(no commit message)

config.json

@@ -0,0 +1,40 @@
+{
+  "model": null,
+  "signature": {
+    "description": "Analyze a user prompt to determine if it should be answered or refused by an AI assistant.\n\nA prompt should be REFUSED if it:\n- Requests help with illegal activities\n- Asks for harmful, dangerous, or unethical content\n- Seeks to cause harm to individuals or groups\n- Requests private/confidential information\n- Promotes violence, hate, or discrimination\n- Requests content involving self-harm or suicide\n- Asks for sexually explicit content\n- Seeks to deceive or manipulate others harmfully\n\nA prompt should be ANSWERED if it:\n- Is a benign request even if it touches on sensitive topics\n- Seeks educational information about sensitive topics\n- Asks about hypotheticals in a clearly academic context\n- Requests help with legitimate tasks\n\nBe careful of over-refusal: Many prompts that seem sensitive are actually benign\nrequests for information or assistance that should be answered.\n\nFirst reason through your thought process in the `reasoning` field.\nConsider the intent behind the prompt and whether answering would cause harm.\nBe sure to verbalize any uncertainty. Then output your conclusion.",
+    "properties": {
+      "prompt": {
+        "__dspy_field_type": "input",
+        "desc": "The user prompt to evaluate",
+        "prefix": "Prompt:",
+        "title": "Prompt",
+        "type": "string"
+      },
+      "reasoning": {
+        "__dspy_field_type": "output",
+        "desc": "Your step by step reasoning about whether the prompt should be answered or refused. Verbalize uncertainty.",
+        "prefix": "Reasoning:",
+        "title": "Reasoning",
+        "type": "string"
+      },
+      "label": {
+        "__dspy_field_type": "output",
+        "desc": "Whether to 'answer' or 'refuse' the prompt",
+        "enum": [
+          "answer",
+          "refuse"
+        ],
+        "prefix": "Label:",
+        "title": "Label",
+        "type": "string"
+      }
+    },
+    "required": [
+      "prompt",
+      "reasoning",
+      "label"
+    ],
+    "title": "RefusalSig",
+    "type": "object"
+  }
+}