Repository navigation
TMKB v0.1.0 — Initial Release
The Threat Model Knowledge Base (TMKB) is a security context source that agents, tools, and humans can consume—starting with threat patterns for authorization enforcement in multi-tenant applications.
Highlights
- 12 authorization threat patterns covering the architectural security gaps that LLMs systematically miss
- CLI tool with sub-second query response (~120ms)
- MCP server for seamless integration with Claude Code and other AI coding assistants
- Validated: 6 independent baseline tests across 3 providers (Anthropic, OpenAI, Google) confirm 100% async boundary failure rate without TMKB; 0% with TMKB
Pattern Library
Tier A — Full Depth (5 patterns)
| ID | Pattern | Severity |
|---|---|---|
| TMKB-AUTHZ-001 | Background Job Authorization Context Loss | High |
| TMKB-AUTHZ-002 | List/Detail Authorization Inconsistency | High |
| TMKB-AUTHZ-003 | Soft-Delete Resurrection Attack | High |
| TMKB-AUTHZ-004 | Tenant Isolation via Application Logic | Critical |
| TMKB-AUTHZ-005 | User-Account-Resource Ownership Confusion | High |
Tier B — Essential Coverage (7 patterns)
| ID | Pattern | Severity |
|---|---|---|
| TMKB-AUTHZ-006 | Mass Assignment of Ownership Fields | High |
| TMKB-AUTHZ-007 | Insecure Direct Object Reference via Sequential IDs | High |
| TMKB-AUTHZ-008 | Authorization Bypass via HTTP Method Override | Medium |
| TMKB-AUTHZ-009 | State Transition Authorization Bypass | High |
| TMKB-AUTHZ-010 | Unauthorized Access via Relationship Traversal | High |
| TMKB-AUTHZ-011 | Authorization Check in Wrong Layer | High |
| TMKB-AUTHZ-012 | Inconsistent Authorization in Bulk Operations | High |
Each pattern includes: threat description, attack scenarios, mitigations with code examples (Python/Flask), testing guidance, and validation data.
CLI Tool
# Query by context
./tmkb query --context "background job processing"
# Filter by language/framework
./tmkb query --context "file upload" --language python --framework flask
# List all patterns
./tmkb list
# Get specific pattern
./tmkb get TMKB-AUTHZ-001
# Validate pattern files
./tmkb validate --allMCP Server Integration
Add to ~/.claude/mcp_settings.json:
{
"mcpServers": {
"tmkb": {
"command": "/path/to/tmkb",
"args": ["serve"]
}
}
}See the MCP Integration Guide and Claude Code Integration Example for details.
Validation Results
| Invariant | Baseline (6 runs) | Enhanced (with TMKB) |
|---|---|---|
| Auth on mutating endpoints | 5/6 pass | Pass |
| Object ownership validated | 4/6 pass | Pass |
| List/detail consistency | 5/6 pass | Pass |
| Async boundary re-auth | 0/6 pass | Pass |
The async boundary invariant failed 100% across Claude Sonnet 4.5, Claude Opus 4.6, GPT-5.2, and Gemini—confirming this is a systematic, provider-invariant LLM blindspot that TMKB addresses.
Installation
Download Binary
Download the appropriate binary for your platform from the assets below, then:
chmod +x tmkb-*
./tmkb-linux-amd64 list # or your platform's binaryBuild from Source
git clone https://github.com/mark-chris/tmkb.git
cd tmkb
go build -o bin/tmkb ./cmd/tmkbRequires Go 1.25+.
What's Next
- Additional pattern categories beyond authorization
- Support for more languages and frameworks
- Enhanced validation with broader prompt coverage
- Community pattern contributions