@modusensus/dsh-mneme 0.4.1 → 0.4.3-beta.1

package/src/config.js

@@ -4,6 +4,11 @@ export const Config = z.object({
   memoryDir: z.string().default("~/.dsh/memory"),
   autoInject: z.boolean().default(true),
   autoSummarize: z.boolean().default(true),
+  // Optional model override for summarization. When both are non-empty, they
+  // take priority over the session's current model. Empty = use the session's
+  // active provider/model (same as before).
+  summarizeProvider: z.string().default(""),
+  summarizeModel: z.string().default(""),
   maxInjectedItems: z.natural().min(1).max(20).default(5),
   importanceThreshold: z.natural().min(1).max(5).default(3),
   autoDream: z.boolean().default(true),
@@ -12,7 +17,18 @@ export const Config = z.object({
   dreamDelayMs: z.natural().min(0).max(60000).default(2000),
   dreamProvider: z.string(),
   dreamModel: z.string(),
-  dreamMaxTokens: z.natural().min(256).max(32768).default(4096),
+  dreamMaxTokens: z.natural().min(256).max(131072).default(4096),
+  // Pass-through reasoning effort for dream's LLM calls. 'none' (default)
+  // omits the field so the provider's own default applies; low/medium/high
+  // are forwarded verbatim. Useful to cap reasoning spend on thinking-type
+  // models that would otherwise drain the whole token budget and return an
+  // empty body ("no json array in llm output").
+  dreamReasoningEffort: z.union([
+    z.const("low"),
+    z.const("medium"),
+    z.const("high"),
+    z.const("none")
+  ]).default("none"),
   // Rule version for dream adjudication: when this bumps, older dream_runs
   // degrade to historical evidence (their receipts no longer drive live
   // decisions). Default 0 = no versioning in use yet.
@@ -92,31 +108,50 @@ export const Config = z.object({
   // Prefix/semantic search over entity names (used by recall).
   entitySearchEnabled: z.boolean().default(true),
 
-  // --- system-level sleep (v0.4.1) -----------------------------------------
-  // Opt-in: when false (default) the plugin never runs a sleep cycle, so the
-  // access-touch bookkeeping on recall/inject paths stays off too. Sleep is
-  // three phases: conflict resolution (reuses the dream conflict machinery),
-  // archival demotion (unrecalled memories tier down to summary then archive),
-  // and pattern discovery (LLM scans recent memories and mints type=pattern
-  // entries with evidence references).
-  sleepEnabled: z.boolean().default(false),
-  // A sleep run only fires when the store has been idle for this long and the
-  // last run is older than sleepMinIntervalHours. Idle detection replaces a
-  // cron-like schedule (DSH plugins have no resident crontab).
-  sleepIdleMinutes: z.natural().min(1).max(1440).default(30),
+  // --- sleep mode: idle-triggered deep maintenance (v0.4.0) ---------------
+  // Opt-in, off by default. Unlike autoDream (threshold-triggered, lightweight)
+  // sleep fires when the store has been quiet for sleepIdleMinutes and deep-
+  // maintains the whole library: conflict resolution, archival demotion,
+  // pattern discovery and entity relation completion. Abortable on user
+  // activity, audited into dream_runs (run_type='sleep'), and serialized with
+  // autoDream so the two never overlap.
+  sleepModeEnabled: z.boolean().default(false),
+  // Quiet window before a cycle fires (minutes).
+  sleepIdleMinutes: z.natural().min(1).max(60).default(5),
+  // Minimum gap between two sleep runs (hours) — a second idle window within
+  // this interval does not retrigger.
   sleepMinIntervalHours: z.natural().min(1).max(168).default(8),
-  // Unrecalled (COALESCE(last_accessed_at, updated_at, created_at)) beyond
-  // sleepArchiveDays → demote: full body moves to _full_content, content
-  // becomes a one-line summary. Beyond sleepDeepArchiveDays → archive.
-  sleepArchiveDays: z.natural().min(1).max(365).default(30),
-  sleepDeepArchiveDays: z.natural().min(1).max(3650).default(90),
-  // How many most-recent memories the pattern-discovery pass scans.
-  sleepPatternScanCount: z.natural().min(10).max(500).default(100),
-  // Max patterns minted per sleep run (mirrors decisions maxCreatePerRun).
-  sleepMaxPatterns: z.natural().min(1).max(20).default(5),
-  // Optional LLM route override; empty = fall back to agentDefaultModel then
-  // the dream route. Distinct from dreamProvider/dreamModel so the sleep pass
-  // can pin a cheaper model for its bulk summarization.
+  // Conflict adjudication strictness:
+  //   gentle    only high-confidence conflicts (threshold 0.92) are resolved
+  //   normal    standard dream-level (threshold 0.85)
+  //   aggressive low-confidence pairs are also adjudicated (threshold 0.75)
+  sleepConflictStrictness: z.union([
+    z.const("gentle"),
+    z.const("normal"),
+    z.const("aggressive")
+  ]).default("normal"),
+  // Archival demotion tiering (days since last access):
+  //   >= sleepArchiveDays  → shrink to summary, full body kept in _full_content
+  //   >= sleepCompressDays → archived outright (entity relations preserved)
+  sleepArchiveDays: z.natural().min(7).max(365).default(30),
+  sleepCompressDays: z.natural().min(7).max(365).default(90),
+  // Pattern discovery scan window (most recent memories to scan).
+  sleepPatternMinMemories: z.natural().min(10).max(1000).default(100),
+  // How far back pattern discovery considers entity attr changes (days).
+  sleepPatternLookbackDays: z.natural().min(1).max(90).default(30),
+  // Max pattern memories minted per run (0 = disabled).
+  sleepMaxPatternPerRun: z.natural().min(0).max(10).default(3),
+  // Optional LLM route override for sleep's bulk passes (empty = use dream
+  // route / agent default model).
   sleepProvider: z.string().default(""),
   sleepModel: z.string().default(""),
+  // Pass-through reasoning effort for sleep's LLM passes, same semantics as
+  // dreamReasoningEffort: 'none' (default) omits the field; low/medium/high
+  // are forwarded verbatim.
+  sleepReasoningEffort: z.union([
+    z.const("low"),
+    z.const("medium"),
+    z.const("high"),
+    z.const("none")
+  ]).default("none"),
 });

package/src/dream/decisions.js

@@ -1,9 +1,5 @@
 const ACTIONS = new Set(["keep", "merge", "archive", "conflict", "update", "create"]);
 
-// create is used by sleep pattern discovery (v0.4.1). It fabricates a new
-// memory of any known type (default pattern) rather than touching existing ids.
-const CREATE_TYPES = new Set(["pattern", "preference", "project", "decision", "history", "summary"]);
-
 /**
  * Validate a dream decision list against a snapshot of eligible memories.
  * @param decisions - LLM-produced decision list.
@@ -13,7 +9,6 @@ const CREATE_TYPES = new Set(["pattern", "preference", "project", "decision", "h
 export function validateDecisions(decisions, snapshot, options = {}) {
   const errors = [];
   const maxUpdatePerRun = options.maxUpdatePerRun ?? 2;
-  const maxCreatePerRun = options.maxCreatePerRun ?? 5;
   const minAgeHours = options.minAgeHours ?? 24;
   if (!Array.isArray(decisions) || decisions.length === 0) {
     return { ok: false, errors: ["decision list must be a non-empty array"] };
@@ -25,9 +20,16 @@ export function validateDecisions(decisions, snapshot, options = {}) {
       errors.push(`${at}: invalid action ${JSON.stringify(d?.action)}`);
       continue;
     }
-    // create claims no existing id: it fabricates a new memory, so it runs its
-    // own field validation and skips the ids-required check + the claimed set.
-    if (d.action === "create") {
+    const ids = d.action === "conflict" ? [d.winner, d.loser] : (d.ids ?? []);
+    if (d.action === "conflict") {
+      if (!d.winner || !d.loser || d.winner === d.loser) {
+        errors.push(`${at}: conflict needs distinct winner and loser`);
+        continue;
+      }
+    } else if (d.action === "create") {
+      // Mint a fresh memory (sleep pattern discovery). Claims no existing id,
+      // so it skips the claiming loop below; evidence is optional provenance
+      // (already filtered to real ids by the caller) and is stored in content.
       if (typeof d.title !== "string" || !d.title.trim()) {
         errors.push(`${at}: create needs non-empty title`);
         continue;
@@ -38,28 +40,11 @@ export function validateDecisions(decisions, snapshot, options = {}) {
       }
       if (d.importance !== undefined && (!Number.isInteger(d.importance) || d.importance < 1 || d.importance > 5)) {
         errors.push(`${at}: create importance must be an integer 1-5 when provided`);
-        continue;
-      }
-      if (d.type !== undefined && (typeof d.type !== "string" || !CREATE_TYPES.has(d.type))) {
-        errors.push(`${at}: create type must be one of ${[...CREATE_TYPES].join(", ")}`);
-        continue;
       }
-      if (d.evidence !== undefined && !Array.isArray(d.evidence)) {
-        errors.push(`${at}: create evidence must be an array of memory ids`);
-        continue;
-      }
-      if (d.tags !== undefined && !Array.isArray(d.tags)) {
-        errors.push(`${at}: create tags must be an array`);
-        continue;
+      if (typeof d.type !== "string" || !d.type.trim()) {
+        errors.push(`${at}: create needs non-empty type`);
       }
       continue;
-    }
-    const ids = d.action === "conflict" ? [d.winner, d.loser] : (d.ids ?? []);
-    if (d.action === "conflict") {
-      if (!d.winner || !d.loser || d.winner === d.loser) {
-        errors.push(`${at}: conflict needs distinct winner and loser`);
-        continue;
-      }
     } else if (!Array.isArray(d.ids) || d.ids.length === 0) {
       errors.push(`${at}: ${d.action} needs non-empty ids`);
       continue;
@@ -129,9 +114,9 @@ export function validateDecisions(decisions, snapshot, options = {}) {
   if (updateCount > maxUpdatePerRun) {
     errors.push(`too many update decisions: ${updateCount} > ${maxUpdatePerRun}`);
   }
-  // Cap create churn: a pattern-discovery loop fabricating endless new memories
-  // would bloat the store, so a run can mint at most maxCreatePerRun.
+  // Cap pattern minting per run (sleepMaxPatternPerRun passes through here).
   const createCount = decisions.filter((d) => d.action === "create").length;
+  const maxCreatePerRun = options.maxCreatePerRun ?? 5;
   if (createCount > maxCreatePerRun) {
     errors.push(`too many create decisions: ${createCount} > ${maxCreatePerRun}`);
   }
@@ -242,46 +227,31 @@ function applyOne(d, service, snapshot, config = {}) {
     case "archive": return applyArchive(d, service, snapshot);
     case "merge": return applyMerge(d, service, snapshot, config);
     case "conflict": return applyConflict(d, service, snapshot);
-    case "create": return applyCreate(d, service, snapshot);
+    case "create": return applyCreate(d, service, config);
     default: return applyUpdate(d, service, snapshot, config);
   }
 }
 
 /**
- * Mint a new memory (sleep pattern discovery). saveWithDedupe dedupes by
- * (type, title) so a replayed create merges instead of duplicating — the
- * idempotency guard. Evidence ids are folded into tags as `ev:<id>` so a
- * pattern's provenance stays queryable after creation.
+ * Mint a fresh memory (pattern discovery). No existing target, so no CAS guard.
+ * Evidence ids ride in the content so a pattern stays traceable to its source
+ * memories. saveWithDedupe dedupes identical mints (idempotent replay-safe).
  */
-function applyCreate(d, service, snapshot) {
-  const evidence = Array.isArray(d.evidence) ? d.evidence : [];
-  const tags = [
-    ...(Array.isArray(d.tags) ? d.tags : []),
-    ...evidence.map((id) => `ev:${id}`)
-  ];
-  const result = service.saveWithDedupe({
-    type: d.type ?? "pattern",
-    title: d.title,
-    content: d.content,
-    importance: d.importance ?? 3,
-    tags,
-    source: "dream-create"
-  });
-  if (!result?.memory) return "skipped";
-  return {
-    applied: 1,
-    committed: {
-      action: "create",
-      id: result.memory.id,
-      type: d.type ?? "pattern",
-      title: d.title,
-      content: d.content,
-      importance: d.importance ?? 3,
-      evidence,
-      count_before: 0,
-      count_after: 1
-    }
-  };
+function applyCreate(d, service, config = {}) {
+  const title = String(d.title ?? "").trim();
+  const content = String(d.content ?? "").trim();
+  const importance = Number.isInteger(d.importance) ? d.importance : 3;
+  const type = typeof d.type === "string" ? d.type : "pattern";
+  const evidence = Array.isArray(d.evidence)
+    ? d.evidence.filter((id) => typeof id === "string")
+    : [];
+  const body = evidence.length > 0
+    ? `${content}\n\n[证据: ${evidence.join(", ")}]`
+    : content;
+  const created = service.saveWithDedupe({ type, title, content: body, importance });
+  const memory = created?.memory;
+  if (!memory) return "skipped"; // deduped/subsumed: nothing minted, clean no-op
+  return { applied: 1, committed: { action: "create", id: memory.id, type } };
 }
 
 function applyArchive(d, service, snapshot) {

package/src/{sleep.js → dream/sleep.js}

@@ -1,22 +1,30 @@
-// System-level sleep (v0.4.1): an idle-triggered, LLM-assisted deep pass over
-// the memory store. Three independent, fail-safe phases:
+// System-level sleep (v0.4.0): an idle-triggered, LLM-assisted deep pass over
+// the memory store. Four independent, fail-safe phases:
 //   1. conflict resolution — high-similarity same-type pairs are either parked
 //      for review (freeze mode) or adjudicated by the LLM (winner kept / loser
-//      archived), reusing the dream conflict machinery.
+//      archived), reusing the dream conflict machinery. Strictness-graded.
 //   2. archival demotion — memories unreferenced past sleepArchiveDays shrink
 //      to a one-line summary with the full body moved to _full_content; past
-//      sleepDeepArchiveDays they are archived outright.
+//      sleepCompressDays they are archived outright.
 //   3. pattern discovery — the LLM scans the most recent memories and mints
 //      type=pattern entries carrying evidence id references.
+//   4. relation completion — orphan entities (zero relations) get implied
+//      relations completed from memory co-occurrence.
 // Each phase is wrapped so one failure never aborts the others, and a missing
-// LLM route / semantic embedder only skips the phases that need it.
+// LLM route / semantic embedder only skips the phases that need it. A run is
+// abortable via an AbortController signal (user activity) — phases check the
+// signal between batches so a running cycle yields promptly.
 import { randomUUID, createHash } from "node:crypto";
-import { validateDecisions, applyDecisions } from "./dream/decisions.js";
-import { findPotentialConflicts } from "./dream/clustering.js";
-import { buildReceipt } from "./dream.js";
+import { validateDecisions, applyDecisions } from "./decisions.js";
+import { findPotentialConflicts } from "./clustering.js";
+import { buildReceipt } from "../dream.js";
 
 const SUMMARY_MAX = 120;
-const CONFLICT_THRESHOLD = 0.85;
+// Conflict similarity threshold per strictness level (v0.4.0):
+//   gentle     only high-confidence pairs (0.92) — first-time users
+//   normal     standard dream-level (0.85) — default
+//   aggressive low-confidence pairs too (0.75) — bloated stores
+const CONFLICT_THRESHOLDS = { gentle: 0.92, normal: 0.85, aggressive: 0.75 };
 
 const CONFLICT_PROMPT = `你是记忆库冲突仲裁助手。下面是检测到的高相似度记忆对，可能内容矛盾或重复。
 对每一对输出一个 decision 对象：
@@ -92,14 +100,17 @@ function makeSummary(m) {
  * conflict_pending for human review (no LLM). Otherwise the LLM adjudicates:
  * each pair → winner kept / loser archived. Returns a per-run summary.
  */
-async function phaseConflicts(ctx, service, config, logger, runId, semantic = null) {
+async function phaseConflicts(ctx, service, config, logger, runId, semantic = null, signal = null) {
   const embedder = semantic?.embedder;
   const vectorIndex = semantic?.vectorIndex;
   if (!embedder || !vectorIndex || typeof embedder.embed !== "function") {
     return { status: "skipped", reason: "no semantic embedder" };
   }
+  const strictness = config.sleepConflictStrictness ?? "normal";
+  const threshold = CONFLICT_THRESHOLDS[strictness] ?? CONFLICT_THRESHOLDS.normal;
   const memories = service.all().filter((m) => !m.archived && !m.forgotten && m.type !== "summary");
   if (memories.length < 2) return { status: "skipped", reason: "too few memories" };
+  if (signal?.aborted) return { status: "aborted", reason: "user activity" };
 
   // Backfill + collect vectors for every eligible memory (best effort).
   const vectors = new Array(memories.length);
@@ -131,7 +142,7 @@ async function phaseConflicts(ctx, service, config, logger, runId, semantic = nu
   const usableMemories = usable.map((i) => memories[i]);
   const usableVectors = usable.map((i) => vectors[i]);
 
-  const pairs = findPotentialConflicts(usableMemories, usableVectors, CONFLICT_THRESHOLD);
+  const pairs = findPotentialConflicts(usableMemories, usableVectors, threshold);
   if (pairs.length === 0) return { status: "skipped", reason: "no conflicts found" };
 
   // Dedupe: each memory participates in at most one pair, highest similarity
@@ -176,6 +187,9 @@ async function phaseConflicts(ctx, service, config, logger, runId, semantic = nu
     model: route.model,
     purpose: "sleep-conflict",
     maxTokens: 2048,
+    ...(config.sleepReasoningEffort && config.sleepReasoningEffort !== "none"
+      ? { reasoningEffort: config.sleepReasoningEffort }
+      : {}),
     messages: [
       { role: "system", content: [{ type: "text", text: CONFLICT_PROMPT }] },
       { role: "user", content: [{ type: "text", text: listText }] }
@@ -218,20 +232,21 @@ async function phaseConflicts(ctx, service, config, logger, runId, semantic = nu
  * truncations, and the full body is preserved in _full_content so nothing is
  * lost. Deterministic and cheap, so it runs even with no LLM route.
  */
-function phaseDemotion(service, config, logger, runId) {
+function phaseDemotion(service, config, logger, runId, signal = null) {
   const archiveDays = config.sleepArchiveDays ?? 30;
-  const deepArchiveDays = config.sleepDeepArchiveDays ?? 90;
+  const compressDays = config.sleepCompressDays ?? 90;
   const archiveCut = Date.now() - archiveDays * 86400000;
-  const deepCut = Date.now() - deepArchiveDays * 86400000;
+  const compressCut = Date.now() - compressDays * 86400000;
   const demoted = [];
   const archived = [];
   for (const m of service.all()) {
+    if (signal?.aborted) break;
     if (m.archived || m.forgotten) continue;
     const ref = m.last_accessed_at ?? m.updated_at ?? m.created_at;
     if (!ref) continue;
     const t = new Date(ref).getTime();
     if (Number.isNaN(t)) continue;
-    if (t < deepCut) {
+    if (t < compressCut) {
       service.setArchived(m.id, true);
       archived.push(m.id);
     } else if (t < archiveCut) {
@@ -254,25 +269,29 @@ function phaseDemotion(service, config, logger, runId) {
  * The empty snapshot is intentional: create claims no existing id, so the
  * "every id claimed" invariant is trivially satisfied for pure-create lists.
  */
-async function phasePatterns(ctx, service, config, logger, runId) {
+async function phasePatterns(ctx, service, config, logger, runId, signal = null) {
   const route = resolveSleepRoute(ctx, config, logger);
   if (!route) return { status: "skipped", reason: "no llm route" };
-  const limit = config.sleepPatternScanCount ?? 100;
+  const limit = config.sleepPatternMinMemories ?? 100;
   const memories = service
     .list({ limit: 200, includeForgotten: false })
     .filter((m) => !m.archived && m.type !== "summary" && m.type !== "pattern")
     .sort((a, b) => (a.updated_at < b.updated_at ? 1 : -1))
     .slice(0, limit);
   if (memories.length === 0) return { status: "skipped", reason: "no memories to scan" };
+  if (signal?.aborted) return { status: "aborted", reason: "user activity" };
   const listText = memories
     .map((m) => `id=${m.id} | type=${m.type} | importance=${m.importance} | updated=${m.updated_at} | title=${m.title} | content=${m.content}`)
     .join("\n");
-  const maxPatterns = config.sleepMaxPatterns ?? 5;
+  const maxPatterns = config.sleepMaxPatternPerRun ?? 3;
   const text = await streamText(ctx, {
     provider: route.provider,
     model: route.model,
     purpose: "sleep-pattern",
     maxTokens: 2048,
+    ...(config.sleepReasoningEffort && config.sleepReasoningEffort !== "none"
+      ? { reasoningEffort: config.sleepReasoningEffort }
+      : {}),
     messages: [
       { role: "system", content: [{ type: "text", text: PATTERN_PROMPT.replace("N", String(maxPatterns)) }] },
       { role: "user", content: [{ type: "text", text: listText }] }
@@ -305,6 +324,61 @@ async function phasePatterns(ctx, service, config, logger, runId) {
   };
 }
 
+/**
+ * Phase 4 — entity relation completion. Detects orphan entities (zero
+ * relations) and completes implied relations from memory co-occurrence:
+ * entities named in the same memory → related_to; container kinds
+ * (project/module) → part_of; tech-ish pairs → depends_on. Deterministic,
+ * no LLM — cheap, so it runs even without a route. saveRelation is
+ * bookkeeping (no write hook), so it never re-triggers the scheduler.
+ */
+function inferRelationType(a, b) {
+  if ((a.type === "project" || a.type === "module") && a.type !== b.type) return "part_of";
+  if ((b.type === "project" || b.type === "module") && b.type !== a.type) return "part_of";
+  if (/npm|plugin|api|sdk|lib|framework|package|deps?|build/i.test(`${a.name} ${b.name}`)) return "depends_on";
+  return "related_to";
+}
+
+function phaseRelations(service, config, logger, runId, signal = null) {
+  const entities = service.listEntities({ limit: 1000 }) ?? [];
+  if (entities.length < 2) return { status: "skipped", reason: "too few entities" };
+  const orphans = entities.filter((e) => (service.getRelations(e.id) ?? []).length === 0);
+  if (orphans.length === 0) return { status: "skipped", reason: "no orphan entities" };
+  const memories = service.all().filter((m) => !m.archived && !m.forgotten);
+  const seen = new Set();
+  const related = [];
+  const MAX_RELATIONS_PER_ORPHAN = 3;
+  for (const o of orphans) {
+    if (signal?.aborted) break;
+    let made = 0;
+    for (const m of memories) {
+      if (signal?.aborted || made >= MAX_RELATIONS_PER_ORPHAN) break;
+      const text = `${m.title ?? ""} ${m.content ?? ""}`;
+      if (!text.includes(o.name)) continue;
+      for (const other of entities) {
+        if (other.id === o.id || other.name === o.name) continue;
+        const key = [o.id, other.id].sort().join("|");
+        if (seen.has(key)) continue;
+        if (!text.includes(other.name)) continue;
+        const relationType = inferRelationType(o, other);
+        try {
+          service.saveRelation({ from_entity: o.id, to_entity: other.id, relation_type: relationType, memory_id: m.id, metadata: { source: "sleep_relation_completion" } });
+          seen.add(key);
+          related.push({ from: o.id, to: other.id, type: relationType });
+          made++;
+        } catch (error) {
+          logger?.warn?.(`dsh-mneme sleep: relation ${o.id}/${other.id} failed: ${String(error)}`);
+        }
+      }
+    }
+  }
+  return {
+    status: related.length > 0 ? "ok" : "noop",
+    orphanCount: orphans.length,
+    related
+  };
+}
+
 // ---------------------------------------------------------------- run
 
 function deriveStatus(phases) {
@@ -323,10 +397,11 @@ function deriveStatus(phases) {
  * run_type='sleep' audit row (same dream_runs table) so sleep activity is
  * observable alongside consolidation runs.
  */
-export async function runSleep(ctx, service, config, logger, semantic = null) {
+export async function runSleep(ctx, service, config, logger, semantic = null, signal = null) {
   const runId = randomUUID();
   const phases = {};
   const attempt = async (name, fn) => {
+    if (signal?.aborted) return; // user resumed activity — stop before next phase
     try {
       phases[name] = await fn();
     } catch (error) {
@@ -334,9 +409,10 @@ export async function runSleep(ctx, service, config, logger, semantic = null) {
       logger?.warn?.(`dsh-mneme sleep: ${name} phase failed: ${error?.message ?? error}`);
     }
   };
-  await attempt("conflicts", () => phaseConflicts(ctx, service, config, logger, runId, semantic));
-  await attempt("demotion", () => phaseDemotion(service, config, logger, runId));
-  await attempt("patterns", () => phasePatterns(ctx, service, config, logger, runId));
+  await attempt("conflicts", () => phaseConflicts(ctx, service, config, logger, runId, semantic, signal));
+  await attempt("demotion", () => phaseDemotion(service, config, logger, runId, signal));
+  await attempt("patterns", () => phasePatterns(ctx, service, config, logger, runId, signal));
+  await attempt("relations", () => phaseRelations(service, config, logger, runId, signal));
 
   const status = deriveStatus(phases);
   const route = resolveSleepRoute(ctx, config, logger);
@@ -398,11 +474,12 @@ export function createSleepScheduler({
   let running = false;
   let disposed = false;
   let idleTimer = null;
+  let sleepAbort = null;
 
   function armIdleTimer() {
     if (disposed || idleTimer) return;
-    if (config.sleepEnabled !== true) return;
-    const idleMs = (config.sleepIdleMinutes ?? 30) * 60000;
+    if (config.sleepModeEnabled !== true) return;
+    const idleMs = (config.sleepIdleMinutes ?? 5) * 60000;
     const delay = Math.max(0, idleMs - (now() - lastWriteAt)) + 1000;
     idleTimer = setTimeoutFn(async () => {
       idleTimer = null;
@@ -413,18 +490,29 @@ export function createSleepScheduler({
 
   function shouldRun(at = now()) {
     if (disposed || running) return false;
-    if (config.sleepEnabled !== true) return false;
-    if (at - lastWriteAt < (config.sleepIdleMinutes ?? 30) * 60000) return false;
-    if (at - lastRunAt < (config.sleepMinIntervalHours ?? 8) * 3600000) return false;
+    if (config.sleepModeEnabled !== true) return false;
+    if (at - lastWriteAt < (config.sleepIdleMinutes ?? 5) * 60000) return false;
+    // lastRunAt === 0 means never ran — the min-interval check must not block
+    // the very first cycle (a real run stamps a nonzero timestamp).
+    if (lastRunAt > 0 && at - lastRunAt < (config.sleepMinIntervalHours ?? 8) * 3600000) return false;
     return true;
   }
 
   /** Called on writes: resets the idle clock and re-arms the fire timer. The
    *  pending timer is cleared first — a stale timer armed against the old idle
    *  window would otherwise fire early, fail shouldRun, and leave nothing armed
-   *  for the next window (a missed trigger until the next write). */
+   *  for the next window (a missed trigger until the next write).
+   *
+   *  While a sleep run is executing (running=true) the in-flight AbortController
+   *  is NOT aborted: the run's own writes (demoteToSummary / setArchived ride
+   *  the normal write-hook path) would otherwise self-abort the cycle. External
+   *  activity during the run still resets the idle clock here, so no new cycle
+   *  fires until the store is quiet again. */
   function noteWrite() {
     lastWriteAt = now();
+    if (!running && sleepAbort) {
+      sleepAbort.abort(); // user resumed activity — interrupt an idle run
+    }
     if (idleTimer) {
       clearTimeoutFn(idleTimer);
       idleTimer = null;
@@ -435,16 +523,19 @@ export function createSleepScheduler({
   async function maybeSchedule() {
     if (!shouldRun()) return false;
     running = true;
+    const abort = new AbortController();
+    sleepAbort = abort;
     try {
       lastRunAt = now();
       const result = await service.enqueue(() =>
-        onRun ? onRun() : Promise.resolve({ ok: true, skipped: true })
+        onRun ? onRun(abort.signal) : Promise.resolve({ ok: true, skipped: true })
       );
       return !!(result && result.ok);
     } catch (error) {
       logger?.warn?.(`dsh-mneme sleep: run failed: ${error?.message ?? error}`);
       return false;
     } finally {
+      sleepAbort = null;
       running = false;
     }
   }
@@ -455,6 +546,10 @@ export function createSleepScheduler({
       clearTimeoutFn(idleTimer);
       idleTimer = null;
     }
+    if (sleepAbort) {
+      sleepAbort.abort();
+      sleepAbort = null;
+    }
   }
 
   return { noteWrite, maybeSchedule, shouldRun, dispose };

package/src/dream.js

@@ -457,6 +457,9 @@ export function createDreamScheduler({ onRun, thresholdCount = 10, thresholdChar
         model: route.model,
         purpose: "compaction",
         maxTokens: config.dreamMaxTokens ?? 4096,
+        ...(config.dreamReasoningEffort && config.dreamReasoningEffort !== "none"
+          ? { reasoningEffort: config.dreamReasoningEffort }
+          : {}),
         messages: [
           { role: "system", content: [{ type: "text", text: consolidationPrompt }] },
           { role: "user", content: [{ type: "text", text: listText }] }
@@ -600,6 +603,9 @@ export function createDreamScheduler({ onRun, thresholdCount = 10, thresholdChar
         model: route.model,
         purpose: "compaction",
         maxTokens: config.dreamMaxTokens ?? 2048,
+        ...(config.dreamReasoningEffort && config.dreamReasoningEffort !== "none"
+          ? { reasoningEffort: config.dreamReasoningEffort }
+          : {}),
         messages: [
           { role: "system", content: [{ type: "text", text: SUMMARY_PROMPT }] },
           { role: "user", content: [{ type: "text", text: service.all().filter((m) => !m.archived && m.type !== "summary").map((m) => `- ${m.title}: ${m.content}`).join("\n") }] }

package/src/index.js

@@ -5,7 +5,7 @@ import { createTools } from "./tools.js";
 import { createInjector } from "./inject.js";
 import { createSummarizer } from "./summarize.js";
 import { createDreamScheduler } from "./dream.js";
-import { createSleepScheduler, runSleep } from "./sleep.js";
+import { createSleepScheduler, runSleep } from "./dream/sleep.js";
 import { createApi } from "./api.js";
 import { createSettings } from "./settings.js";
 import { createCommandManager } from "./commands.js";
@@ -182,21 +182,18 @@ export const apply = (ctx, config) => {
     service.setDreamHook(() => dream.maybeSchedule(service));
   }
 
-  // Sleep scheduler (v0.4.1): idle-triggered deep pass (conflict resolution +
-  // archival demotion + pattern discovery). opt-in via sleepEnabled; writes
-  // through the service reset the idle clock (setSleepHook), and when the store
-  // stays quiet for sleepIdleMinutes past the sleepMinIntervalHours gate, the
-  // scheduler fires one cycle. The onRun closure reuses the same semantic
-  // pipeline as dream for the conflict phase.
+  // Sleep scheduler (v0.4.0): idle-triggered deep maintenance. Fires when the
+  // store has been quiet for sleepIdleMinutes and re-arms on every write via
+  // noteWrite (hooked to the service's write path). Runs go through
+  // service.enqueue so they serialize with autoDream — the two never overlap.
+  // Abortable on user activity; audited into dream_runs with run_type='sleep'.
   let sleep = null;
-  if (cfg.sleepEnabled) {
+  if (cfg.sleepModeEnabled) {
     sleep = createSleepScheduler({
       service,
       config: cfg,
       logger: ctx.logger,
-      onRun: () => (sleep
-        ? runSleep(ctx, service, cfg, ctx.logger, { embedder, vectorIndex })
-        : Promise.resolve({ ok: true, skipped: true }))
+      onRun: (signal) => (sleep ? runSleep(ctx, service, cfg, ctx.logger, { embedder, vectorIndex }, signal) : Promise.resolve({ ok: true, skipped: true }))
     });
     service.setSleepHook(() => sleep.noteWrite());
   }
@@ -271,7 +268,7 @@ export const apply = (ctx, config) => {
     }
     commands?.dispose();
     if (dream) await dream.dispose();
-    if (sleep) await sleep.dispose();
+    if (sleep) sleep.dispose();
     store.close();
   };
 };