@ouro.bot/cli 0.1.0-alpha.296 → 0.1.0-alpha.297

package/changelog.json

@@ -1,6 +1,12 @@
 {
   "_note": "This changelog is maintained as part of the PR/version-bump workflow. Agent-curated, not auto-generated. Agents read this file directly via read_file to understand what changed between versions.",
   "versions": [
+    {
+      "version": "0.1.0-alpha.297",
+      "changes": [
+        "feat(mind): add integrity scanning for diary entries at write time. New detectSuspiciousContent() in src/mind/diary-integrity.ts checks entry text for prompt injection patterns (instruction framing, override language, role injection, boundary markers) before persisting. Detection emits a mind.diary_integrity_warning nerves event with matched pattern categories, text preview, and entry ID. Entries still save (flag, don't block) to avoid false-positive data loss. 41 unit tests + 4 integration tests with 100% coverage on new code."
+      ]
+    },
     {
       "version": "0.1.0-alpha.296",
       "changes": [

package/dist/mind/diary-integrity.js

@@ -0,0 +1,60 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.detectSuspiciousContent = detectSuspiciousContent;
+const PATTERN_CATEGORIES = [
+    {
+        name: "instruction_framing",
+        patterns: [
+            /\byou are (?:a|an) (?:ai|assistant|language model|helpful assistant)\b/i,
+            /\byour (?:new )?instructions are\b/i,
+            /\bsystem\s*:/i,
+            /\bignore (?:all |my )?previous instructions\b/i,
+            /\bdo not reveal\b/i,
+        ],
+    },
+    {
+        name: "override_language",
+        patterns: [
+            /\bdisregard\b/i,
+            /\bforget everything\b/i,
+            /\bnew instructions:/i,
+            /\boverride (?:all |any |previous )?instructions\b/i,
+        ],
+    },
+    {
+        name: "role_injection",
+        patterns: [
+            /\bas (?:a|an) (?:ai|language model)\b/i,
+            /\byou must always\b/i,
+            /\byou are now\b/i,
+        ],
+    },
+    {
+        name: "boundary_markers",
+        patterns: [
+            /```system/i,
+            /<<SYS>>/i,
+            /\[INST\]/i,
+            /<\/?system>/i,
+            /\[system\]/i,
+        ],
+    },
+];
+function detectSuspiciousContent(text) {
+    if (!text) {
+        return { suspicious: false, patterns: [] };
+    }
+    const matched = new Set();
+    for (const category of PATTERN_CATEGORIES) {
+        for (const pattern of category.patterns) {
+            if (pattern.test(text)) {
+                matched.add(category.name);
+                break;
+            }
+        }
+    }
+    return {
+        suspicious: matched.size > 0,
+        patterns: [...matched],
+    };
+}

package/dist/mind/diary.js

@@ -46,6 +46,7 @@ const crypto_1 = require("crypto");
 const identity_1 = require("../heart/identity");
 const runtime_1 = require("../nerves/runtime");
 const associative_recall_1 = require("./associative-recall");
+const diary_integrity_1 = require("./diary-integrity");
 const embedding_provider_1 = require("./embedding-provider");
 const DEDUP_THRESHOLD = 0.6;
 const SEMANTIC_DEDUP_THRESHOLD = 0.95;
@@ -237,6 +238,20 @@ async function saveDiaryEntry(options) {
         embedding,
         ...(options.provenance ? { provenance: options.provenance } : {}),
     };
+    const integrity = (0, diary_integrity_1.detectSuspiciousContent)(text);
+    if (integrity.suspicious) {
+        (0, runtime_1.emitNervesEvent)({
+            level: "warn",
+            component: "mind",
+            event: "mind.diary_integrity_warning",
+            message: "suspicious content detected in diary entry",
+            meta: {
+                patterns: integrity.patterns,
+                textPreview: text.slice(0, 200),
+                entryId: fact.id,
+            },
+        });
+    }
     return appendEntriesWithDedup(stores, [fact], { semanticThreshold: SEMANTIC_DEDUP_THRESHOLD });
 }
 async function backfillEmbeddings(options) {

package/dist/nerves/coverage/file-completeness.js

@@ -69,6 +69,9 @@ const DISPATCH_EXEMPT_PATTERNS = [
     "mind/embedding-provider",
     // Commerce utility module: error classes and pure helpers (no independent side effects).
     "repertoire/commerce-errors",
+    // Diary integrity: pure detection utility (pattern matching only). The caller
+    // (diary.ts saveDiaryEntry) owns observability via mind.diary_integrity_warning.
+    "mind/diary-integrity",
     // Log redaction: pure utility consumed by the NDJSON sink (no independent side effects).
     "nerves/redact",
     // Bundle templates: pure constants (gitignore template string, PII

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ouro.bot/cli",
-  "version": "0.1.0-alpha.296",
+  "version": "0.1.0-alpha.297",
   "main": "dist/heart/daemon/ouro-entry.js",
   "bin": {
     "cli": "dist/heart/daemon/ouro-bot-entry.js",