clawmem 0.1.7 → 0.1.8

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "clawmem",
-  "version": "0.1.7",
+  "version": "0.1.8",
   "description": "On-device context engine and memory for AI agents. Claude Code and OpenClaw. Hooks + MCP server + hybrid RAG search.",
   "type": "module",
   "bin": {

package/src/clawmem.ts

@@ -1237,8 +1237,22 @@ async function cmdWatch() {
     // Skill vault not configured — skip
   }
 
+  // Periodic WAL checkpoint: the watcher holds a long-lived DB connection which
+  // prevents SQLite auto-checkpoint from shrinking the WAL file. Without this,
+  // the WAL grows unbounded (observed 77MB+), slowing every concurrent DB access
+  // (hooks, MCP) and eventually causing UserPromptSubmit hook timeouts.
+  const WAL_CHECKPOINT_INTERVAL = 5 * 60 * 1000; // 5 minutes
+  const checkpointTimer = setInterval(() => {
+    try {
+      s.db.exec("PRAGMA wal_checkpoint(PASSIVE)");
+    } catch {
+      // Checkpoint failed (busy) — will retry next interval
+    }
+  }, WAL_CHECKPOINT_INTERVAL);
+
   // Keep running until Ctrl+C
   process.on("SIGINT", () => {
+    clearInterval(checkpointTimer);
     watcher.close();
     skillWatcher?.close();
     closeStore();

package/src/hooks/staleness-check.ts

@@ -51,6 +51,33 @@ export async function stalenessCheck(
     }
   }
 
+  // Auto-archive if lifecycle policy is configured (runs regardless of stale report results)
+  try {
+    const { loadVaultConfig } = await import("../config.ts");
+    const config = loadVaultConfig();
+    if (config.lifecycle) {
+      const candidates = store.getArchiveCandidates(config.lifecycle);
+      if (candidates.length > 0 && !config.lifecycle.dry_run) {
+        const archived = store.archiveDocuments(candidates.map(c => c.id));
+        if (archived > 0 && input.sessionId) {
+          store.insertUsage({
+            sessionId: input.sessionId,
+            timestamp: now.toISOString(),
+            hookName: "lifecycle-archive",
+            injectedPaths: candidates.map(c => `${c.collection}/${c.path}`),
+            estimatedTokens: 0,
+            wasReferenced: 0,
+          });
+        }
+      }
+      if (config.lifecycle.purge_after_days && !config.lifecycle.dry_run) {
+        store.purgeArchivedDocuments(config.lifecycle.purge_after_days);
+      }
+    }
+  } catch {
+    // Fail-open: lifecycle errors never block the hook
+  }
+
   if (allStale.size === 0) return makeEmptyOutput("staleness-check");
 
   // Build context within budget

package/src/hooks.ts

@@ -84,6 +84,7 @@ const HOOK_EVENT_MAP: Record<string, string | null> = {
   "feedback-loop": null,           // Stop — no hookSpecificOutput
   "precompact-extract": null,      // PreCompact — side-effect only, no context injection
   "postcompact-inject": "SessionStart", // SessionStart(compact) — injects additionalContext
+  "curator-nudge": "SessionStart",     // SessionStart — surfaces curator report actions
   "pretool-inject": null,          // PreToolUse — disabled (cannot inject additionalContext; E13 folded into context-surfacing)
 };
 

package/src/openclaw/index.ts

@@ -19,6 +19,36 @@ import { resolveClawMemBin, execHook, parseHookOutput, extractContext } from "./
 import type { ClawMemConfig } from "./shell.js";
 import { createTools } from "./tools.js";
 
+// =============================================================================
+// Prompt Cleaning (strips OpenClaw noise for better retrieval)
+// Pattern extracted from memory-core-plus (MIT, aloong-planet)
+// =============================================================================
+
+/**
+ * Strip OpenClaw-specific noise from the user prompt before using it as a
+ * search query. Gateway prompts contain metadata, system events, timestamps,
+ * and previously injected context that degrade embedding/BM25 quality.
+ */
+function cleanPromptForSearch(prompt: string): string {
+  let cleaned = prompt;
+  // Strip previously injected vault-context (avoid re-searching our own output)
+  cleaned = cleaned.replace(/<vault-context>[\s\S]*?<\/vault-context>/g, "");
+  cleaned = cleaned.replace(/<vault-routing>[\s\S]*?<\/vault-routing>/g, "");
+  cleaned = cleaned.replace(/<vault-session>[\s\S]*?<\/vault-session>/g, "");
+  // Strip OpenClaw sender metadata block
+  cleaned = cleaned.replace(/Sender\s*\(untrusted metadata\)\s*:\s*```json\n[\s\S]*?```/g, "");
+  cleaned = cleaned.replace(/Sender\s*\(untrusted metadata\)\s*:\s*\{[\s\S]*?\}\s*/g, "");
+  // Strip OpenClaw runtime context blocks
+  cleaned = cleaned.replace(/OpenClaw runtime context \(internal\):[\s\S]*?(?=\n\n|\n?$)/g, "");
+  // Strip "System: ..." single-line event entries
+  cleaned = cleaned.replace(/^System:.*$/gm, "");
+  // Strip timestamp prefixes e.g. "[Sat 2026-03-14 16:19 GMT+8] "
+  cleaned = cleaned.replace(/^\[.*?GMT[+-]\d+\]\s*/gm, "");
+  // Collapse excessive whitespace
+  cleaned = cleaned.replace(/\n{3,}/g, "\n\n").trim();
+  return cleaned || prompt;
+}
+
 // =============================================================================
 // Plugin Definition
 // =============================================================================
@@ -101,9 +131,11 @@ const clawmemPlugin = {
       }
 
       // Every turn: run context-surfacing for prompt-aware retrieval
+      // Clean the prompt to remove OpenClaw noise before search
+      const searchPrompt = cleanPromptForSearch(event.prompt);
       const surfacingResult = await execHook(cfg, "context-surfacing", {
         session_id: sessionId,
-        prompt: event.prompt,
+        prompt: searchPrompt,
       });
 
       if (surfacingResult.exitCode === 0) {