@highflame/policy 2.0.9 → 2.1.0

package/_schemas/guardrails/templates/templates.json

@@ -0,0 +1,213 @@
+{
+  "service": "guardrails",
+  "version": "1.0.0",
+  "description": "Guardrails policy templates for LLM application security",
+  "categories": [
+    {
+      "id": "security",
+      "name": "Security",
+      "description": "Detect and block prompt injection, jailbreak attempts, and credential leakage"
+    },
+    {
+      "id": "privacy",
+      "name": "Privacy",
+      "description": "Detect and block personally identifiable information (PII) in prompts and responses"
+    },
+    {
+      "id": "trust_safety",
+      "name": "Trust & Safety",
+      "description": "Detect and block toxic, violent, hateful, sexual, or profane content"
+    },
+    {
+      "id": "agentic_security",
+      "name": "Agentic Security",
+      "description": "Detect tool abuse, data exfiltration patterns, infinite loops, and budget violations"
+    },
+    {
+      "id": "organization",
+      "name": "Organization",
+      "description": "Organization-wide baselines and default permit/deny policies"
+    }
+  ],
+  "defaults": [
+    {
+      "id": "baseline-default",
+      "name": "Baseline Permit",
+      "description": "Permits all actions by default — threat-specific forbid policies override this when threats are detected",
+      "category": "organization",
+      "file": "defaults/baseline.cedar",
+      "severity": "low",
+      "tags": ["baseline", "permit-default", "organization"],
+      "is_active": true
+    },
+    {
+      "id": "secrets-default",
+      "name": "Secrets Detection",
+      "description": "Block content containing API keys, tokens, credentials, or other secrets",
+      "category": "security",
+      "file": "defaults/secrets.cedar",
+      "severity": "critical",
+      "tags": ["secrets", "api-keys", "credentials", "data-leak"],
+      "is_active": true
+    },
+    {
+      "id": "injection-default",
+      "name": "Injection & Jailbreak Detection",
+      "description": "Block prompt injection, jailbreak attempts, and command injection using ML confidence scores",
+      "category": "security",
+      "file": "defaults/injection.cedar",
+      "severity": "high",
+      "tags": ["injection", "jailbreak", "security"],
+      "is_active": true
+    },
+    {
+      "id": "pii-default",
+      "name": "PII Detection",
+      "description": "Block content containing PII such as SSN, credit cards, or passport numbers in outputs",
+      "category": "privacy",
+      "file": "defaults/pii.cedar",
+      "severity": "high",
+      "tags": ["pii", "privacy", "data-protection"],
+      "is_active": true
+    },
+    {
+      "id": "toxicity-default",
+      "name": "Toxicity & Content Moderation",
+      "description": "Block toxic, violent, hateful, sexual, and profane content based on classifier scores",
+      "category": "trust_safety",
+      "file": "defaults/toxicity.cedar",
+      "severity": "critical",
+      "tags": ["toxicity", "trust-safety", "content-moderation"],
+      "is_active": true
+    },
+    {
+      "id": "tool-risk-default",
+      "name": "Tool Risk",
+      "description": "Block dangerous tool calls, shell execution, and sensitive tool usage based on risk scoring",
+      "category": "agentic_security",
+      "file": "defaults/tool_risk.cedar",
+      "severity": "critical",
+      "tags": ["tools", "agentic", "security"],
+      "is_active": true
+    },
+    {
+      "id": "agentic-safety-default",
+      "name": "Agentic Safety",
+      "description": "Block tool call loops, data exfiltration patterns, high-risk sequences, and budget violations",
+      "category": "agentic_security",
+      "file": "defaults/agentic_safety.cedar",
+      "severity": "high",
+      "tags": ["agentic", "safety", "loops", "exfiltration", "budget"],
+      "is_active": true
+    }
+  ],
+  "templates": [
+    {
+      "id": "mcp-tool-permissions",
+      "name": "MCP Tool Permissions",
+      "description": "Per-tool access control for MCP servers — allow specific tools while denying others, exclude servers org-wide, block unverified sources",
+      "category": "agentic_security",
+      "file": "mcp_tool_permissions.cedar",
+      "severity": "high",
+      "tags": ["mcp", "tools", "least-privilege", "per-server", "exclusion"]
+    },
+    {
+      "id": "chat-assistant-security",
+      "name": "Chat Assistant — Security",
+      "description": "Aggressive injection and jailbreak defense for customer-facing chatbots with lower thresholds",
+      "category": "security",
+      "file": "profiles/chat_assistant/security.cedar",
+      "severity": "high",
+      "tags": ["profile", "chat-assistant", "injection", "jailbreak", "security"]
+    },
+    {
+      "id": "chat-assistant-privacy",
+      "name": "Chat Assistant — Privacy",
+      "description": "Block PII in both user inputs and assistant outputs for chat applications",
+      "category": "privacy",
+      "file": "profiles/chat_assistant/privacy.cedar",
+      "severity": "high",
+      "tags": ["profile", "chat-assistant", "pii", "privacy"]
+    },
+    {
+      "id": "chat-assistant-trust-safety",
+      "name": "Chat Assistant — Trust & Safety",
+      "description": "Strict content moderation with lower toxicity thresholds and topic restrictions for public-facing chat",
+      "category": "trust_safety",
+      "file": "profiles/chat_assistant/trust_safety.cedar",
+      "severity": "critical",
+      "tags": ["profile", "chat-assistant", "toxicity", "trust-safety", "topics"]
+    },
+    {
+      "id": "code-agent-agentic-security",
+      "name": "Code Agent — Agentic Security",
+      "description": "Tool risk controls, shell blocking, loop detection, exfiltration prevention, and budget enforcement for coding assistants",
+      "category": "agentic_security",
+      "file": "profiles/code_agent/agentic_security.cedar",
+      "severity": "high",
+      "tags": ["profile", "code-agent", "tools", "agentic", "exfiltration", "budget"]
+    },
+    {
+      "id": "code-agent-security",
+      "name": "Code Agent — Security",
+      "description": "Prevent code agents from writing detected secrets to output files",
+      "category": "security",
+      "file": "profiles/code_agent/security.cedar",
+      "severity": "critical",
+      "tags": ["profile", "code-agent", "secrets", "security"]
+    },
+    {
+      "id": "data-pipeline-privacy",
+      "name": "Data Pipeline — Privacy",
+      "description": "Strict PII protection with zero-tolerance for sensitive PII types in data pipelines",
+      "category": "privacy",
+      "file": "profiles/data_pipeline/privacy.cedar",
+      "severity": "critical",
+      "tags": ["profile", "data-pipeline", "pii", "privacy", "compliance"]
+    },
+    {
+      "id": "data-pipeline-security",
+      "name": "Data Pipeline — Security",
+      "description": "Strict secrets detection and lower injection thresholds for RAG and data processing pipelines",
+      "category": "security",
+      "file": "profiles/data_pipeline/security.cedar",
+      "severity": "critical",
+      "tags": ["profile", "data-pipeline", "secrets", "injection", "security"]
+    },
+    {
+      "id": "data-pipeline-agentic-security",
+      "name": "Data Pipeline — Agentic Security",
+      "description": "Exfiltration prevention and tool risk controls for data processing pipelines",
+      "category": "agentic_security",
+      "file": "profiles/data_pipeline/agentic_security.cedar",
+      "severity": "critical",
+      "tags": ["profile", "data-pipeline", "exfiltration", "tools"]
+    }
+  ],
+  "profiles": [
+    {
+      "id": "chat-assistant",
+      "name": "Chat Assistant",
+      "description": "Optimized for customer-facing chatbots — strict toxicity, PII blocking, aggressive injection defense, topic restrictions",
+      "severity": "high",
+      "tags": ["chat-assistant", "toxicity", "pii", "injection"],
+      "template_ids": ["chat-assistant-security", "chat-assistant-privacy", "chat-assistant-trust-safety"]
+    },
+    {
+      "id": "code-agent",
+      "name": "Code Agent",
+      "description": "Optimized for coding assistants — tool risk controls, shell blocking, loop detection, exfiltration prevention, budget enforcement",
+      "severity": "high",
+      "tags": ["code-agent", "tools", "agentic", "exfiltration"],
+      "template_ids": ["code-agent-agentic-security", "code-agent-security"]
+    },
+    {
+      "id": "data-pipeline",
+      "name": "Data Pipeline",
+      "description": "Optimized for RAG and data processing — strict PII/secrets protection, exfiltration detection, pipeline injection defense",
+      "severity": "critical",
+      "tags": ["data-pipeline", "pii", "secrets", "exfiltration"],
+      "template_ids": ["data-pipeline-privacy", "data-pipeline-security", "data-pipeline-agentic-security"]
+    }
+  ]
+}

package/_schemas/overwatch/context.json

@@ -100,61 +100,61 @@
         {
           "key": "violence_score",
           "type": "number",
-          "required": false,
+          "required": true,
           "description": "Violence content detection score (0-100)"
         },
         {
           "key": "weapons_score",
           "type": "number",
-          "required": false,
+          "required": true,
           "description": "Weapons content detection score (0-100)"
         },
         {
           "key": "hate_speech_score",
           "type": "number",
-          "required": false,
+          "required": true,
           "description": "Hate speech detection score (0-100)"
         },
         {
           "key": "crime_score",
           "type": "number",
-          "required": false,
+          "required": true,
           "description": "Criminal content detection score (0-100)"
         },
         {
           "key": "sexual_score",
           "type": "number",
-          "required": false,
+          "required": true,
           "description": "Sexual content detection score (0-100)"
         },
         {
           "key": "profanity_score",
           "type": "number",
-          "required": false,
+          "required": true,
           "description": "Profanity detection score (0-100)"
         },
         {
           "key": "pii_confidence",
           "type": "number",
-          "required": false,
+          "required": true,
           "description": "PII detection classifier confidence (0-100)"
         },
         {
           "key": "injection_confidence",
           "type": "number",
-          "required": false,
+          "required": true,
           "description": "Prompt injection classifier confidence (0-100)"
         },
         {
           "key": "jailbreak_confidence",
           "type": "number",
-          "required": false,
+          "required": true,
           "description": "Jailbreak detection classifier confidence (0-100)"
         },
         {
           "key": "indirect_injection_score",
           "type": "number",
-          "required": false,
+          "required": true,
           "description": "Indirect prompt injection risk score (0-100)"
         }
       ]
@@ -226,44 +226,44 @@
         {
           "key": "threat_count",
           "type": "number",
-          "required": true,
-          "description": "Total threats detected"
+          "required": false,
+          "description": "Total threats detected (if scanning ran)"
         },
         {
           "key": "highest_severity",
           "type": "string",
-          "required": true,
-          "description": "Highest severity: critical, high, medium, low"
+          "required": false,
+          "description": "Highest severity (if scanning ran)"
         },
         {
           "key": "threat_categories",
           "type": "array",
-          "required": true,
-          "description": "Threat category names"
+          "required": false,
+          "description": "Threat category names (if scanning ran)"
         },
         {
           "key": "threat_types",
           "type": "array",
-          "required": true,
-          "description": "YARA threat categories"
+          "required": false,
+          "description": "YARA threat categories (if scanning ran)"
         },
         {
           "key": "yara_threats",
           "type": "array",
-          "required": true,
-          "description": "YARA rule names"
+          "required": false,
+          "description": "YARA rule names (if scanning ran)"
         },
         {
           "key": "max_threat_severity",
           "type": "number",
-          "required": true,
-          "description": "Numeric severity (0-4)"
+          "required": false,
+          "description": "Numeric severity 0-4 (if scanning ran)"
         },
         {
           "key": "contains_secrets",
           "type": "boolean",
-          "required": true,
-          "description": "Whether secrets detected"
+          "required": false,
+          "description": "Whether secrets detected (if scanning ran)"
         },
         {
           "key": "response_content",
@@ -358,8 +358,8 @@
         {
           "key": "content",
           "type": "string",
-          "required": true,
-          "description": "Raw content being scanned"
+          "required": false,
+          "description": "Raw content being scanned (if available)"
         },
         {
           "key": "source",
@@ -388,26 +388,26 @@
         {
           "key": "threat_count",
           "type": "number",
-          "required": true,
-          "description": "Total threats detected"
+          "required": false,
+          "description": "Total threats detected (if scanning ran)"
         },
         {
           "key": "highest_severity",
           "type": "string",
-          "required": true,
-          "description": "Highest severity level"
+          "required": false,
+          "description": "Highest severity level (if scanning ran)"
         },
         {
           "key": "threat_categories",
           "type": "array",
-          "required": true,
-          "description": "Threat category names"
+          "required": false,
+          "description": "Threat category names (if scanning ran)"
         },
         {
           "key": "max_threat_severity",
           "type": "number",
-          "required": true,
-          "description": "Numeric severity (0-4)"
+          "required": false,
+          "description": "Numeric severity 0-4 (if scanning ran)"
         },
         {
           "key": "tool_poisoning_score",
@@ -484,32 +484,32 @@
         {
           "key": "threat_count",
           "type": "number",
-          "required": true,
-          "description": "Total threats detected"
+          "required": false,
+          "description": "Total threats detected (if scanning ran)"
         },
         {
           "key": "highest_severity",
           "type": "string",
-          "required": true,
-          "description": "Highest severity level"
+          "required": false,
+          "description": "Highest severity level (if scanning ran)"
         },
         {
           "key": "threat_categories",
           "type": "array",
-          "required": true,
-          "description": "Threat categories"
+          "required": false,
+          "description": "Threat categories (if scanning ran)"
         },
         {
           "key": "max_threat_severity",
           "type": "number",
-          "required": true,
-          "description": "Numeric severity (0-4)"
+          "required": false,
+          "description": "Numeric severity 0-4 (if scanning ran)"
         },
         {
           "key": "contains_secrets",
           "type": "boolean",
-          "required": true,
-          "description": "Whether secrets detected"
+          "required": false,
+          "description": "Whether secrets detected (if scanning ran)"
         }
       ]
     },
@@ -562,32 +562,32 @@
         {
           "key": "threat_count",
           "type": "number",
-          "required": true,
-          "description": "Total threats detected"
+          "required": false,
+          "description": "Total threats detected (if scanning ran)"
         },
         {
           "key": "highest_severity",
           "type": "string",
-          "required": true,
-          "description": "Highest severity level"
+          "required": false,
+          "description": "Highest severity level (if scanning ran)"
         },
         {
           "key": "threat_categories",
           "type": "array",
-          "required": true,
-          "description": "Threat categories"
+          "required": false,
+          "description": "Threat categories (if scanning ran)"
         },
         {
           "key": "max_threat_severity",
           "type": "number",
-          "required": true,
-          "description": "Numeric severity (0-4)"
+          "required": false,
+          "description": "Numeric severity 0-4 (if scanning ran)"
         },
         {
           "key": "contains_secrets",
           "type": "boolean",
-          "required": true,
-          "description": "Whether secrets detected"
+          "required": false,
+          "description": "Whether secrets detected (if scanning ran)"
         }
       ]
     }