@wolfx/pi-magic-context 0.25.0 → 0.26.0

package/dist/subagent-entry.js

@@ -9582,6 +9582,41 @@ 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 +142593,9 @@ 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 +142608,9 @@ 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 +143125,8 @@ 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 +143190,23 @@ 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 +143358,8 @@ 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 +143438,22 @@ 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 +143540,9 @@ 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 +143618,7 @@ function openDatabase(dbPathOrOptions) {
     setDatabase(db);
     loadToolDefinitionMeasurements(db);
     databases.set(dbPath, db);
+    pathByDatabase.set(db, dbPath);
     persistenceByDatabase.set(db, true);
     persistenceErrorByDatabase.delete(db);
     return db;
@@ -158141,14 +158220,16 @@ 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."),
   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) => {
@@ -158178,6 +158259,7 @@ var EmbeddingConfigSchema = BaseEmbeddingConfigSchema.transform((data) => {
   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 +158267,7 @@ 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 } : {}
     };
@@ -158908,6 +158991,16 @@ 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/)) {
@@ -159502,7 +159595,7 @@ class LocalEmbeddingProvider {
       waiter();
     }
   }
-  async embed(text, signal) {
+  async embed(text, signal, _purpose) {
     if (signal?.aborted)
       return null;
     if (this.disposing)
@@ -159528,7 +159621,7 @@ class LocalEmbeddingProvider {
       this.finishInFlight();
     }
   }
-  async embedBatch(texts, signal) {
+  async embedBatch(texts, signal, _purpose) {
     if (texts.length === 0) {
       return [];
     }
@@ -159667,6 +159760,7 @@ class OpenAICompatibleEmbeddingProvider {
   model;
   apiKey;
   inputType;
+  queryInputType;
   truncate;
   initialized = false;
   failureTimes = [];
@@ -159679,6 +159773,7 @@ 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 +159801,17 @@ 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 +159837,7 @@ 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 +159847,7 @@ 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",
@@ -160449,6 +160551,7 @@ function resolveEmbeddingConfig(config2) {
   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 +160559,7 @@ 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)
@@ -160477,6 +160581,7 @@ 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
     });
@@ -160659,7 +160764,7 @@ 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 +160773,7 @@ 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 +160782,7 @@ 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 +160798,7 @@ 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 +160855,7 @@ 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;
@@ -160959,6 +161064,7 @@ 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
     });
@@ -161194,6 +161300,8 @@ 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 +161351,8 @@ 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"
 };
@@ -161944,6 +162054,24 @@ 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 +162254,7 @@ 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;
     }
@@ -168076,7 +168204,9 @@ 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 +168220,9 @@ 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 +168230,7 @@ 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 +168274,17 @@ 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 +168303,30 @@ 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 +168334,8 @@ 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 +168345,25 @@ 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 +169385,17 @@ 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 +169422,12 @@ 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 +169546,52 @@ 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);
@@ -172105,7 +172330,7 @@ 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,

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@wolfx/pi-magic-context",
-	"version": "0.25.0",
+	"version": "0.26.0",
 	"type": "module",
 	"description": "Pi coding agent extension for Magic Context — cross-session memory and context management",
 	"main": "dist/index.js",
@@ -26,7 +26,7 @@
 		"README.md"
 	],
 	"scripts": {
-		"build": "bun build src/index.ts src/subagent-entry.ts --outdir dist --target node --format esm --external @earendil-works/pi-coding-agent --external @earendil-works/pi-tui --external @huggingface/transformers --external better-sqlite3",
+		"build": "bun build src/index.ts src/subagent-entry.ts --outdir dist --target node --format esm --external @earendil-works/pi-coding-agent --external @earendil-works/pi-tui --external @huggingface/transformers --external node:sqlite",
 		"typecheck": "tsc --noEmit",
 		"test": "bun test",
 		"lint": "biome check src",
@@ -49,7 +49,6 @@
 		"@earendil-works/pi-tui": "^0.74.0",
 		"@types/better-sqlite3": "^7.6.13",
 		"@types/bun": "^1.3.10",
-		"better-sqlite3": "^12.9.0",
 		"@types/node": "^22.0.0",
 		"typescript": "^5.8.0"
 	},