@framers/agentos-ext-ml-classifiers 0.1.0 → 0.2.1

package/dist/types.js

@@ -1,62 +1,16 @@
 /**
- * @fileoverview Core type definitions for the ML Classifier Guardrail Extension Pack.
+ * @file types.ts
+ * @description Core type definitions for the ML Classifiers extension pack.
  *
- * This file defines all configuration shapes, runtime result types, and
- * service-identifier constants used by the ML classifier pipeline. All
- * classifiers in this pack evaluate text content against learned models
- * (toxicity, prompt-injection, jailbreak) and emit structured results that
- * feed into the AgentOS guardrail decision tree.
+ * Defines the shared interfaces used across the ML classification system:
+ * classifier categories, confidence results, option shapes, and the LLM
+ * invoker callback signature.
  *
- * Import hierarchy
- * ----------------
- * ```
- * IUtilityAI  ──── ClassificationResult, ClassificationScore
- * IGuardrailService ── GuardrailAction
- *                   │
- *                   ▼
- *              types.ts  (this file)
- *                   │
- *                   ▼
- *           IContentClassifier.ts  /  SlidingWindowBuffer.ts  /  …
- * ```
- *
- * @module agentos/extensions/packs/ml-classifiers/types
- */
-/**
- * Sensible defaults for {@link ClassifierThresholds}.
- *
- * These values reflect a conservative-but-pragmatic policy:
- * - block at 90 % confidence → very high bar, minimises false positives
- * - flag at 70 % → surfaced for human review, not blocked
- * - warn at 40 % → low-confidence signal, handled with a light touch
+ * @module ml-classifiers/types
  */
-export const DEFAULT_THRESHOLDS = {
-    blockThreshold: 0.9,
-    flagThreshold: 0.7,
-    warnThreshold: 0.4,
-};
-// ---------------------------------------------------------------------------
-// Service identifiers
-// ---------------------------------------------------------------------------
 /**
- * Well-known service identifier strings for the three built-in ML classifier
- * pipelines.
- *
- * These IDs follow the `agentos:<domain>:<name>` naming convention used
- * throughout the AgentOS extension ecosystem.  Use them to retrieve specific
- * classifier services from the shared service registry.
- *
- * @example
- * ```typescript
- * const toxicity = serviceRegistry.get(ML_CLASSIFIER_SERVICE_IDS.TOXICITY_PIPELINE);
- * ```
+ * All supported classifier categories as a constant array, used for
+ * iteration and default configuration.
  */
-export const ML_CLASSIFIER_SERVICE_IDS = {
-    /** Classifier that detects toxic, hateful, or abusive language. */
-    TOXICITY_PIPELINE: 'agentos:ml-classifiers:toxicity-pipeline',
-    /** Classifier that detects prompt-injection attempts. */
-    INJECTION_PIPELINE: 'agentos:ml-classifiers:injection-pipeline',
-    /** Classifier that detects jailbreak / system-override attempts. */
-    JAILBREAK_PIPELINE: 'agentos:ml-classifiers:jailbreak-pipeline',
-};
+export const ALL_CATEGORIES = ['toxic', 'injection', 'nsfw', 'threat'];
 //# sourceMappingURL=types.js.map

package/dist/types.js.map

@@ -1 +1 @@
-{"version":3,"file":"types.js","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AA2CH;;;;;;;GAOG;AACH,MAAM,CAAC,MAAM,kBAAkB,GAAyB;IACtD,cAAc,EAAE,GAAG;IACnB,aAAa,EAAE,GAAG;IAClB,aAAa,EAAE,GAAG;CACV,CAAC;AA6NX,8EAA8E;AAC9E,sBAAsB;AACtB,8EAA8E;AAE9E;;;;;;;;;;;;GAYG;AACH,MAAM,CAAC,MAAM,yBAAyB,GAAG;IACvC,mEAAmE;IACnE,iBAAiB,EAAE,0CAA0C;IAE7D,yDAAyD;IACzD,kBAAkB,EAAE,2CAA2C;IAE/D,oEAAoE;IACpE,kBAAkB,EAAE,2CAA2C;CACvD,CAAC"}
+{"version":3,"file":"types.js","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAgBH;;;GAGG;AACH,MAAM,CAAC,MAAM,cAAc,GAAyB,CAAC,OAAO,EAAE,WAAW,EAAE,MAAM,EAAE,QAAQ,CAAC,CAAC"}

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@framers/agentos-ext-ml-classifiers",
-  "version": "0.1.0",
-  "description": "ML-based content classification guardrail (toxicity, injection, jailbreak) for AgentOS",
+  "version": "0.2.1",
+  "description": "ML-based content classifiers for AgentOS — toxicity, prompt injection, and NSFW detection via ONNX models or LLM fallback",
   "type": "module",
   "main": "./dist/index.js",
   "types": "./dist/index.d.ts",
@@ -11,23 +11,12 @@
       "types": "./dist/index.d.ts"
     }
   },
-  "files": [
-    "dist",
-    "src",
-    "SKILL.md",
-    "manifest.json"
-  ],
   "peerDependencies": {
-    "@framers/agentos": "^0.1.0"
+    "@framers/agentos": ">=0.1.0"
   },
   "optionalDependencies": {
     "@huggingface/transformers": "^3.0.0"
   },
-  "devDependencies": {
-    "typescript": "^5.5.0",
-    "vitest": "^1.6.0",
-    "@framers/agentos": "0.1.47"
-  },
   "license": "MIT",
   "author": "Frame.dev",
   "repository": {
@@ -38,8 +27,13 @@
   "publishConfig": {
     "access": "public"
   },
+  "devDependencies": {
+    "semantic-release": "^24.0.0",
+    "@semantic-release/github": "^11.0.0"
+  },
   "scripts": {
-    "build": "tsc -p tsconfig.json",
-    "test": "vitest run"
+    "build": "tsc",
+    "clean": "rm -rf dist",
+    "typecheck": "tsc --noEmit"
   }
 }