> ## Documentation Index
> Fetch the complete documentation index at: https://agdiai-597e782f.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Threat Model (MITRE ATLAS)

# Agdi Threat Model v1.0

## MITRE ATLAS Framework

**Version:** 1.0-draft
**Last Updated:** 2026-02-04
**Methodology:** MITRE ATLAS + Data Flow Diagrams
**Framework:** [MITRE ATLAS](https://atlas.mitre.org/) (Adversarial Threat Landscape for AI Systems)

### Framework Attribution

This threat model is built on [MITRE ATLAS](https://atlas.mitre.org/), the industry-standard framework for documenting adversarial threats to AI/ML systems. ATLAS is maintained by [MITRE](https://www.mitre.org/) in collaboration with the AI security community.

**Key ATLAS Resources:**

* [ATLAS Techniques](https://atlas.mitre.org/techniques/)
* [ATLAS Tactics](https://atlas.mitre.org/tactics/)
* [ATLAS Case Studies](https://atlas.mitre.org/studies/)
* [ATLAS GitHub](https://github.com/mitre-atlas/atlas-data)
* [Contributing to ATLAS](https://atlas.mitre.org/resources/contribute)

### Contributing to This Threat Model

This is a living document maintained by the Agdi community. See [CONTRIBUTING-THREAT-MODEL.md](https://agdiai-597e782f.mintlify.site/security/CONTRIBUTING-THREAT-MODEL.md) for guidelines on contributing:

* Reporting new threats
* Updating existing threats
* Proposing attack chains
* Suggesting mitigations

***

## 1. Introduction

### 1.1 Purpose

This threat model documents adversarial threats to the Agdi AI agent platform and ClawHub skill marketplace, using the MITRE ATLAS framework designed specifically for AI/ML systems.

### 1.2 Scope

| Component | Included | Notes |
| - | - | - |
| Agdi Agent Runtime | Yes | Core agent execution, tool calls, sessions |
| Gateway | Yes | Authentication, routing, channel integration |
| Channel Integrations | Yes | WhatsApp, Telegram, Discord, Signal, Slack, etc. |
| ClawHub Marketplace | Yes | Skill publishing, moderation, distribution |
| MCP Servers | Yes | External tool providers |
| User Devices | Partial | Mobile apps, desktop clients |

### 1.3 Out of Scope

Nothing is explicitly out of scope for this threat model.

***

## 2. System Architecture

### 2.1 Trust Boundaries

```
┌─────────────────────────────────────────────────────────────────┐
│                    UNTRUSTED ZONE                                │
│  ┌─────────────┐  ┌─────────────┐  ┌─────────────┐              │
│  │  WhatsApp   │  │  Telegram   │  │   Discord   │  ...         │
│  └──────┬──────┘  └──────┬──────┘  └──────┬──────┘              │
│         │                │                │                      │
└─────────┼────────────────┼────────────────┼──────────────────────┘
          │                │                │
          ▼                ▼                ▼
┌─────────────────────────────────────────────────────────────────┐
│                 TRUST BOUNDARY 1: Channel Access                 │
│  ┌──────────────────────────────────────────────────────────┐   │
│  │                      GATEWAY                              │   │
│  │  • Device Pairing (30s grace period)                      │   │
│  │  • AllowFrom / AllowList validation                       │   │
│  │  • Token/Password/Tailscale auth                          │   │
│  └──────────────────────────────────────────────────────────┘   │
└─────────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────────┐
│                 TRUST BOUNDARY 2: Session Isolation              │
│  ┌──────────────────────────────────────────────────────────┐   │
│  │                   AGENT SESSIONS                          │   │
│  │  • Session key = agent:channel:peer                       │   │
│  │  • Tool policies per agent                                │   │
│  │  • Transcript logging                                     │   │
│  └──────────────────────────────────────────────────────────┘   │
└─────────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────────┐
│                 TRUST BOUNDARY 3: Tool Execution                 │
│  ┌──────────────────────────────────────────────────────────┐   │
│  │                  EXECUTION SANDBOX                        │   │
│  │  • Docker sandbox OR Host (exec-approvals)                │   │
│  │  • Node remote execution                                  │   │
│  │  • SSRF protection (DNS pinning + IP blocking)            │   │
│  └──────────────────────────────────────────────────────────┘   │
└─────────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────────┐
│                 TRUST BOUNDARY 4: External Content               │
│  ┌──────────────────────────────────────────────────────────┐   │
│  │              FETCHED URLs / EMAILS / WEBHOOKS             │   │
│  │  • External content wrapping (XML tags)                   │   │
│  │  • Security notice injection                              │   │
│  └──────────────────────────────────────────────────────────┘   │
└─────────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────────┐
│                 TRUST BOUNDARY 5: Supply Chain                   │
│  ┌──────────────────────────────────────────────────────────┐   │
│  │                      CLAWHUB                              │   │
│  │  • Skill publishing (semver, SKILL.md required)           │   │
│  │  • Pattern-based moderation flags                         │   │
│  │  • VirusTotal scanning (coming soon)                      │   │
│  │  • GitHub account age verification                        │   │
│  └──────────────────────────────────────────────────────────┘   │
└─────────────────────────────────────────────────────────────────┘
```

### 2.2 Data Flows

| Flow | Source | Destination | Data | Protection |
| - | - | - | - | - |
| F1 | Channel | Gateway | User messages | TLS, AllowFrom |
| F2 | Gateway | Agent | Routed messages | Session isolation |
| F3 | Agent | Tools | Tool invocations | Policy enforcement |
| F4 | Agent | External | web\_fetch requests | SSRF blocking |
| F5 | ClawHub | Agent | Skill code | Moderation, scanning |
| F6 | Agent | Channel | Responses | Output filtering |

***

## 3. Threat Analysis by ATLAS Tactic

### 3.1 Reconnaissance (AML.TA0002)

#### T-RECON-001: Agent Endpoint Discovery

| Attribute | Value |
| - | - |
| **ATLAS ID** | AML.T0006 - Active Scanning |
| **Description** | Attacker scans for exposed Agdi gateway endpoints |
| **Attack Vector** | Network scanning, shodan queries, DNS enumeration |
| **Affected Components** | Gateway, exposed API endpoints |
| **Current Mitigations** | Tailscale auth option, bind to loopback by default |
| **Residual Risk** | Medium - Public gateways discoverable |
| **Recommendations** | Document secure deployment, add rate limiting on discovery endpoints |

#### T-RECON-002: Channel Integration Probing

| Attribute | Value |
| - | - |
| **ATLAS ID** | AML.T0006 - Active Scanning |
| **Description** | Attacker probes messaging channels to identify AI-managed accounts |
| **Attack Vector** | Sending test messages, observing response patterns |
| **Affected Components** | All channel integrations |
| **Current Mitigations** | None specific |
| **Residual Risk** | Low - Limited value from discovery alone |
| **Recommendations** | Consider response timing randomization |

***

### 3.2 Initial Access (AML.TA0004)

#### T-ACCESS-001: Pairing Code Interception

| Attribute | Value |
| - | - |
| **ATLAS ID** | AML.T0040 - AI Model Inference API Access |
| **Description** | Attacker intercepts pairing code during 30s grace period |
| **Attack Vector** | Shoulder surfing, network sniffing, social engineering |
| **Affected Components** | Device pairing system |
| **Current Mitigations** | 30s expiry, codes sent via existing channel |
| **Residual Risk** | Medium - Grace period exploitable |
| **Recommendations** | Reduce grace period, add confirmation step |

#### T-ACCESS-002: AllowFrom Spoofing

| Attribute | Value |
| - | - |
| **ATLAS ID** | AML.T0040 - AI Model Inference API Access |
| **Description** | Attacker spoofs allowed sender identity in channel |
| **Attack Vector** | Depends on channel - phone number spoofing, username impersonation |
| **Affected Components** | AllowFrom validation per channel |
| **Current Mitigations** | Channel-specific identity verification |
| **Residual Risk** | Medium - Some channels vulnerable to spoofing |
| **Recommendations** | Document channel-specific risks, add cryptographic verification where possible |

#### T-ACCESS-003: Token Theft

| Attribute | Value |
| - | - |
| **ATLAS ID** | AML.T0040 - AI Model Inference API Access |
| **Description** | Attacker steals authentication tokens from config files |
| **Attack Vector** | Malware, unauthorized device access, config backup exposure |
| **Affected Components** | \~/.agdi/credentials/, config storage |
| **Current Mitigations** | File permissions |
| **Residual Risk** | High - Tokens stored in plaintext |
| **Recommendations** | Implement token encryption at rest, add token rotation |

***

### 3.3 Execution (AML.TA0005)

#### T-EXEC-001: Direct Prompt Injection

| Attribute | Value |
| - | - |
| **ATLAS ID** | AML.T0051.000 - LLM Prompt Injection: Direct |
| **Description** | Attacker sends crafted prompts to manipulate agent behavior |
| **Attack Vector** | Channel messages containing adversarial instructions |
| **Affected Components** | Agent LLM, all input surfaces |
| **Current Mitigations** | Pattern detection, external content wrapping |
| **Residual Risk** | Critical - Detection only, no blocking; sophisticated attacks bypass |
| **Recommendations** | Implement multi-layer defense, output validation, user confirmation for sensitive actions |

#### T-EXEC-002: Indirect Prompt Injection

| Attribute | Value |
| - | - |
| **ATLAS ID** | AML.T0051.001 - LLM Prompt Injection: Indirect |
| **Description** | Attacker embeds malicious instructions in fetched content |
| **Attack Vector** | Malicious URLs, poisoned emails, compromised webhooks |
| **Affected Components** | web\_fetch, email ingestion, external data sources |
| **Current Mitigations** | Content wrapping with XML tags and security notice |
| **Residual Risk** | High - LLM may ignore wrapper instructions |
| **Recommendations** | Implement content sanitization, separate execution contexts |

#### T-EXEC-003: Tool Argument Injection

| Attribute | Value |
| - | - |
| **ATLAS ID** | AML.T0051.000 - LLM Prompt Injection: Direct |
| **Description** | Attacker manipulates tool arguments through prompt injection |
| **Attack Vector** | Crafted prompts that influence tool parameter values |
| **Affected Components** | All tool invocations |
| **Current Mitigations** | Exec approvals for dangerous commands |
| **Residual Risk** | High - Relies on user judgment |
| **Recommendations** | Implement argument validation, parameterized tool calls |

#### T-EXEC-004: Exec Approval Bypass

| Attribute | Value |
| - | - |
| **ATLAS ID** | AML.T0043 - Craft Adversarial Data |
| **Description** | Attacker crafts commands that bypass approval allowlist |
| **Attack Vector** | Command obfuscation, alias exploitation, path manipulation |
| **Affected Components** | exec-approvals.ts, command allowlist |
| **Current Mitigations** | Allowlist + ask mode |
| **Residual Risk** | High - No command sanitization |
| **Recommendations** | Implement command normalization, expand blocklist |

***

### 3.4 Persistence (AML.TA0006)

#### T-PERSIST-001: Malicious Skill Installation

| Attribute | Value |
| - | - |
| **ATLAS ID** | AML.T0010.001 - Supply Chain Compromise: AI Software |
| **Description** | Attacker publishes malicious skill to ClawHub |
| **Attack Vector** | Create account, publish skill with hidden malicious code |
| **Affected Components** | ClawHub, skill loading, agent execution |
| **Current Mitigations** | GitHub account age verification, pattern-based moderation flags |
| **Residual Risk** | Critical - No sandboxing, limited review |
| **Recommendations** | VirusTotal integration (in progress), skill sandboxing, community review |

#### T-PERSIST-002: Skill Update Poisoning

| Attribute | Value |
| - | - |
| **ATLAS ID** | AML.T0010.001 - Supply Chain Compromise: AI Software |
| **Description** | Attacker compromises popular skill and pushes malicious update |
| **Attack Vector** | Account compromise, social engineering of skill owner |
| **Affected Components** | ClawHub versioning, auto-update flows |
| **Current Mitigations** | Version fingerprinting |
| **Residual Risk** | High - Auto-updates may pull malicious versions |
| **Recommendations** | Implement update signing, rollback capability, version pinning |

#### T-PERSIST-003: Agent Configuration Tampering

| Attribute | Value |
| - | - |
| **ATLAS ID** | AML.T0010.002 - Supply Chain Compromise: Data |
| **Description** | Attacker modifies agent configuration to persist access |
| **Attack Vector** | Config file modification, settings injection |
| **Affected Components** | Agent config, tool policies |
| **Current Mitigations** | File permissions |
| **Residual Risk** | Medium - Requires local access |
| **Recommendations** | Config integrity verification, audit logging for config changes |

***

### 3.5 Defense Evasion (AML.TA0007)

#### T-EVADE-001: Moderation Pattern Bypass

| Attribute | Value |
| - | - |
| **ATLAS ID** | AML.T0043 - Craft Adversarial Data |
| **Description** | Attacker crafts skill content to evade moderation patterns |
| **Attack Vector** | Unicode homoglyphs, encoding tricks, dynamic loading |
| **Affected Components** | ClawHub moderation.ts |
| **Current Mitigations** | Pattern-based FLAG\_RULES |
| **Residual Risk** | High - Simple regex easily bypassed |
| **Recommendations** | Add behavioral analysis (VirusTotal Code Insight), AST-based detection |

#### T-EVADE-002: Content Wrapper Escape

| Attribute | Value |
| - | - |
| **ATLAS ID** | AML.T0043 - Craft Adversarial Data |
| **Description** | Attacker crafts content that escapes XML wrapper context |
| **Attack Vector** | Tag manipulation, context confusion, instruction override |
| **Affected Components** | External content wrapping |
| **Current Mitigations** | XML tags + security notice |
| **Residual Risk** | Medium - Novel escapes discovered regularly |
| **Recommendations** | Multiple wrapper layers, output-side validation |

***

### 3.6 Discovery (AML.TA0008)

#### T-DISC-001: Tool Enumeration

| Attribute | Value |
| - | - |
| **ATLAS ID** | AML.T0040 - AI Model Inference API Access |
| **Description** | Attacker enumerates available tools through prompting |
| **Attack Vector** | "What tools do you have?" style queries |
| **Affected Components** | Agent tool registry |
| **Current Mitigations** | None specific |
| **Residual Risk** | Low - Tools generally documented |
| **Recommendations** | Consider tool visibility controls |

#### T-DISC-002: Session Data Extraction

| Attribute | Value |
| - | - |
| **ATLAS ID** | AML.T0040 - AI Model Inference API Access |
| **Description** | Attacker extracts sensitive data from session context |
| **Attack Vector** | "What did we discuss?" queries, context probing |
| **Affected Components** | Session transcripts, context window |
| **Current Mitigations** | Session isolation per sender |
| **Residual Risk** | Medium - Within-session data accessible |
| **Recommendations** | Implement sensitive data redaction in context |

***

### 3.7 Collection & Exfiltration (AML.TA0009, AML.TA0010)

#### T-EXFIL-001: Data Theft via web\_fetch

| Attribute | Value |
| - | - |
| **ATLAS ID** | AML.T0009 - Collection |
| **Description** | Attacker exfiltrates data by instructing agent to send to external URL |
| **Attack Vector** | Prompt injection causing agent to POST data to attacker server |
| **Affected Components** | web\_fetch tool |
| **Current Mitigations** | SSRF blocking for internal networks |
| **Residual Risk** | High - External URLs permitted |
| **Recommendations** | Implement URL allowlisting, data classification awareness |

#### T-EXFIL-002: Unauthorized Message Sending

| Attribute | Value |
| - | - |
| **ATLAS ID** | AML.T0009 - Collection |
| **Description** | Attacker causes agent to send messages containing sensitive data |
| **Attack Vector** | Prompt injection causing agent to message attacker |
| **Affected Components** | Message tool, channel integrations |
| **Current Mitigations** | Outbound messaging gating |
| **Residual Risk** | Medium - Gating may be bypassed |
| **Recommendations** | Require explicit confirmation for new recipients |

#### T-EXFIL-003: Credential Harvesting

| Attribute | Value |
| - | - |
| **ATLAS ID** | AML.T0009 - Collection |
| **Description** | Malicious skill harvests credentials from agent context |
| **Attack Vector** | Skill code reads environment variables, config files |
| **Affected Components** | Skill execution environment |
| **Current Mitigations** | None specific to skills |
| **Residual Risk** | Critical - Skills run with agent privileges |
| **Recommendations** | Skill sandboxing, credential isolation |

***

### 3.8 Impact (AML.TA0011)

#### T-IMPACT-001: Unauthorized Command Execution

| Attribute | Value |
| - | - |
| **ATLAS ID** | AML.T0031 - Erode AI Model Integrity |
| **Description** | Attacker executes arbitrary commands on user system |
| **Attack Vector** | Prompt injection combined with exec approval bypass |
| **Affected Components** | Bash tool, command execution |
| **Current Mitigations** | Exec approvals, Docker sandbox option |
| **Residual Risk** | Critical - Host execution without sandbox |
| **Recommendations** | Default to sandbox, improve approval UX |

#### T-IMPACT-002: Resource Exhaustion (DoS)

| Attribute | Value |
| - | - |
| **ATLAS ID** | AML.T0031 - Erode AI Model Integrity |
| **Description** | Attacker exhausts API credits or compute resources |
| **Attack Vector** | Automated message flooding, expensive tool calls |
| **Affected Components** | Gateway, agent sessions, API provider |
| **Current Mitigations** | None |
| **Residual Risk** | High - No rate limiting |
| **Recommendations** | Implement per-sender rate limits, cost budgets |

#### T-IMPACT-003: Reputation Damage

| Attribute | Value |
| - | - |
| **ATLAS ID** | AML.T0031 - Erode AI Model Integrity |
| **Description** | Attacker causes agent to send harmful/offensive content |
| **Attack Vector** | Prompt injection causing inappropriate responses |
| **Affected Components** | Output generation, channel messaging |
| **Current Mitigations** | LLM provider content policies |
| **Residual Risk** | Medium - Provider filters imperfect |
| **Recommendations** | Output filtering layer, user controls |

***

## 4. ClawHub Supply Chain Analysis

### 4.1 Current Security Controls

| Control | Implementation | Effectiveness |
| - | - | - |
| GitHub Account Age | `requireGitHubAccountAge()` | Medium - Raises bar for new attackers |
| Path Sanitization | `sanitizePath()` | High - Prevents path traversal |
| File Type Validation | `isTextFile()` | Medium - Only text files, but can still be malicious |
| Size Limits | 50MB total bundle | High - Prevents resource exhaustion |
| Required SKILL.md | Mandatory readme | Low security value - Informational only |
| Pattern Moderation | FLAG\_RULES in moderation.ts | Low - Easily bypassed |
| Moderation Status | `moderationStatus` field | Medium - Manual review possible |

### 4.2 Moderation Flag Patterns

Current patterns in `moderation.ts`:

```javascript theme={"theme":{"light":"min-light","dark":"min-dark"}}
// Known-bad identifiers
/(keepcold131\/ClawdAuthenticatorTool|ClawdAuthenticatorTool)/i

// Suspicious keywords
/(malware|stealer|phish|phishing|keylogger)/i
/(api[-_ ]?key|token|password|private key|secret)/i
/(wallet|seed phrase|mnemonic|crypto)/i
/(discord\.gg|webhook|hooks\.slack)/i
/(curl[^\n]+\|\s*(sh|bash))/i
/(bit\.ly|tinyurl\.com|t\.co|goo\.gl|is\.gd)/i
```

**Limitations:**

* Only checks slug, displayName, summary, frontmatter, metadata, file paths
* Does not analyze actual skill code content
* Simple regex easily bypassed with obfuscation
* No behavioral analysis

### 4.3 Planned Improvements

| Improvement | Status | Impact |
| - | - | - |
| VirusTotal Integration | In Progress | High - Code Insight behavioral analysis |
| Community Reporting | Partial (`skillReports` table exists) | Medium |
| Audit Logging | Partial (`auditLogs` table exists) | Medium |
| Badge System | Implemented | Medium - `highlighted`, `official`, `deprecated`, `redactionApproved` |

***

## 5. Risk Matrix

### 5.1 Likelihood vs Impact

| Threat ID | Likelihood | Impact | Risk Level | Priority |
| - | - | - | - | - |
| T-EXEC-001 | High | Critical | **Critical** | P0 |
| T-PERSIST-001 | High | Critical | **Critical** | P0 |
| T-EXFIL-003 | Medium | Critical | **Critical** | P0 |
| T-IMPACT-001 | Medium | Critical | **High** | P1 |
| T-EXEC-002 | High | High | **High** | P1 |
| T-EXEC-004 | Medium | High | **High** | P1 |
| T-ACCESS-003 | Medium | High | **High** | P1 |
| T-EXFIL-001 | Medium | High | **High** | P1 |
| T-IMPACT-002 | High | Medium | **High** | P1 |
| T-EVADE-001 | High | Medium | **Medium** | P2 |
| T-ACCESS-001 | Low | High | **Medium** | P2 |
| T-ACCESS-002 | Low | High | **Medium** | P2 |
| T-PERSIST-002 | Low | High | **Medium** | P2 |

### 5.2 Critical Path Attack Chains

**Attack Chain 1: Skill-Based Data Theft**

```
T-PERSIST-001 → T-EVADE-001 → T-EXFIL-003
(Publish malicious skill) → (Evade moderation) → (Harvest credentials)
```

**Attack Chain 2: Prompt Injection to RCE**

```
T-EXEC-001 → T-EXEC-004 → T-IMPACT-001
(Inject prompt) → (Bypass exec approval) → (Execute commands)
```

**Attack Chain 3: Indirect Injection via Fetched Content**

```
T-EXEC-002 → T-EXFIL-001 → External exfiltration
(Poison URL content) → (Agent fetches & follows instructions) → (Data sent to attacker)
```

***

## 6. Recommendations Summary

### 6.1 Immediate (P0)

| ID | Recommendation | Addresses |
| - | - | - |
| R-001 | Complete VirusTotal integration | T-PERSIST-001, T-EVADE-001 |
| R-002 | Implement skill sandboxing | T-PERSIST-001, T-EXFIL-003 |
| R-003 | Add output validation for sensitive actions | T-EXEC-001, T-EXEC-002 |

### 6.2 Short-term (P1)

| ID | Recommendation | Addresses |
| - | - | - |
| R-004 | Implement rate limiting | T-IMPACT-002 |
| R-005 | Add token encryption at rest | T-ACCESS-003 |
| R-006 | Improve exec approval UX and validation | T-EXEC-004 |
| R-007 | Implement URL allowlisting for web\_fetch | T-EXFIL-001 |

### 6.3 Medium-term (P2)

| ID | Recommendation | Addresses |
| - | - | - |
| R-008 | Add cryptographic channel verification where possible | T-ACCESS-002 |
| R-009 | Implement config integrity verification | T-PERSIST-003 |
| R-010 | Add update signing and version pinning | T-PERSIST-002 |

***

## 7. Appendices

### 7.1 ATLAS Technique Mapping

| ATLAS ID | Technique Name | Agdi Threats |
| - | - | - |
| AML.T0006 | Active Scanning | T-RECON-001, T-RECON-002 |
| AML.T0009 | Collection | T-EXFIL-001, T-EXFIL-002, T-EXFIL-003 |
| AML.T0010.001 | Supply Chain: AI Software | T-PERSIST-001, T-PERSIST-002 |
| AML.T0010.002 | Supply Chain: Data | T-PERSIST-003 |
| AML.T0031 | Erode AI Model Integrity | T-IMPACT-001, T-IMPACT-002, T-IMPACT-003 |
| AML.T0040 | AI Model Inference API Access | T-ACCESS-001, T-ACCESS-002, T-ACCESS-003, T-DISC-001, T-DISC-002 |
| AML.T0043 | Craft Adversarial Data | T-EXEC-004, T-EVADE-001, T-EVADE-002 |
| AML.T0051.000 | LLM Prompt Injection: Direct | T-EXEC-001, T-EXEC-003 |
| AML.T0051.001 | LLM Prompt Injection: Indirect | T-EXEC-002 |

### 7.2 Key Security Files

| Path | Purpose | Risk Level |
| - | - | - |
| `src/infra/exec-approvals.ts` | Command approval logic | **Critical** |
| `src/gateway/auth.ts` | Gateway authentication | **Critical** |
| `src/web/inbound/access-control.ts` | Channel access control | **Critical** |
| `src/infra/net/ssrf.ts` | SSRF protection | **Critical** |
| `src/security/external-content.ts` | Prompt injection mitigation | **Critical** |
| `src/agents/sandbox/tool-policy.ts` | Tool policy enforcement | **Critical** |
| `convex/lib/moderation.ts` | ClawHub moderation | **High** |
| `convex/lib/skillPublish.ts` | Skill publishing flow | **High** |
| `src/routing/resolve-route.ts` | Session isolation | **Medium** |

### 7.3 Glossary

| Term | Definition |
| - | - |
| **ATLAS** | MITRE's Adversarial Threat Landscape for AI Systems |
| **ClawHub** | Agdi's skill marketplace |
| **Gateway** | Agdi's message routing and authentication layer |
| **MCP** | Model Context Protocol - tool provider interface |
| **Prompt Injection** | Attack where malicious instructions are embedded in input |
| **Skill** | Downloadable extension for Agdi agents |
| **SSRF** | Server-Side Request Forgery |

***

*This threat model is a living document. Report security issues to [security@agdi.ai](mailto:security@agdi.ai)*


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.