fai-threat-model — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited fai-threat-model (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Model threats for AI applications with STRIDE and trust boundary analysis.
| Threat | AI Example | Mitigation |
|---|---|---|
| Spoofing | Fake user identity to access AI | OAuth2/MI auth, no anonymous access |
| Tampering | Modify training data or prompts | Input validation, prompt integrity |
| Repudiation | Deny AI-generated actions | Audit logging, immutable trails |
| Information Disclosure | PII leaked via LLM output | PII filtering, output validation |
| Denial of Service | Token exhaustion attack | Rate limiting, token budgets |
| Elevation of Privilege | Prompt injection to bypass controls | Input sanitization, guardrails |
┌─────────────────────────────────────────────┐
│ UNTRUSTED: Internet │
│ [User Browser] ──HTTPS──→ [WAF/CDN] │
└────────────────────────────┬────────────────┘
│
┌────────────────────────────▼────────────────┐
│ DMZ: API Gateway (APIM) │
│ - Rate limiting │
│ - Auth validation │
│ - Input sanitization │
└────────────────────────────┬────────────────┘
│ MI Auth
┌────────────────────────────▼────────────────┐
│ TRUSTED: Application VNet │
│ [API] → [Azure OpenAI] (Private Endpoint) │
│ [API] → [AI Search] (Private Endpoint) │
│ [API] → [Cosmos DB] (Private Endpoint) │
│ [API] → [Key Vault] (Private Endpoint) │
└─────────────────────────────────────────────┘| Threat | Description | Mitigation |
|---|---|---|
| Prompt injection | User manipulates system behavior | Input sanitization, system prompt protection |
| Data exfiltration | LLM reveals training data | Content filtering, output validation |
| Model abuse | Excessive API calls for non-intended use | Rate limiting, usage monitoring |
| Supply chain | Compromised model or package | Pin versions, verify checksums |
| Jailbreak | Bypassing safety guardrails | Multi-layer content safety, red teaming |
| # | Threat | Category | Likelihood | Impact | Risk | Mitigation | Status |
|---|--------|----------|-----------|--------|------|-----------|--------|
| 1 | Prompt injection | Elevation | High | High | Critical | Input sanitization + guardrails | In progress |
| 2 | PII in responses | Info Disc. | Medium | High | High | Output filtering | Done |
| 3 | Token exhaustion | DoS | Medium | Medium | Medium | Rate limiting | Done |
| 4 | Model version drift | Tampering | Low | Medium | Low | Pin model version | Planned || Issue | Cause | Fix |
|---|---|---|
| Threats missed | No systematic approach | Use STRIDE for each component |
| Risk overestimated | No likelihood assessment | Rate likelihood AND impact separately |
| Model incomplete | Missing trust boundaries | Draw boundary diagram FIRST |
| Mitigations not implemented | No tracking | Add status column to threat register |
| Practice | Rationale |
|---|---|
| Defense in depth | Multiple layers, not single point |
| Least privilege always | Minimize blast radius |
| Audit log everything | Evidence for compliance and debugging |
| Automate scanning in CI | Catch issues before merge |
| Rotate credentials regularly | Reduce exposure window |
| Test security controls | Verify they actually work |
fai-security-review-skill — OWASP LLM Top 10 reviewfai-secret-scanning — Pre-commit secret detectionfai-threat-model — STRIDE threat analysisfai-guardrails-policy — AI safety guardrails~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.