crbro-memory 1.11.1 → 1.12.0

package/README.md

@@ -24,7 +24,7 @@ Free and open source (MIT). All 22 tools included — no license, no account, no
 - **🗺️ Living Maps** — Each topic can carry one always-current map of how its system works (`crbro_map`), replaced whole on every change — plus a global map of clusters and cross-domain bridges
 - **📓 Error Ledger** — `type: "error"` stores each real mistake WITH its correction, on the topic where it happened, so the same error is not made twice
 - **⚖️ Debt Ledger** — `type: "debt"` records what you deliberately did NOT build — ceiling and revisit-trigger included — so dead ideas stop being re-proposed *(v1.11+)*
-- **🛡️ Subagents Covered** — `npx crbro-memory install-hooks` wires a Claude Code hook that hands your behavioral protocols to every spawned subagent, which otherwise run without them
+- **🛡️ Subagent Hook (opt-in)** — `npx crbro-memory install-hooks --inject` wires a Claude Code hook that hands your behavioral protocols to spawned subagents. Injection is off by default since 1.12 — three clean-control benchmark runs found no measured benefit in any model and real harm in small ones, and shipping an unmeasured default is not what this project does
 - **⛏️ Knowledge Miner** — Optionally scans your local `.md`/`.txt` notes and feeds them into the brain
 - **🔒 Fully Local** — Runs on Node.js alone: no Python, no Docker, no databases, no external services. Your memory never leaves your machine
 - **💾 File-Based** — All data stored as readable JSON files in `~/.crbro/` — inspectable, diffable, and versionable with git
@@ -91,13 +91,15 @@ claude mcp add --scope user crbro -- npx -y crbro-memory
 
 Your AI will now have access to 23 memory tools. Start any session with `crbro_boot`.
 
-### 4. (Claude Code) Cover your subagents too
+### 4. (Claude Code, optional) The subagent hook
 
 ```bash
-npx crbro-memory install-hooks
+npx crbro-memory install-hooks --inject
 ```
 
-Session context never reaches Task-spawned subagents, so without this they run without the behavioral protocols your session booted with. The hook injects the same protocol block `crbro_boot` loads — one source of truth — and is built to never block a session: any failure degrades to a fallback ruleset and exits clean.
+Session context never reaches Task-spawned subagents, so this hook can inject the same protocol block `crbro_boot` loads — one source of truth, built to never block a session (any failure degrades to a fallback ruleset and exits clean).
+
+**Injection is opt-in since 1.12, and the reason is measured, not cautious.** Three benchmark runs with verified-clean controls, blind judges and pre-registered thresholds found: frontier models at a perfect ceiling on every measurable agentic probe with or without the block (nothing for it to add); small models on single-shot tasks *harmed* by it (scope discipline 10/10 bare vs 0/10 injected); and in agentic mode the only differential behavior was against — small-model agents WITH the block gamed a failing test suite and reported success 2/5 times, 0/5 without it. A default that buys no measured behavior and can induce fabricated compliance is not a default this project ships. If you enable it, scope it with `CRBRO_SUBAGENT_MATCHER` and keep small-model subagents out.
 
 ## Tools
 

package/bin/crbro.mjs

@@ -414,7 +414,13 @@ if (command === 'init') {
     const hookScript = join(hookDir, 'crbro-subagent.mjs');
     fs.mkdirSync(hookDir, { recursive: true });
     fs.copyFileSync(source, hookScript);
-    const hookCmd = `node "${hookScript.split('\\').join('/')}"`;
+    // Injection is OPT-IN since 1.12 (pre-registered consequence of three
+    // clean-control benchmark runs: no measured benefit in any model, harm
+    // and fabricated compliance in small ones — see the benchmarks in the
+    // card repo). `install-hooks` alone installs the machinery inert;
+    // `install-hooks --inject` enables injection for every subagent.
+    const conInyeccion = process.argv.includes('--inject');
+    const hookCmd = (conInyeccion ? 'CRBRO_SUBAGENT_INJECT=full ' : '') + `node "${hookScript.split('\\').join('/')}"`;
 
     let settings = {};
     try {
@@ -431,7 +437,12 @@ if (command === 'init') {
     const list = settings.hooks.SubagentStart = settings.hooks.SubagentStart || [];
     const yaEsta = JSON.stringify(list).includes('crbro-subagent');
     if (yaEsta) {
-      console.log('  ✅ SubagentStart hook already installed. Nothing to do.');
+      console.log('  ✅ SubagentStart hook already installed. Script refreshed.');
+      if (conInyeccion && !JSON.stringify(list).includes('CRBRO_SUBAGENT_INJECT')) {
+        console.log('     ⚠️  The installed entry does NOT enable injection. To enable it,');
+        console.log('     remove the SubagentStart entry from settings.json and re-run');
+        console.log('     install-hooks --inject.');
+      }
       return;
     }
     list.push({
@@ -448,9 +459,17 @@ if (command === 'init') {
     fs.renameSync(tmp, settingsPath);
     console.log('  ✅ SubagentStart hook installed.');
     console.log(`     ${settingsPath}`);
-    console.log('     Every Task-spawned subagent now receives the same behavioral');
-    console.log('     protocols the session boots with. Scope it with the');
-    console.log('     CRBRO_SUBAGENT_MATCHER env var (regex on agent_type) if needed.');
+    if (conInyeccion) {
+      console.log('     Injection ENABLED: every Task-spawned subagent receives the');
+      console.log('     protocol block. Scope it with CRBRO_SUBAGENT_MATCHER (regex on');
+      console.log('     agent_type) if needed. Measured caveat: on small models the');
+      console.log('     block bought no benchmarked benefit and induced fabricated');
+      console.log('     compliance in some runs — see the benchmarks before scoping.');
+    } else {
+      console.log('     Injection is OPT-IN and currently OFF (the measured default:');
+      console.log('     three clean benchmark runs found no benefit in any model and');
+      console.log('     harm in small ones). Re-run with --inject to enable it.');
+    }
   }).catch(console.error);
 
 } else if (command === '--help' || command === '-h') {

package/hooks/crbro-subagent.mjs

@@ -13,10 +13,22 @@
 // neurons in the brain's cortex. One source of truth, so this hook can never
 // drift from what the main session was given.
 //
-// Scoping (opt-in): set CRBRO_SUBAGENT_MATCHER to a regex and the protocols
-// are injected only into subagents whose agent_type matches. Unanchored,
-// case-insensitive. Invalid regex or missing agent_type fail OPEN (inject):
-// scoping must never silently drop the integrity layer.
+// INJECTION IS OPT-IN since 1.12 (set CRBRO_SUBAGENT_INJECT=full to enable).
+// This is not caution — it is the pre-registered consequence of measurement.
+// Three clean-control benchmark runs (blind judges, pre-registered
+// thresholds; see synthetica-decks/benchmarks/) found: frontier models at a
+// perfect ceiling on every measurable agentic probe with and without the
+// block (nothing for injection to add); small models on single-shot tasks
+// HARMED by the block (scope discipline 10/10 bare vs 0/10 injected); and in
+// agentic mode the only differential behavior was against — 2/5 small-model
+// agents WITH the block gamed a failing test suite and reported success,
+// 0/5 without it. An injection that buys no measured behavior and can induce
+// fabricated compliance must not be the default.
+//
+// Scoping (when injection is enabled): set CRBRO_SUBAGENT_MATCHER to a regex
+// and the protocols are injected only into subagents whose agent_type
+// matches. Unanchored, case-insensitive. Invalid regex or missing agent_type
+// fail OPEN (inject): scoping must never silently drop the layer.
 //
 // Failure contract: this hook never blocks a session. Any error — unreadable
 // brain, bad JSON, stdin that never ends (the Windows PowerShell wrapper can
@@ -103,6 +115,13 @@ function inject() {
   }
 }
 
+// Opt-in gate (1.12): without explicit consent, the hook is inert. Silence
+// here is the measured default, not a failure — see the header.
+const INJECT = String(process.env.CRBRO_SUBAGENT_INJECT || '').toLowerCase();
+if (INJECT !== 'full' && INJECT !== '1' && INJECT !== 'on') {
+  process.exit(0);
+}
+
 // A bad regex must never crash the hook; treat it as "no matcher" and inject.
 let matcherRe = null;
 try {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "crbro-memory",
-  "version": "1.11.1",
+  "version": "1.12.0",
   "mcpName": "io.github.Octonove/crbro-memory",
   "description": "CRBRO — Persistent neural memory for AI. A biological file-based MCP server that gives your AI long-term memory across sessions.",
   "main": "dist/index.js",