Threat modeling

Threat Modeling

Identify security threats BEFORE implementation using STRIDE methodology. This is proactive security — find threats at design time when mitigations are cheap, not at review time when they're expensive.

Use this when: the feature involves authentication, user input, external APIs, payment processing, PII, multi-tenant access, or any data flow that crosses trust boundaries.

Use super-agent-skills:security-and-hardening instead when: you're implementing and need reactive security checks (input validation, output encoding, etc.).

These skills complement each other: threat-modeling identifies WHAT to protect against (design time), security-and-hardening ensures the protection is correctly IMPLEMENTED (build time).

Announce at start: "I'm using threat-modeling to identify security threats before implementation."

When to Use

• Building authentication or authorization features
• Handling user input that flows to databases, APIs, or other systems
• Integrating with external APIs or third-party services
• Processing payments or sensitive financial data
• Storing or transmitting PII (personally identifiable information)
• Building multi-tenant features (data isolation between users/orgs)
• Adding webhooks, callbacks, or any external-facing endpoints

When NOT to use:

• Static content pages with no data flow
• Internal tools with no authentication
• Pure UI styling changes
• Documentation updates

STRIDE Framework

For each component or data flow in the feature, systematically check:

Threat	Question	Examples
Spoofing	Can someone pretend to be another user/system?	Forged JWT, session hijacking, IP spoofing, stolen API keys
Tampering	Can someone modify data they shouldn't?	SQL injection, request body manipulation, unsigned webhooks, MITM attacks
Repudiation	Can someone deny performing an action?	Missing audit logs, unsigned transactions, no request logging
Information Disclosure	Can someone access data they shouldn't?	Exposed API keys, verbose error messages, directory traversal, IDOR
Denial of Service	Can someone make the system unavailable?	Unbounded queries, resource exhaustion, regex DoS, missing rate limits
Elevation of Privilege	Can someone gain permissions they shouldn't?	IDOR, missing auth checks, role confusion, JWT claim manipulation

The Process

Step 1: Scope

Define what you're threat modeling:

• Which feature or component?
• What data does it handle?
• Who are the users (and potential attackers)?

Step 2: Diagram Data Flows

Map the system's data flows, identifying trust boundaries:

┌─────────────────────────────────────────────────┐
│ TRUST BOUNDARY: Public Internet                  │
│                                                  │
│  [User Browser] ──HTTP──→ [Load Balancer]        │
│                                                  │
├─────────────────────────────────────────────────┤
│ TRUST BOUNDARY: DMZ                              │
│                                                  │
│  [Load Balancer] ──→ [API Server]                │
│                                                  │
├─────────────────────────────────────────────────┤
│ TRUST BOUNDARY: Internal Network                 │
│                                                  │
│  [API Server] ──→ [Database]                     │
│  [API Server] ──→ [External Payment API]         │
│  [API Server] ──→ [Email Service]                │
│                                                  │
└─────────────────────────────────────────────────┘

Every arrow crossing a trust boundary is a potential attack surface.

Step 3: Apply STRIDE

For each component and each data flow crossing a trust boundary, check all 6 STRIDE categories:

### Component: API Server — POST /api/tasks

| STRIDE | Threat? | Detail | Severity |
|--------|---------|--------|----------|
| Spoofing | Yes | No auth check — anyone can create tasks | High |
| Tampering | Yes | Request body not validated — SQL injection risk | Critical |
| Repudiation | No | API logs all requests with user ID | - |
| Info Disclosure | Yes | Error messages expose stack trace | Medium |
| DoS | Yes | No rate limiting on endpoint | Medium |
| Elevation | Yes | No check that user belongs to the org | High |

Step 4: Prioritize

Rate each identified threat:

Likelihood × Impact	Priority
High × High	Critical — must mitigate before shipping
High × Medium or Medium × High	High — mitigate before shipping
Medium × Medium	Medium — mitigate if feasible
Low × any or any × Low	Low — accept risk or defer

Step 5: Define Mitigations

For each High/Critical threat, define a specific countermeasure:

| Threat | Mitigation | Implementation |
|--------|-----------|----------------|
| No auth on POST /api/tasks | Add JWT validation middleware | Use existing authMiddleware from auth.ts |
| SQL injection in request body | Parameterized queries + Zod validation | Add TaskCreateSchema, apply at route level |
| Stack trace in errors | Sanitize error responses | Use existing ErrorHandler, disable stack in prod |
| No rate limiting | Add rate limiter middleware | Use express-rate-limit, 100 req/min per IP |
| Missing org membership check | Add org authorization check | Query org_members table before allowing action |

Step 6: Add to Spec

Append the threat model as a new section in the design spec:

## Threat Model

### Scope
[Feature being modeled]

### Threats Identified
[STRIDE table from Step 3]

### Mitigations
[Table from Step 5]

### Accepted Risks
[Low-priority threats that won't be mitigated, with justification]

These mitigations become acceptance criteria — the implementer must address them during build, and the reviewer must verify them during code review.

Output Format

The threat model document should contain:

1. Scope and data flow diagram
2. STRIDE analysis per component
3. Prioritized threat list
4. Mitigation plan for High/Critical threats
5. Accepted risks for Low/Medium threats

Anti-Rationalizations

Thought	Reality
"We'll add security later"	Security bolted on after implementation is 10x more expensive than security designed in.
"This feature isn't security-sensitive"	If it handles user data, auth, or external input, it's security-sensitive. Most features are.
"STRIDE is overkill"	STRIDE takes 15 minutes for a simple feature. Finding a vulnerability in production takes days.
"Our framework handles security"	Frameworks handle common cases. Your specific data flows, business logic, and integration points need specific analysis.
"We have security-and-hardening for this"	security-and-hardening catches implementation bugs. Threat modeling catches design flaws. You need both.

Red Flags

• Building auth features without a threat model
• External-facing endpoints with no STRIDE analysis
• "Accepted risks" that are actually High severity (accepting what should be mitigated)
• Threat model that only covers Spoofing and Tampering (STRIDE has 6 categories — check all of them)
• No data flow diagram — you can't identify threats you can't see

Verification

After threat modeling:

• Data flow diagram exists showing trust boundaries
• STRIDE applied to every component crossing a trust boundary
• Every High/Critical threat has a specific mitigation
• Mitigations added as acceptance criteria in the spec
• Accepted risks are documented with justification
• security-and-hardening skill referenced for implementation phase