@solongate/proxy 0.59.1 → 0.59.2

package/hooks/guard.bundled.mjs

@@ -6535,7 +6535,7 @@ import { resolve, join, dirname, isAbsolute } from "node:path";
 import { homedir } from "node:os";
 import { gunzipSync } from "node:zlib";
 import { createHash } from "node:crypto";
-var HOOK_VERSION = 30;
+var HOOK_VERSION = 31;
 function localLogsOnly(security) {
   const l = security && security.localLogs;
   return !!(l && l.enabled && typeof l.path === "string" && l.path.trim());
@@ -7646,7 +7646,7 @@ process.stdin.on("end", async () => {
     let securityCfg = null;
     const agentKey = (AGENT_ID || "default").replace(/[^a-zA-Z0-9_-]/g, "_");
     const policyCacheFile = join(resolve(homedir(), ".solongate"), ".policy-cache-" + agentKey + ".json");
-    const POLICY_TTL_MS = 6e4;
+    const POLICY_TTL_MS = 1e4;
     try {
       let dashboardPolicy = null;
       let staleCache = null;

package/hooks/guard.mjs

@@ -31,7 +31,7 @@ import { createHash } from 'node:crypto';
 // the installed hook self-updates when the cloud version is higher (see
 // maybeSelfUpdate). This is what makes guard fixes propagate without a manual
 // reinstall — the same trust model as the OPA WASM this hook already runs.
-const HOOK_VERSION = 30;
+const HOOK_VERSION = 31;
 
 // True when local log storage is ON. In that mode logs are kept LOCAL ONLY and
 // nothing is sent to the cloud audit log.
@@ -1432,11 +1432,14 @@ process.stdin.on('end', async () => {
     const agentKey = (AGENT_ID || 'default').replace(/[^a-zA-Z0-9_-]/g, '_');
     const policyCacheFile = join(resolve(homedir(), '.solongate'), '.policy-cache-' + agentKey + '.json');
     // Serve the cached policy without any network for TTL seconds. This is a HOT
-    // PATH — it runs before every single tool call — so a short TTL means we
-    // block on a cloud round-trip on nearly every action. A dashboard-managed
-    // policy changes rarely, so 60s freshness is plenty; the previous 3s made the
-    // cache useless and put an 8s-timeout fetch in front of most tool calls.
-    const POLICY_TTL_MS = 60_000;
+    // PATH — it runs before every tool call. The original 3s TTL was slow not
+    // because of the number but because it paired an 8s-timeout blocking fetch
+    // with an every-3s cadence (and stacked behind the WASM fetch + a shared
+    // rate limit). With those fixed — 2s timeout, WASM decoupled (24h/fp),
+    // last-known-good stale fallback — a short TTL is cheap again: one small
+    // ~200ms GET on a single call per window, others read the cache with no
+    // network. So keep freshness tight (10s propagation) at negligible cost.
+    const POLICY_TTL_MS = 10_000;
     try {
       let dashboardPolicy = null;
       // Last-known-good cache, kept even past the TTL so we can fall back on it

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@solongate/proxy",
-  "version": "0.59.1",
+  "version": "0.59.2",
   "description": "AI tool security proxy: protect any AI tool server with customizable policies, path/command constraints, rate limiting, and audit logging. No code changes required.",
   "type": "module",
   "bin": {