v0.7.0 Release Notes¶

Release Date: 2026-06-01

This release introduces agentic AI threat modeling capabilities, credential flow tracking, WebSocket security configuration, and LLM-as-Judge evaluation integration.

Highlights¶

Agentic AI threat modeling with agent capabilities and execution context
Credential flow tracking for token exfiltration and replay attack modeling
WebSocket security configuration and Cross-Site WebSocket Hijacking (CSWSH) modeling
Trust modeling fields for implicit trust assumptions and breach vectors
New boundary types: container, sandbox, agent, origin
New flow types: credential, websocket, cswsh, lateral
LLM-as-Judge evaluation package with structured-evaluation integration
Windows compatibility fix for embedded rubric loading

What's New¶

Agentic AI Threat Modeling¶

Model AI agent capabilities and execution contexts:

{
  "elements": [{
    "id": "agent",
    "type": "agent",
    "agentCapabilities": {
      "toolAccess": ["file_read", "file_write", "code_execution"],
      "autonomyLevel": "high",
      "humanOversight": "minimal",
      "permissions": ["network", "filesystem"]
    },
    "executionContext": {
      "runtime": "python",
      "sandboxed": false,
      "privileges": "user",
      "networkAccess": true
    }
  }]
}

New boundary types for agent modeling:

Type	Description
`container`	Container boundary
`sandbox`	Application sandbox boundary
`agent`	AI agent execution boundary
`origin`	Browser same-origin boundary

Credential Flow Tracking¶

Track credential lifecycles through systems for token exfiltration and replay attack modeling:

{
  "credentialFlows": [{
    "id": "session-token",
    "type": "bearer",
    "source": "auth-server",
    "targets": ["api-gateway", "websocket"],
    "exfiltrationRisk": "high",
    "replayable": true
  }]
}

New flow types for attack patterns:

Type	Description
`credential`	Credential/token transmission
`websocket`	WebSocket connection
`cswsh`	Cross-Site WebSocket Hijacking
`lateral`	Lateral movement

WebSocket Security¶

Model WebSocket endpoints with security configuration:

{
  "network": {
    "webSocket": {
      "endpoint": "ws://localhost:8080/ws",
      "protocols": ["v1"],
      "originValidation": false,
      "authenticationRequired": false
    },
    "allowedOrigins": ["https://example.com"],
    "rateLimitRps": 100
  }
}

Trust Modeling¶

Model implicit trust assumptions and breach vectors:

{
  "boundaries": [{
    "id": "localhost",
    "type": "localhost",
    "implicitlyTrusted": true,
    "trustAssumption": "Localhost connections trusted without authentication",
    "trustViolationRisk": "CSWSH bypasses localhost restriction via browser",
    "authenticationRequired": false
  }]
}

Evaluation Package¶

Integration with structured-evaluation for LLM-as-Judge workflows:

import "github.com/grokify/threat-model-spec/evaluation"

// Load embedded rubrics
rubric, _ := evaluation.VulnerabilityArticleRubric()
rubric, _ := evaluation.ThreatModelRubric()
rubric, _ := evaluation.DiagramRubric()

// Convert evaluation results to reports
report := result.ToEvaluationReport("article.md")
claimsReport := result.ToClaimsReport("article.md")

New Types¶

Type	Description
`AgentCapabilities`	AI agent tool access and autonomy
`ExecutionContext`	Runtime environment and privileges
`CredentialFlow`	Token lifecycle tracking
`AttackPattern`	Reusable attack pattern templates
`WebSocketConfig`	WebSocket security configuration

Bug Fixes¶

Fixed Windows compatibility issue with embedded rubric loading (path.Join for embed.FS)

Installation¶

Go Library¶

go get github.com/grokify/threat-model-spec@v0.7.0

CLI Tool¶

go install github.com/grokify/threat-model-spec/cmd/tms@v0.7.0

Requirements¶

Go 1.25+
D2 v0.6+ (for SVG rendering)