tags_session_message_id ON tags(session_id, message_id);
     CREATE INDEX IF NOT EXISTS idx_pending_ops_session ON pending_ops(session_id);
@@ -143300,6 +143356,8 @@ CREATE INDEX IF NOT EXISTS idx_dream_queue_pending ON dream_queue(started_at, en
   ensureColumn(db, "session_meta", "recovery_no_eligible_head_count", "INTEGER NOT NULL DEFAULT 0");
   ensureColumn(db, "session_meta", "force_emergency_bypass_window_start", "INTEGER NOT NULL DEFAULT 0");
   ensureColumn(db, "session_meta", "force_emergency_bypass_used", "INTEGER NOT NULL DEFAULT 0");
+  ensureColumn(db, "session_meta", "emergency_drain_active", "INTEGER NOT NULL DEFAULT 0");
+  ensureColumn(db, "session_meta", "historian_drain_failure_at", "INTEGER NOT NULL DEFAULT 0");
   ensureColumn(db, "session_meta", "cached_m0_materialized_at", "INTEGER");
   ensureColumn(db, "session_meta", "cached_m0_session_facts_version", "INTEGER");
   ensureColumn(db, "session_meta", "cached_m0_upgrade_state", "TEXT");
@@ -143378,6 +143436,22 @@ CREATE INDEX IF NOT EXISTS idx_dream_queue_pending ON dream_queue(started_at, en
         failed_at INTEGER NOT NULL,
         UNIQUE(table_name, row_id)
       );
+      CREATE TABLE IF NOT EXISTS transform_decisions (
+        session_id         TEXT    NOT NULL,
+        harness            TEXT    NOT NULL DEFAULT 'opencode',
+        message_id         TEXT    NOT NULL,
+        ts_ms              INTEGER NOT NULL,
+        decision           TEXT    NOT NULL,
+        materialized       INTEGER NOT NULL DEFAULT 0,
+        materialize_reason TEXT,
+        emergency          INTEGER NOT NULL DEFAULT 0,
+        dropped_tokens     INTEGER NOT NULL DEFAULT 0,
+        dropped_count      INTEGER NOT NULL DEFAULT 0,
+        input_tokens       INTEGER NOT NULL DEFAULT 0,
+        PRIMARY KEY (session_id, harness, message_id)
+      );
+      CREATE INDEX IF NOT EXISTS idx_transform_decisions_session_harness
+        ON transform_decisions(session_id, harness);
     `);
   ensureColumn(db, "tags", "harness", "TEXT NOT NULL DEFAULT 'opencode'");
   ensureColumn(db, "pending_ops", "harness", "TEXT NOT NULL DEFAULT 'opencode'");
@@ -143464,7 +143538,9 @@ function healNullIntegerColumns(db) {
     ["protected_tail_drain_tokens", 0],
     ["recovery_no_eligible_head_count", 0],
     ["force_emergency_bypass_window_start", 0],
-    ["force_emergency_bypass_used", 0]
+    ["force_emergency_bypass_used", 0],
+    ["emergency_drain_active", 0],
+    ["historian_drain_failure_at", 0]
   ];
   for (const [column, fallback] of columns) {
     try {
@@ -143540,6 +143616,7 @@ function openDatabase(dbPathOrOptions) {
     setDatabase(db);
     loadToolDefinitionMeasurements(db);
     databases.set(dbPath, db);
+    pathByDatabase.set(db, dbPath);
     persistenceByDatabase.set(db, true);
     persistenceErrorByDatabase.delete(db);
     return db;
@@ -158101,7 +158178,6 @@ var DEFAULT_EXECUTE_THRESHOLD_PERCENTAGE = 65;
 var EXECUTE_THRESHOLD_CAP_MESSAGE = "execute_threshold is capped at 80% for cache safety: a single large agent step can overflow the context window before Magic Context can compact between turns, forcing OpenCode's native compaction (hard to recover from). 80% also leaves headroom below the 85%/95% emergency bands. Use a value between 20 and 80.";
 var DEFAULT_HISTORIAN_TIMEOUT_MS = 300000;
 var DEFAULT_HISTORY_BUDGET_PERCENTAGE = 0.15;
-var DEFAULT_LOCAL_EMBEDDING_MODEL = "Xenova/all-MiniLM-L6-v2";
 var DREAMER_TASKS = [
   "consolidate",
   "verify",
@@ -158141,14 +158217,16 @@ var SidekickConfigSchema = AgentOverrideConfigSchema.extend({
 }).optional();
 var HistorianConfigSchema = AgentOverrideConfigSchema.extend({
   two_pass: exports_external.boolean().default(false).describe("Run a second editor pass over historian output to clean low-signal U: lines and cross-compartment duplicates. Adds ~1 extra API call and ~1.3x cost per historian run. Useful for models without extended thinking support. (default: false)"),
-  thinking_level: PiThinkingLevelSchema.describe("Pi only: explicit thinking level passed as --thinking <level> to Pi historian subagent invocations. Required when using reasoning models (e.g. github-copilot/gpt-5.4) because Pi's default thinking-level resolution can pick a value the provider rejects. OpenCode users set variant instead. Valid: off | minimal | low | medium | high | xhigh")
+  thinking_level: PiThinkingLevelSchema.describe("Pi only: explicit thinking level passed as --thinking <level> to Pi historian subagent invocations. Required when using reasoning models (e.g. github-copilot/gpt-5.4) because Pi's default thinking-level resolution can pick a value the provider rejects. OpenCode users set variant instead. Valid: off | minimal | low | medium | high | xhigh"),
+  disallowed_tools: exports_external.array(exports_external.enum(["*", "read", "aft_outline", "aft_zoom", "aft_search"])).default([]).describe(`OpenCode only. Tools to REMOVE from the historian's default allow-list [read, aft_outline, aft_zoom, aft_search]. Applies to both historian and historian-editor agents. Use ["*"] to strip all tool definitions from the model request — this prevents weak instruction-following models (e.g. mistral-small-latest) from entering tool-calling loops. Individual tool names remove just that tool. Note: a user-supplied historian.permission override can re-allow a tool that disallowed_tools removed — disallowed_tools sets the baseline, permission overrides take precedence. (default: [])`)
 }).optional();
 var BaseEmbeddingConfigSchema = exports_external.object({
-  provider: exports_external.enum(["local", "openai-compatible", "off"]).default("local").describe("Embedding provider. 'local' uses Xenova/all-MiniLM-L6-v2, 'openai-compatible' requires endpoint and model, 'off' disables embeddings."),
-  model: exports_external.string().optional().describe("Embedding model name. Required for openai-compatible, ignored for local."),
+  provider: exports_external.enum(["openai-compatible", "off"]).default("off").describe("Embedding provider. 'openai-compatible' requires endpoint and model, 'off' disables embeddings (default)."),
+  model: exports_external.string().optional().describe("Embedding model name. Required for openai-compatible."),
   endpoint: exports_external.string().optional().describe("API endpoint URL. Required when provider is openai-compatible."),
   api_key: exports_external.string().optional().describe("API key for remote embedding provider (optional)"),
-  input_type: exports_external.string().optional().describe("Optional input_type sent in the embedding request body. Required by some openai-compatible providers (e.g. NVIDIA NIM expects 'query' or 'passage'). Omitted from the request when unset."),
+  input_type: exports_external.string().optional().describe("Default input_type for stored/indexed (passage) embeddings in the request body. Required by some openai-compatible providers (e.g. NVIDIA NIM). Omitted from the request when unset."),
+  query_input_type: exports_external.string().optional().describe("Optional input_type for query (search) embeddings on asymmetric models (e.g. NVIDIA NIM 'query'). When unset, query embeddings use embedding.input_type. Passage/stored content always uses embedding.input_type."),
   truncate: exports_external.string().optional().describe("Optional truncate mode sent in the embedding request body (e.g. NVIDIA NIM accepts 'NONE' | 'START' | 'END'). Omitted from the request when unset."),
   max_input_tokens: exports_external.number().int().positive().optional().describe("Optional maximum input tokens for chunk embeddings. Defaults conservatively to 512 when omitted.")
 }).superRefine((data, ctx) => {
@@ -158168,16 +158246,10 @@ var BaseEmbeddingConfigSchema = exports_external.object({
   }
 });
 var EmbeddingConfigSchema = BaseEmbeddingConfigSchema.transform((data) => {
-  if (data.provider === "local") {
-    return {
-      provider: "local",
-      model: data.model?.trim() || DEFAULT_LOCAL_EMBEDDING_MODEL,
-      ...data.max_input_tokens ? { max_input_tokens: data.max_input_tokens } : {}
-    };
-  }
   if (data.provider === "openai-compatible") {
     const apiKey = data.api_key?.trim();
     const inputType = data.input_type?.trim();
+    const queryInputType = data.query_input_type?.trim();
     const truncate = data.truncate?.trim();
     return {
       provider: "openai-compatible",
@@ -158185,6 +158257,7 @@ var EmbeddingConfigSchema = BaseEmbeddingConfigSchema.transform((data) => {
       endpoint: data.endpoint?.trim() ?? "",
       ...apiKey ? { api_key: apiKey } : {},
       ...inputType ? { input_type: inputType } : {},
+      ...queryInputType ? { query_input_type: queryInputType } : {},
       ...truncate ? { truncate } : {},
       ...data.max_input_tokens ? { max_input_tokens: data.max_input_tokens } : {}
     };
@@ -158224,8 +158297,7 @@ var MagicContextConfigSchema = exports_external.object({
     mmap_size_mb: exports_external.number().min(0).max(8192).default(0).describe("Memory-mapped I/O size in MiB (PRAGMA mmap_size). 0 disables mmap (SQLite default). Raising it can cut read overhead on large DBs at the cost of address space. (min 0, max 8192, default 0)")
   }).default({ cache_size_mb: 64, mmap_size_mb: 0 }).describe("SQLite connection tuning for Magic Context's own context.db. These are per-connection PRAGMAs applied at open; they do not change the schema or what is stored."),
   embedding: EmbeddingConfigSchema.default({
-    provider: "local",
-    model: DEFAULT_LOCAL_EMBEDDING_MODEL
+    provider: "off"
   }).describe("Embedding provider configuration"),
   temporal_awareness: exports_external.boolean().default(true).describe('Inject wall-clock gap markers (<!-- +Xm -->) between user messages where > 5 min elapsed since the previous message, and add start/end date attributes on compartments. Gives the agent a sense of session pacing and "how long ago" across multi-day sessions. Graduated from experimental.temporal_awareness; default: true (set false to opt out).'),
   keep_subagents: exports_external.boolean().default(false).describe("Debug: keep the child sessions Magic Context spawns for its own subagents (historian, dreamer, sidekick, memory-migration) instead of deleting them on success. Useful for short-term inspection/data collection — their full transcript (prompt, tool calls, token usage, output) stays in the host session store. Kept sessions accumulate until manually cleared; leave false for normal use. Requires a restart to take effect."),
@@ -158908,6 +158980,16 @@ function buildCanonicalChunkTextFromFts(db, sessionId, startOrdinal, endOrdinal)
   return lines.join(`
 `);
 }
+function buildCompartmentSummaryFallbackText(db, compartmentId) {
+  const row = db.prepare("SELECT title, p1, content FROM compartments WHERE id = ?").get(compartmentId);
+  if (!row)
+    return "";
+  const title = typeof row.title === "string" ? row.title.trim() : "";
+  const p1 = typeof row.p1 === "string" ? row.p1.trim() : "";
+  const body = p1.length > 0 ? p1 : typeof row.content === "string" ? row.content.trim() : "";
+  return [title, body].filter((s) => s.length > 0).join(`
+`);
+}
 function canonicalizeInMemoryChunkTextForEmbedding(chunkText, startOrdinal, endOrdinal) {
   const lines = [];
   for (const rawLine of chunkText.split(/\r?\n/)) {
@@ -159215,377 +159297,11 @@ function getEmbeddingProviderIdentity(config2) {
     endpoint: normalizeEndpoint2(config2.endpoint),
     apiKeyPresent: Boolean(config2.api_key?.trim()),
     inputType: config2.input_type?.trim() || ""
-  } : {
-    provider: "local",
-    model: config2.model?.trim() || DEFAULT_LOCAL_EMBEDDING_MODEL,
-    endpoint: "",
-    apiKeyPresent: false
-  };
-  return `embedding-provider:${computeNormalizedHash(JSON.stringify(identityInput))}`;
-}
-
-// ../plugin/src/features/magic-context/memory/embedding-local.ts
-import { mkdirSync as mkdirSync3 } from "node:fs";
-import { open, stat, unlink, writeFile } from "node:fs/promises";
-import { dirname as dirname4, join as join6 } from "node:path";
-import { pathToFileURL } from "node:url";
-init_data_path();
-init_logger();
-var LOCK_POLL_MS = 150;
-var STALE_LOCK_MS = 3 * 60000;
-var MAX_LOCK_WAIT_MS = 5 * 60000;
-async function acquireModelLoadLock(lockPath) {
-  const waitStart = Date.now();
-  while (true) {
-    try {
-      const handle = await open(lockPath, "wx");
-      try {
-        await handle.writeFile(`pid=${process.pid} started=${Date.now()}
-`);
-      } catch {}
-      await handle.close();
-      return async () => {
-        try {
-          await unlink(lockPath);
-        } catch {}
-      };
-    } catch (error51) {
-      const code = error51.code;
-      if (code !== "EEXIST" && code !== "EPERM") {
-        throw error51;
-      }
-      try {
-        const info = await stat(lockPath);
-        if (Date.now() - info.mtimeMs > STALE_LOCK_MS) {
-          log(`[magic-context] embedding-load lock stale (>${STALE_LOCK_MS}ms), taking over`);
-          try {
-            await unlink(lockPath);
-          } catch {}
-          continue;
-        }
-      } catch {
-        continue;
-      }
-      if (Date.now() - waitStart > MAX_LOCK_WAIT_MS) {
-        throw new Error(`[magic-context] embedding-load lock wait exceeded ${MAX_LOCK_WAIT_MS}ms; another process is still loading the model. Skipping this init attempt to avoid an unsynchronized native load.`);
-      }
-      await new Promise((resolve3) => setTimeout(resolve3, LOCK_POLL_MS));
-    }
-  }
-}
-function startLockHeartbeat(lockPath) {
-  const HEARTBEAT_MS = Math.floor(STALE_LOCK_MS / 3);
-  const timer = setInterval(() => {
-    writeFile(lockPath, `pid=${process.pid} alive=${Date.now()}
-`).catch(() => {});
-  }, HEARTBEAT_MS);
-  timer.unref?.();
-  return () => clearInterval(timer);
-}
-async function injectWasmOrtForElectron() {
-  if (typeof process === "undefined" || !process.versions?.electron) {
-    return false;
-  }
-  try {
-    const ortWebSpec = `onnxruntime-${"web"}`;
-    const ortWeb = await import(ortWebSpec);
-    try {
-      const { createRequire: createRequireFn } = await import("node:module");
-      const requireFn = createRequireFn(import.meta.url);
-      const pkgPath = requireFn.resolve("onnxruntime-web/package.json");
-      const distDir = join6(dirname4(pkgPath), "dist");
-      const wasmPathsPrefix = `${pathToFileURL(distDir).href}/`;
-      if (ortWeb.env?.wasm) {
-        ortWeb.env.wasm.wasmPaths = wasmPathsPrefix;
-      }
-    } catch (pathError) {
-      log("[magic-context] could not resolve local onnxruntime-web/dist, falling back to default WASM paths:", pathError instanceof Error ? pathError.message : String(pathError));
-    }
-    globalThis[Symbol.for("onnxruntime")] = ortWeb;
-    log("[magic-context] Electron detected — using onnxruntime-web (WASM) for embeddings (bypasses onnxruntime-node native load)");
-    return true;
-  } catch (error51) {
-    log("[magic-context] failed to inject onnxruntime-web for Electron — letting transformers fall back to native:", error51 instanceof Error ? error51.message : String(error51));
-    return false;
-  }
-}
-async function withQuietConsole(fn) {
-  const origWarn = console.warn;
-  const origError = console.error;
-  const redirect = (...args) => {
-    const message = args.map((a) => typeof a === "string" ? a : String(a)).join(" ");
-    log(`[transformers] ${message}`);
-  };
-  console.warn = redirect;
-  console.error = redirect;
-  try {
-    return await fn();
-  } finally {
-    console.warn = origWarn;
-    console.error = origError;
-  }
-}
-var nativeRuntimeMissing = false;
-function isNativeRuntimeMissingError(error51) {
-  const message = error51 instanceof Error ? error51.message : String(error51 ?? "");
-  const lower = message.toLowerCase();
-  const code = error51?.code;
-  const name2 = error51?.name;
-  if (code === "ERR_DLOPEN_FAILED" && lower.includes("onnxruntime")) {
-    return true;
-  }
-  if (!lower.includes("onnxruntime-node"))
-    return false;
-  return code === "ERR_MODULE_NOT_FOUND" || name2 === "ResolveMessage" || lower.includes("cannot find package") || lower.includes("cannot find module") || lower.includes("err_module_not_found");
-}
-function isTransientLoadError(error51) {
-  const message = error51 instanceof Error ? error51.message : String(error51 ?? "");
-  if (!message)
-    return false;
-  const lower = message.toLowerCase();
-  return lower.includes("protobuf parsing failed") || lower.includes("unable to get model file path or buffer") || lower.includes("ebusy") || lower.includes("resource busy") || lower.includes("resource temporarily unavailable");
-}
-function isArrayLikeNumber(value) {
-  if (typeof value !== "object" || value === null || !("length" in value)) {
-    return false;
-  }
-  const arr = value;
-  if (typeof arr.length !== "number") {
-    return false;
-  }
-  return arr.length === 0 || typeof arr[0] === "number";
-}
-function toFloat32Array2(values) {
-  return values instanceof Float32Array ? new Float32Array(values) : Float32Array.from(Array.from(values));
-}
-function extractBatchEmbeddings(result, expectedCount) {
-  const { data } = result;
-  if (Array.isArray(data) && data.length === expectedCount && data.every((entry) => typeof entry !== "number" && isArrayLikeNumber(entry))) {
-    return data.map((entry) => toFloat32Array2(entry));
-  }
-  if (!isArrayLikeNumber(data)) {
-    log("[magic-context] embedding batch returned unexpected data shape");
-    return Array.from({ length: expectedCount }, () => null);
-  }
-  const flatData = toFloat32Array2(data);
-  const dimension = result.dims?.at(-1) ?? flatData.length / expectedCount;
-  if (!Number.isInteger(dimension) || dimension <= 0 || flatData.length !== expectedCount * dimension) {
-    log("[magic-context] embedding batch returned invalid dimensions");
-    return Array.from({ length: expectedCount }, () => null);
-  }
-  const embeddings = [];
-  for (let index = 0;index < expectedCount; index++) {
-    embeddings.push(flatData.slice(index * dimension, (index + 1) * dimension));
-  }
-  return embeddings;
-}
-
-class LocalEmbeddingProvider {
-  modelId;
-  maxInputTokens;
-  model;
-  pipeline = null;
-  initPromise = null;
-  inFlight = 0;
-  disposing = false;
-  disposePromise = null;
-  inFlightWaiters = [];
-  constructor(model = DEFAULT_LOCAL_EMBEDDING_MODEL, maxInputTokens = 512) {
-    this.model = model;
-    this.maxInputTokens = maxInputTokens;
-    this.modelId = getEmbeddingProviderIdentity({ provider: "local", model });
-  }
-  async initialize() {
-    if (this.disposing) {
-      return false;
-    }
-    if (this.pipeline) {
-      return true;
-    }
-    if (nativeRuntimeMissing) {
-      return false;
-    }
-    if (this.initPromise) {
-      await this.initPromise;
-      return this.pipeline !== null;
-    }
-    this.initPromise = (async () => {
-      try {
-        if (this.disposing) {
-          return;
-        }
-        await injectWasmOrtForElectron();
-        const transformersSpec = "@huggingface/transformers";
-        const transformersModule = await import(transformersSpec);
-        const env = transformersModule.env;
-        const LogLevel = transformersModule.LogLevel;
-        if (LogLevel && "ERROR" in LogLevel) {
-          env.logLevel = LogLevel.ERROR;
-        }
-        const modelCacheDir = join6(getMagicContextStorageDir(), "models");
-        try {
-          mkdirSync3(modelCacheDir, { recursive: true });
-          env.cacheDir = modelCacheDir;
-        } catch {
-          log("[magic-context] could not create model cache dir, using library default");
-        }
-        const createPipeline = transformersModule.pipeline;
-        const lockPath = join6(modelCacheDir, ".load.lock");
-        const releaseLock = await acquireModelLoadLock(lockPath);
-        const stopHeartbeat = startLockHeartbeat(lockPath);
-        try {
-          const MAX_ATTEMPTS = 3;
-          let lastError;
-          for (let attempt = 1;attempt <= MAX_ATTEMPTS; attempt++) {
-            try {
-              const pipeline = await withQuietConsole(() => createPipeline("feature-extraction", this.model, {
-                dtype: "fp32"
-              }));
-              if (this.disposing) {
-                await pipeline.dispose?.();
-                this.pipeline = null;
-              } else {
-                this.pipeline = pipeline;
-              }
-              lastError = undefined;
-              break;
-            } catch (error51) {
-              lastError = error51;
-              if (!isTransientLoadError(error51) || attempt === MAX_ATTEMPTS) {
-                break;
-              }
-              const delayMs = 300 * attempt + Math.floor(Math.random() * 200);
-              log(`[magic-context] embedding model load attempt ${attempt}/${MAX_ATTEMPTS} failed transiently, retrying in ${delayMs}ms`);
-              await new Promise((resolve3) => setTimeout(resolve3, delayMs));
-            }
-          }
-          if (this.pipeline) {
-            log(`[magic-context] embedding model loaded: ${this.model}`);
-          } else if (this.disposing) {
-            return;
-          } else {
-            throw lastError ?? new Error("unknown embedding load failure");
-          }
-        } finally {
-          stopHeartbeat();
-          await releaseLock();
-        }
-      } catch (error51) {
-        if (isNativeRuntimeMissingError(error51)) {
-          nativeRuntimeMissing = true;
-          log("[magic-context] local embedding runtime is not installed (onnxruntime-node missing from this install). Local embeddings are disabled. Fix: reinstall the plugin (run `npx @wolfx/magic-context@latest doctor --force`), or configure an `openai-compatible`/`ollama` embedding endpoint instead. Existing memories are unaffected.");
-        } else {
-          log("[magic-context] embedding model failed to load:", error51);
-        }
-        this.pipeline = null;
-      } finally {
-        this.initPromise = null;
-      }
-    })();
-    await this.initPromise;
-    return this.pipeline !== null;
-  }
-  waitForInFlightToDrain() {
-    if (this.inFlight === 0) {
-      return Promise.resolve();
-    }
-    return new Promise((resolve3) => {
-      this.inFlightWaiters.push(resolve3);
-    });
-  }
-  finishInFlight() {
-    this.inFlight = Math.max(0, this.inFlight - 1);
-    if (this.inFlight !== 0)
-      return;
-    const waiters = this.inFlightWaiters.splice(0);
-    for (const waiter of waiters) {
-      waiter();
-    }
-  }
-  async embed(text, signal) {
-    if (signal?.aborted)
-      return null;
-    if (this.disposing)
-      return null;
-    this.inFlight += 1;
-    try {
-      if (!await this.initialize()) {
-        return null;
-      }
-      const pipeline = this.pipeline;
-      if (!pipeline) {
-        return null;
-      }
-      const result = await withQuietConsole(() => pipeline(text, {
-        pooling: "mean",
-        normalize: true
-      }));
-      return extractBatchEmbeddings(result, 1)[0] ?? null;
-    } catch (error51) {
-      log("[magic-context] embedding failed:", error51);
-      return null;
-    } finally {
-      this.finishInFlight();
-    }
-  }
-  async embedBatch(texts, signal) {
-    if (texts.length === 0) {
-      return [];
-    }
-    if (signal?.aborted) {
-      return Array.from({ length: texts.length }, () => null);
-    }
-    if (this.disposing) {
-      return Array.from({ length: texts.length }, () => null);
-    }
-    this.inFlight += 1;
-    try {
-      if (!await this.initialize()) {
-        return Array.from({ length: texts.length }, () => null);
-      }
-      const pipeline = this.pipeline;
-      if (!pipeline) {
-        return Array.from({ length: texts.length }, () => null);
-      }
-      const result = await withQuietConsole(() => pipeline(texts, {
-        pooling: "mean",
-        normalize: true
-      }));
-      return extractBatchEmbeddings(result, texts.length);
-    } catch (error51) {
-      log("[magic-context] embedding batch failed:", error51);
-      return Array.from({ length: texts.length }, () => null);
-    } finally {
-      this.finishInFlight();
-    }
-  }
-  async dispose() {
-    if (this.disposePromise) {
-      return this.disposePromise;
-    }
-    this.disposing = true;
-    this.disposePromise = (async () => {
-      if (this.initPromise) {
-        await this.initPromise;
-      }
-      await this.waitForInFlightToDrain();
-      const pipelineToDispose = this.pipeline;
-      this.pipeline = null;
-      this.initPromise = null;
-      if (!pipelineToDispose) {
-        return;
-      }
-      try {
-        await pipelineToDispose.dispose?.();
-      } catch (error51) {
-        log("[magic-context] embedding model dispose failed:", error51);
-      }
-    })();
-    return this.disposePromise;
-  }
-  isLoaded() {
-    return this.pipeline !== null;
+  } : null;
+  if (!identityInput) {
+    return "embedding-provider:off";
   }
+  return `embedding-provider:${computeNormalizedHash(JSON.stringify(identityInput))}`;
 }
 
 // ../plugin/src/features/magic-context/memory/embedding-openai.ts
@@ -159667,6 +159383,7 @@ class OpenAICompatibleEmbeddingProvider {
   model;
   apiKey;
   inputType;
+  queryInputType;
   truncate;
   initialized = false;
   failureTimes = [];
@@ -159679,6 +159396,7 @@ class OpenAICompatibleEmbeddingProvider {
     this.model = options.model?.trim() ?? "";
     this.apiKey = options.apiKey?.trim() ?? "";
     this.inputType = options.inputType?.trim() ?? "";
+    this.queryInputType = options.queryInputType?.trim() ?? "";
     this.truncate = options.truncate?.trim() ?? "";
     this.maxInputTokens = typeof options.maxInputTokens === "number" && Number.isFinite(options.maxInputTokens) ? Math.max(1, Math.floor(options.maxInputTokens)) : 512;
     this.modelId = getEmbeddingProviderIdentity({
@@ -159706,11 +159424,17 @@ class OpenAICompatibleEmbeddingProvider {
     this.initialized = true;
     return true;
   }
-  async embed(text, signal) {
-    const [embedding] = await this.embedBatch([text], signal);
+  resolveInputTypeForPurpose(purpose = "passage") {
+    if (purpose === "query") {
+      return this.queryInputType || this.inputType;
+    }
+    return this.inputType;
+  }
+  async embed(text, signal, purpose) {
+    const [embedding] = await this.embedBatch([text], signal, purpose);
     return embedding ?? null;
   }
-  async embedBatch(texts, signal) {
+  async embedBatch(texts, signal, purpose) {
     if (texts.length === 0) {
       return [];
     }
@@ -159736,6 +159460,7 @@ class OpenAICompatibleEmbeddingProvider {
       if (signal) {
         signal.addEventListener("abort", onOuterAbort, { once: true });
       }
+      const inputTypeForRequest = this.resolveInputTypeForPurpose(purpose);
       const response = await fetch(`${this.endpoint}/embeddings`, {
         method: "POST",
         headers: {
@@ -159745,7 +159470,7 @@ class OpenAICompatibleEmbeddingProvider {
         body: JSON.stringify({
           model: this.model,
           input: texts,
-          ...this.inputType ? { input_type: this.inputType } : {},
+          ...inputTypeForRequest ? { input_type: inputTypeForRequest } : {},
           ...this.truncate ? { truncate: this.truncate } : {}
         }),
         redirect: "error",
@@ -159901,7 +159626,7 @@ function isEmbeddingRow(row) {
   const candidate = row;
   return typeof candidate.memoryId === "number" && isEmbeddingBlob(candidate.embedding) && (candidate.modelId === null || typeof candidate.modelId === "string");
 }
-function toFloat32Array3(blob) {
+function toFloat32Array2(blob) {
   if (blob instanceof Uint8Array) {
     const buffer2 = blob.buffer.slice(blob.byteOffset, blob.byteOffset + blob.byteLength);
     return new Float32Array(buffer2);
@@ -159949,7 +159674,7 @@ function loadAllEmbeddings(db, projectPath) {
   const embeddings = new Map;
   for (const row of rows) {
     embeddings.set(row.memoryId, {
-      embedding: toFloat32Array3(row.embedding),
+      embedding: toFloat32Array2(row.embedding),
       modelId: row.modelId
     });
   }
@@ -159978,8 +159703,8 @@ function getMemoryEmbedCoverage(db, projectPath, modelId) {
 }
 
 // ../plugin/src/features/magic-context/project-embedding-registry.ts
-import { createHash as createHash6, randomUUID } from "node:crypto";
 init_logger();
+import { createHash as createHash6, randomUUID } from "node:crypto";
 
 // ../plugin/src/features/magic-context/git-commits/storage-git-commit-embeddings.ts
 var saveStatements = new WeakMap;
@@ -160437,18 +160162,13 @@ var loadUnembeddedMemoriesStatements = new WeakMap;
 var globalRegistrationGeneration = 0;
 var testProviderFactory = null;
 function resolveEmbeddingConfig(config2) {
-  if (!config2 || config2.provider === "local") {
-    return {
-      provider: "local",
-      model: config2?.model?.trim() || DEFAULT_LOCAL_EMBEDDING_MODEL,
-      ...config2?.max_input_tokens ? {
-        max_input_tokens: normalizeCompartmentChunkMaxInputTokens(config2.max_input_tokens)
-      } : {}
-    };
+  if (!config2 || config2.provider === "off") {
+    return { provider: "off" };
   }
   if (config2.provider === "openai-compatible") {
     const apiKey = config2.api_key?.trim();
     const inputType = config2.input_type?.trim();
+    const queryInputType = config2.query_input_type?.trim();
     const truncate = config2.truncate?.trim();
     return {
       provider: "openai-compatible",
@@ -160456,6 +160176,7 @@ function resolveEmbeddingConfig(config2) {
       endpoint: config2.endpoint.trim(),
       ...apiKey ? { api_key: apiKey } : {},
       ...inputType ? { input_type: inputType } : {},
+      ...queryInputType ? { query_input_type: queryInputType } : {},
       ...truncate ? { truncate } : {},
       ...config2.max_input_tokens ? {
         max_input_tokens: normalizeCompartmentChunkMaxInputTokens(config2.max_input_tokens)
@@ -160477,11 +160198,12 @@ function createProvider(config2) {
       model: config2.model,
       apiKey: config2.api_key,
       inputType: config2.input_type,
+      queryInputType: config2.query_input_type,
       truncate: config2.truncate,
       maxInputTokens: config2.max_input_tokens
     });
   }
-  return new LocalEmbeddingProvider(config2.model, config2.max_input_tokens);
+  return null;
 }
 function stableStringify2(value) {
   if (Array.isArray(value)) {
@@ -160533,7 +160255,7 @@ function snapshotFor(registration) {
     modelId: registration.observationMode || !providerIsOn ? "off" : registration.modelId,
     chunkModelId: registration.observationMode || !providerIsOn ? "off" : registration.chunkModelId,
     model: registration.observationMode || !providerIsOn ? "off" : ("model" in registration.config) && registration.config.model.trim() ? registration.config.model.trim() : registration.modelId,
-    provider: registration.observationMode || !providerIsOn ? "off" : registration.config.provider ?? "local"
+    provider: registration.observationMode || !providerIsOn ? "off" : registration.config.provider ?? "off"
   };
 }
 function disposeProvider(provider) {
@@ -160659,7 +160381,7 @@ function getOrCreateProjectProvider(registration) {
   registration.provider = provider;
   return provider;
 }
-async function embedTextForProject(projectIdentity, text, signal) {
+async function embedTextForProject(projectIdentity, text, signal, purpose = "passage") {
   const registration = projectRegistrations.get(projectIdentity);
   if (!registration)
     return null;
@@ -160668,7 +160390,7 @@ async function embedTextForProject(projectIdentity, text, signal) {
   const provider = getOrCreateProjectProvider(registration);
   if (!provider)
     return null;
-  const vector = await provider.embed(text, signal);
+  const vector = await provider.embed(text, signal, purpose);
   if (!vector)
     return null;
   const current = projectRegistrations.get(projectIdentity);
@@ -160677,7 +160399,7 @@ async function embedTextForProject(projectIdentity, text, signal) {
   }
   return { vector, modelId, generation };
 }
-async function embedBatchForProject(projectIdentity, texts, signal) {
+async function embedBatchForProject(projectIdentity, texts, signal, purpose = "passage") {
   if (texts.length === 0) {
     const registration2 = projectRegistrations.get(projectIdentity);
     if (!registration2 || registration2.observationMode)
@@ -160693,7 +160415,7 @@ async function embedBatchForProject(projectIdentity, texts, signal) {
   const provider = getOrCreateProjectProvider(registration);
   if (!provider)
     return null;
-  const vectors = await provider.embedBatch(texts, signal);
+  const vectors = await provider.embedBatch(texts, signal, purpose);
   const current = projectRegistrations.get(projectIdentity);
   if (!current || current.generation !== generation || current.runtimeFingerprint !== runtimeFingerprint) {
     return null;
@@ -160750,7 +160472,7 @@ async function embedCandidateChunkBatch(db, projectIdentity, modelId, candidates
   const maxInputTokens = getProjectEmbeddingMaxInputTokens(projectIdentity);
   const prepared = [];
   for (const candidate of candidates) {
-    const canonicalText = buildCanonicalChunkTextFromFts(db, candidate.sessionId, candidate.startMessage, candidate.endMessage);
+    const canonicalText = buildCanonicalChunkTextFromFts(db, candidate.sessionId, candidate.startMessage, candidate.endMessage) || buildCompartmentSummaryFallbackText(db, candidate.id);
     if (canonicalText.length === 0) {
       noWork.push(candidate.id);
       continue;
@@ -160943,8 +160665,7 @@ function getEmbeddingCoverageStatus(db, projectIdentity, sessionId) {
 
 // ../plugin/src/features/magic-context/memory/embedding.ts
 var DEFAULT_EMBEDDING_CONFIG = {
-  provider: "local",
-  model: DEFAULT_LOCAL_EMBEDDING_MODEL
+  provider: "off"
 };
 var embeddingConfig = DEFAULT_EMBEDDING_CONFIG;
 var provider = null;
@@ -160959,11 +160680,12 @@ function createProvider2(config2) {
       model: config2.model,
       apiKey: config2.api_key,
       inputType: config2.input_type,
+      queryInputType: config2.query_input_type,
       truncate: config2.truncate,
       maxInputTokens: config2.max_input_tokens
     });
   }
-  return new LocalEmbeddingProvider(config2.model, config2.max_input_tokens);
+  return null;
 }
 function getOrCreateProvider() {
   if (provider) {
@@ -161194,6 +160916,8 @@ var SESSION_META_SELECT_COLUMNS = [
   "recovery_no_eligible_head_count",
   "force_emergency_bypass_window_start",
   "force_emergency_bypass_used",
+  "emergency_drain_active",
+  "historian_drain_failure_at",
   "upgrade_reminded_at",
   "pi_stable_id_scheme"
 ];
@@ -161243,6 +160967,8 @@ var META_COLUMNS = {
   recoveryNoEligibleHeadCount: "recovery_no_eligible_head_count",
   forceEmergencyBypassWindowStart: "force_emergency_bypass_window_start",
   forceEmergencyBypassUsed: "force_emergency_bypass_used",
+  emergencyDrainActive: "emergency_drain_active",
+  historianDrainFailureAt: "historian_drain_failure_at",
   upgradeRemindedAt: "upgrade_reminded_at",
   piStableIdScheme: "pi_stable_id_scheme"
 };
@@ -161630,9 +161356,9 @@ function escapeXmlContent(s) {
 init_data_path();
 init_logger();
 import { existsSync as existsSync6 } from "node:fs";
-import { join as join7 } from "node:path";
+import { join as join6 } from "node:path";
 function getOpenCodeDbPath() {
-  return join7(getDataDir(), "opencode", "opencode.db");
+  return join6(getDataDir(), "opencode", "opencode.db");
 }
 function openCodeDbExists() {
   return existsSync6(getOpenCodeDbPath());
@@ -161944,6 +161670,24 @@ class ToolMutationBatch {
 }
 
 // ../plugin/src/hooks/magic-context/read-session-chunk.ts
+var BLOCK_TOKEN_MEMO_MAX = 2048;
+var blockTokenMemo = new Map;
+function estimateBlockTokens(blockText) {
+  const cached2 = blockTokenMemo.get(blockText);
+  if (cached2 !== undefined) {
+    blockTokenMemo.delete(blockText);
+    blockTokenMemo.set(blockText, cached2);
+    return cached2;
+  }
+  const count = estimateTokens(blockText);
+  if (blockTokenMemo.size >= BLOCK_TOKEN_MEMO_MAX) {
+    const oldest = blockTokenMemo.keys().next().value;
+    if (oldest !== undefined)
+      blockTokenMemo.delete(oldest);
+  }
+  blockTokenMemo.set(blockText, count);
+  return count;
+}
 var activeRawMessageCache = null;
 var activeAbsoluteCountCache = null;
 var sessionProviders = new Map;
@@ -162126,7 +161870,7 @@ function readSessionChunk(sessionId, tokenBudget, offset = 1, eligibleEndOrdinal
     if (!currentBlock)
       return true;
     const blockText = formatBlock(currentBlock);
-    const blockTokens = estimateTokens(blockText);
+    const blockTokens = estimateBlockTokens(blockText);
     if (totalTokens + blockTokens > tokenBudget && totalTokens > 0) {
       return false;
     }
@@ -167733,12 +167477,12 @@ function canonicalizeDocContent(raw) {
 }
 function fingerprintFile(filePath) {
   try {
-    const stat2 = lstatSync(filePath);
-    const isReadableDoc = stat2.isFile() && stat2.size <= MAX_PROJECT_DOC_BYTES;
+    const stat = lstatSync(filePath);
+    const isReadableDoc = stat.isFile() && stat.size <= MAX_PROJECT_DOC_BYTES;
     return {
       exists: isReadableDoc,
-      mtimeMs: stat2.mtimeMs,
-      size: stat2.size
+      mtimeMs: stat.mtimeMs,
+      size: stat.size
     };
   } catch {
     return { exists: false, mtimeMs: 0, size: 0 };
@@ -168076,7 +167820,9 @@ var DEFAULT_PROTECTED_TAIL_META = {
   protectedTailDrainTokens: 0,
   recoveryNoEligibleHeadCount: 0,
   forceEmergencyBypassWindowStart: 0,
-  forceEmergencyBypassUsed: 0
+  forceEmergencyBypassUsed: 0,
+  emergencyDrainActive: 0,
+  historianDrainFailureAt: 0
 };
 function toProtectedTailMeta(row) {
   if (row === null || typeof row !== "object")
@@ -168090,7 +167836,9 @@ function toProtectedTailMeta(row) {
     protectedTailDrainTokens: numberOr(r.protected_tail_drain_tokens, 0),
     recoveryNoEligibleHeadCount: numberOr(r.recovery_no_eligible_head_count, 0),
     forceEmergencyBypassWindowStart: numberOr(r.force_emergency_bypass_window_start, 0),
-    forceEmergencyBypassUsed: numberOr(r.force_emergency_bypass_used, 0)
+    forceEmergencyBypassUsed: numberOr(r.force_emergency_bypass_used, 0),
+    emergencyDrainActive: numberOr(r.emergency_drain_active, 0),
+    historianDrainFailureAt: numberOr(r.historian_drain_failure_at, 0)
   };
 }
 function loadProtectedTailMeta(db, sessionId) {
@@ -168098,7 +167846,7 @@ function loadProtectedTailMeta(db, sessionId) {
   const row = db.prepare(`SELECT prior_boundary_ordinal, protected_tail_policy_version,
                     protected_tail_drain_window_started_at, protected_tail_drain_tokens,
                     recovery_no_eligible_head_count, force_emergency_bypass_window_start,
-                    force_emergency_bypass_used
+                    force_emergency_bypass_used, emergency_drain_active, historian_drain_failure_at
              FROM session_meta WHERE session_id = ?`).get(sessionId);
   return toProtectedTailMeta(row);
 }
@@ -168142,6 +167890,17 @@ function protectedTailWindowBudget(usagePercentage, usable, perRunCap) {
     return Math.min(750000, Math.max(3 * perRunCap, Math.round(0.35 * usable)));
   return Math.min(500000, Math.max(perRunCap, Math.round(0.2 * usable)));
 }
+var EMERGENCY_DRAIN_ENTER_PERCENTAGE = 95;
+var EMERGENCY_DRAIN_EXIT_MARGIN = 10;
+var EMERGENCY_DRAIN_FALLBACK_EXIT_PERCENTAGE = 55;
+var EMERGENCY_DRAIN_FAILURE_BACKOFF_MS = 60000;
+var EMERGENCY_DRAIN_MAX_LATCH_MS = 30 * 60 * 1000;
+function emergencyDrainExitThreshold(executeThresholdPercentage) {
+  if (!Number.isFinite(executeThresholdPercentage) || executeThresholdPercentage <= 0) {
+    return EMERGENCY_DRAIN_FALLBACK_EXIT_PERCENTAGE;
+  }
+  return Math.max(0, executeThresholdPercentage - EMERGENCY_DRAIN_EXIT_MARGIN);
+}
 function reserveProtectedTailDrainTokens(args) {
   const now = args.now ?? Date.now();
   const requested = Math.max(0, Math.floor(args.trueRawTokens));
@@ -168160,18 +167919,30 @@ function reserveProtectedTailDrainTokens(args) {
     let meta3 = loadProtectedTailMeta(args.db, args.sessionId);
     if (now - meta3.protectedTailDrainWindowStartedAt > DRAIN_WINDOW_MS) {
       args.db.prepare(`UPDATE session_meta
-                     SET protected_tail_drain_window_started_at = ?, protected_tail_drain_tokens = 0,
-                         force_emergency_bypass_window_start = ?, force_emergency_bypass_used = 0
-                     WHERE session_id = ?`).run(now, now, args.sessionId);
+                     SET protected_tail_drain_window_started_at = ?, protected_tail_drain_tokens = 0
+                     WHERE session_id = ?`).run(now, args.sessionId);
       meta3 = loadProtectedTailMeta(args.db, args.sessionId);
     }
+    const exitThreshold = emergencyDrainExitThreshold(args.executeThresholdPercentage);
+    let latchActiveSince = meta3.emergencyDrainActive;
+    if (args.usagePercentage >= EMERGENCY_DRAIN_ENTER_PERCENTAGE) {
+      if (latchActiveSince <= 0)
+        latchActiveSince = now;
+    } else if (latchActiveSince > 0) {
+      const expired = now - latchActiveSince > EMERGENCY_DRAIN_MAX_LATCH_MS;
+      if (args.usagePercentage < exitThreshold || expired)
+        latchActiveSince = 0;
+    }
+    if (latchActiveSince !== meta3.emergencyDrainActive) {
+      args.db.prepare("UPDATE session_meta SET emergency_drain_active = ? WHERE session_id = ?").run(latchActiveSince, args.sessionId);
+    }
+    const latchActive = latchActiveSince > 0;
     const budget = protectedTailWindowBudget(args.usagePercentage, args.usable, args.perRunCap);
     const remaining = Math.max(0, budget - meta3.protectedTailDrainTokens);
     let reserved = Math.min(requested, args.perRunCap, remaining);
     let bypass = false;
-    const bypassWindowExpired = now - meta3.forceEmergencyBypassWindowStart > DRAIN_WINDOW_MS;
-    const bypassUsed = bypassWindowExpired ? 0 : meta3.forceEmergencyBypassUsed;
-    if (reserved <= 0 && args.usagePercentage >= 95 && bypassUsed === 0) {
+    const inFailureBackoff = meta3.historianDrainFailureAt > 0 && now - meta3.historianDrainFailureAt < EMERGENCY_DRAIN_FAILURE_BACKOFF_MS;
+    if (reserved <= 0 && latchActive && !inFailureBackoff) {
       reserved = Math.min(requested, args.perRunCap);
       bypass = true;
     }
@@ -168179,10 +167950,8 @@ function reserveProtectedTailDrainTokens(args) {
       return;
     args.db.prepare(`UPDATE session_meta
                  SET protected_tail_drain_window_started_at = CASE WHEN protected_tail_drain_window_started_at = 0 THEN ? ELSE protected_tail_drain_window_started_at END,
-                     protected_tail_drain_tokens = COALESCE(protected_tail_drain_tokens, 0) + ?,
-                     force_emergency_bypass_window_start = CASE WHEN ? THEN ? ELSE force_emergency_bypass_window_start END,
-                     force_emergency_bypass_used = CASE WHEN ? THEN 1 ELSE force_emergency_bypass_used END
-                 WHERE session_id = ?`).run(now, reserved, bypass ? 1 : 0, now, bypass ? 1 : 0, args.sessionId);
+                     protected_tail_drain_tokens = COALESCE(protected_tail_drain_tokens, 0) + ?
+                 WHERE session_id = ?`).run(now, reserved, args.sessionId);
     result = {
       ok: true,
       reservedTokens: reserved,
@@ -168192,6 +167961,25 @@ function reserveProtectedTailDrainTokens(args) {
   })();
   return result;
 }
+function clearEmergencyDrainLatch(db, sessionId) {
+  db.transaction(() => {
+    ensureSessionMetaRow(db, sessionId);
+    db.prepare("UPDATE session_meta SET emergency_drain_active = 0 WHERE session_id = ?").run(sessionId);
+  })();
+}
+function recordHistorianDrainFailure(db, sessionId, now) {
+  const ts = now ?? Date.now();
+  db.transaction(() => {
+    ensureSessionMetaRow(db, sessionId);
+    db.prepare("UPDATE session_meta SET historian_drain_failure_at = ? WHERE session_id = ?").run(ts, sessionId);
+  })();
+}
+function clearHistorianDrainFailure(db, sessionId) {
+  db.transaction(() => {
+    ensureSessionMetaRow(db, sessionId);
+    db.prepare("UPDATE session_meta SET historian_drain_failure_at = 0 WHERE session_id = ?").run(sessionId);
+  })();
+}
 function rollbackProtectedTailDrainReservation(db, reservation) {
   if (!reservation || reservation.tokens <= 0)
     return;
@@ -169213,12 +169001,17 @@ function getOldestActiveUnprotectedToolTags(db, sessionId, protectedTags = 0, li
     toolName: typeof row.tool_name === "string" ? row.tool_name : null
   }));
 }
-function getTriggerTagTokenUpperBound(db, sessionId) {
-  const row = db.prepare(`SELECT
+function getTriggerTagTokenUpperBound(db, sessionId, floor = 0) {
+  const sql = floor > 0 ? `SELECT
                 COALESCE(SUM(COALESCE(token_count, 0) + COALESCE(input_token_count, 0) + COALESCE(reasoning_token_count, 0)), 0) AS bound,
                 COALESCE(SUM(CASE WHEN token_count IS NULL THEN 1 ELSE 0 END), 0) AS null_count
              FROM tags
-             WHERE session_id = ? AND status IN ('active', 'dropped')`).get(sessionId);
+             WHERE session_id = ? AND status IN ('active', 'dropped') AND tag_number >= ?` : `SELECT
+                COALESCE(SUM(COALESCE(token_count, 0) + COALESCE(input_token_count, 0) + COALESCE(reasoning_token_count, 0)), 0) AS bound,
+                COALESCE(SUM(CASE WHEN token_count IS NULL THEN 1 ELSE 0 END), 0) AS null_count
+             FROM tags
+             WHERE session_id = ? AND status IN ('active', 'dropped')`;
+  const row = floor > 0 ? db.prepare(sql).get(sessionId, floor) : db.prepare(sql).get(sessionId);
   return { bound: row?.bound ?? 0, nullCount: row?.null_count ?? 0 };
 }
 function updateTagInputByteSize(db, sessionId, tagNumber, newInputByteSize) {
@@ -169245,10 +169038,12 @@ function getUpdateTagInputTokenCountStatement(db) {
 function updateTagTokenCount(db, sessionId, tagNumber, newTokenCount) {
   getUpdateTagTokenCountStatement(db).run(newTokenCount, sessionId, tagNumber);
 }
-function getAllStatusTagTokenTotalsFlat(db, sessionId) {
-  const rows = db.prepare(`SELECT type, message_id, tool_owner_message_id, token_count, input_token_count, reasoning_token_count
-             FROM tags
-             WHERE session_id = ?`).all(sessionId);
+function getAllStatusTagTokenTotalsFlat(db, sessionId, floor = 0) {
+  const rows = floor > 0 ? db.prepare(`SELECT type, message_id, tool_owner_message_id, token_count, input_token_count, reasoning_token_count
+                       FROM tags
+                       WHERE session_id = ? AND tag_number >= ?`).all(sessionId, floor) : db.prepare(`SELECT type, message_id, tool_owner_message_id, token_count, input_token_count, reasoning_token_count
+                       FROM tags
+                       WHERE session_id = ?`).all(sessionId);
   const totals = new Map;
   const nullMessageIds = new Set;
   for (const row of rows) {
@@ -169367,6 +169162,52 @@ function getTagNumberByMessageId(db, sessionId, messageId) {
   const row = getTagNumberByMessageIdStatement(db).get(sessionId, messageId);
   return isTagNumberRow(row) ? row.tag_number : null;
 }
+var getMinMessageTagNumberForRawIdStatements = new WeakMap;
+function isMinTagNumberRow(row) {
+  return row !== null && typeof row === "object" && "m" in row;
+}
+function getMinMessageTagNumberForRawId(db, sessionId, rawId) {
+  if (rawId.includes(":"))
+    return null;
+  let stmt = getMinMessageTagNumberForRawIdStatements.get(db);
+  if (!stmt) {
+    stmt = db.prepare("SELECT MIN(tag_number) AS m FROM tags WHERE session_id = ? AND message_id >= ? AND message_id < ?");
+    getMinMessageTagNumberForRawIdStatements.set(db, stmt);
+  }
+  const row = stmt.get(sessionId, `${rawId}:`, `${rawId};`);
+  return isMinTagNumberRow(row) && typeof row.m === "number" ? row.m : null;
+}
+var TAGGER_FLOOR_SCAN_MESSAGES = 8;
+var TAGGER_FLOOR_MAX_PROBES = 64;
+var TAGGER_FLOOR_SAFETY_MARGIN = 256;
+var TAGGER_FLOOR_PER_SKIP_MARGIN = 64;
+function deriveTagLoadFloor(db, sessionId, rawIds) {
+  let min = Number.POSITIVE_INFINITY;
+  let probes = 0;
+  let hits = 0;
+  let skippedBeforeFirstHit = 0;
+  for (const rawId of rawIds) {
+    if (typeof rawId !== "string" || rawId.length === 0)
+      continue;
+    if (probes >= TAGGER_FLOOR_MAX_PROBES)
+      break;
+    probes++;
+    const m = getMinMessageTagNumberForRawId(db, sessionId, rawId);
+    if (m === null) {
+      if (hits === 0)
+        skippedBeforeFirstHit++;
+      continue;
+    }
+    if (m < min)
+      min = m;
+    if (++hits >= TAGGER_FLOOR_SCAN_MESSAGES)
+      break;
+  }
+  if (!Number.isFinite(min))
+    return 0;
+  const margin = TAGGER_FLOOR_SAFETY_MARGIN + skippedBeforeFirstHit * TAGGER_FLOOR_PER_SKIP_MARGIN;
+  return Math.max(0, min - margin);
+}
 var TAG_SELECT_COLUMNS = "id, message_id, type, status, drop_mode, tool_name, input_byte_size, byte_size, reasoning_byte_size, session_id, tag_number, caveman_depth, tool_owner_message_id";
 function getTagsBySession(db, sessionId) {
   const rows = db.prepare(`SELECT ${TAG_SELECT_COLUMNS} FROM tags WHERE session_id = ? ORDER BY tag_number ASC, id ASC`).all(sessionId).filter(isTagRow);
@@ -171540,13 +171381,13 @@ function extractM0Block(m0Text, tag) {
 
 // ../plugin/src/hooks/magic-context/key-files-block.ts
 import { readFileSync as readFileSync6, realpathSync as realpathSync2 } from "node:fs";
-import { join as join9, sep as sep2 } from "node:path";
+import { join as join8, sep as sep2 } from "node:path";
 
 // ../plugin/src/features/magic-context/key-files/aft-availability.ts
 var import_comment_json2 = __toESM(require_src2(), 1);
 import { existsSync as existsSync7, readFileSync as readFileSync5 } from "node:fs";
 import { homedir as homedir4 } from "node:os";
-import { isAbsolute as isAbsolute2, join as join8, resolve as resolve3 } from "node:path";
+import { isAbsolute as isAbsolute2, join as join7, resolve as resolve3 } from "node:path";
 import { fileURLToPath } from "node:url";
 var overrideAvailability = null;
 function parseConfig(path5) {
@@ -171567,14 +171408,14 @@ function resolveLocalEntryPackageName(value, configDir) {
       return null;
     }
   } else if (value.startsWith("~/")) {
-    dir = join8(homedir4(), value.slice(2));
+    dir = join7(homedir4(), value.slice(2));
   } else if (value.startsWith("/") || value.startsWith("./") || value.startsWith("../")) {
     dir = isAbsolute2(value) ? value : resolve3(configDir, value);
   } else {
     return null;
   }
   try {
-    const pkg = JSON.parse(readFileSync5(join8(dir, "package.json"), "utf-8"));
+    const pkg = JSON.parse(readFileSync5(join7(dir, "package.json"), "utf-8"));
     return typeof pkg.name === "string" ? pkg.name : null;
   } catch {
     return null;
@@ -171605,16 +171446,16 @@ function hasAftAtKeys(value, keys2, configDir) {
 function getAftAvailability() {
   const home = process.env.HOME || homedir4();
   const opencodePaths = [
-    join8(home, ".config", "opencode", "opencode.jsonc"),
-    join8(home, ".config", "opencode", "opencode.json")
+    join7(home, ".config", "opencode", "opencode.jsonc"),
+    join7(home, ".config", "opencode", "opencode.json")
   ];
-  const piPaths = [join8(home, ".pi", "agent", "settings.json")];
+  const piPaths = [join7(home, ".pi", "agent", "settings.json")];
   const checkedPaths = [...opencodePaths, ...piPaths];
   let opencode = false;
   for (const path5 of opencodePaths) {
     try {
       const config2 = parseConfig(path5);
-      const configDir = join8(path5, "..");
+      const configDir = join7(path5, "..");
       if (hasAftAtKeys(config2, ["plugin", "plugins", "mcp", "mcp_servers"], configDir)) {
         opencode = true;
         break;
@@ -171625,7 +171466,7 @@ function getAftAvailability() {
   for (const path5 of piPaths) {
     try {
       const config2 = parseConfig(path5);
-      const configDir = join8(path5, "..");
+      const configDir = join7(path5, "..");
       if (hasAftAtKeys(config2, ["packages", "extensions"], configDir)) {
         pi = true;
         break;
@@ -171704,7 +171545,7 @@ function buildKeyFilesBlock(db, projectPath, config2 = { enabled: true, tokenBud
     let nextStale = null;
     let observed = false;
     try {
-      const absPath = join9(projectPath, row.path);
+      const absPath = join8(projectPath, row.path);
       const real = realpathSync2(absPath);
       if (!isUnderResolvedRoot(projectRoot, real)) {
         nextStale = "missing";
@@ -172105,7 +171946,7 @@ function createCtxSearchTool(deps) {
         memoryEnabled,
         embeddingEnabled,
         embedQuery: async (text, signal) => {
-          const result = await embedTextForProject(projectIdentity, text, signal);
+          const result = await embedTextForProject(projectIdentity, text, signal, "query");
           return result?.vector ?? null;
         },
         isEmbeddingRuntimeEnabled: () => embeddingEnabled === true,