Works with the AI Agents you already use
crg ai agent risk
Generates structured cybersecurity risk scenarios for AI agents and LLMs using CRGO and OWASP frameworks.
$10
crg ai agent risk
Example session with this skill installed
Generate risk scenarios for a RAG-based HR assistant. It uses a self-hosted Llama 3 model, reads internal employee PDF folders, and acts with approval to draft emails to candidates. No logs.
- Read your context and instructions
- Compiled the crg ai agent
The risk analysis for the HR assistant is complete. I have mapped the autonomy levels, identified potential injection vectors via untrusted document ingestion, and provided a management priority list.
1. AI inventory
| System | Purpose | Model | Autonomy | Tools and permissions | Data read / written | Untrusted content | Oversight | Supply chain | Evidence |
|---|---|---|---|---|---|---|---|---|---|
| HR Assistant | Recruitment support | Llama 3 (Self-hosted) | Acts with approval | Draft e-mail | Employee records (Sensitive) | Uploaded CVs/PDFs | Human-in-the-loop | Local model deployment | E1 |
2. Pattern and OWASP mapping
| System | CRGO pattern | OWASP id and CRGO label | Why it applies |
|---|---|---|---|
| HR Assistant | PatPromptInjection | LLM01:2025 manipulation by injected instructions | Ingests untrusted PDFs from candidates |
| HR Assistant | PatAgenticMisuse | ASI01:2026 redirection of an agent's goals | Access to internal employee folders via RAG |
3. Risk scenarios
AI-S1: A candidate uploads a CV containing hidden system instructions (threat source) → the RAG system processes the prompt injection (weakness) → HR Assistant and employee folders (asset) → the agent exfiltrates internal salary data into a drafted email (event) → privacy breach and regulatory non-compliance (consequence).
Analytical estimate — validation required.
5. Priority list for management
| Rank | Scenario | Tier | Reason | Decision for management | Marker |
|---|---|---|---|---|---|
| 1 | AI-S1 | High | Sensitive data access + untrusted input | Implement robust input filtering for PDFs | Management decision required |
Quantification is not performed here; a separate quantification step is needed.
Next steps
- Validate the sensitivity classification of the employee PDF folders.
- Review the recommended SP 800-53 controls for least privilege access.
- Perform a formal quantification of the potential regulatory fines.
Connects securely to your tools. The creator never sees your data.
About this skill
AI agents and LLMs introduce unique security failures, from prompt injection to excessive autonomy. This skill structures cybersecurity risk scenarios for your AI inventory using the CyberRiskGuardian method, mapping your systems to CRGO patterns and OWASP LLM/Agentic identifiers. It converts raw technical specs into structured causal chains and management-ready priority lists to support pre-deployment reviews and risk registers.
What it does
- AI Inventory Mapping builds a detailed table of models, autonomy levels, tool permissions, and data flows.
- Pattern Identification maps assets to CRGO patterns like
PatPromptInjection,PatAgenticMisuse, andPatAISupplyChain. - Causal Scenario Writing generates structured chains linking threat sources and weaknesses to specific operational consequences.
- Control Gap Analysis compares your existing safeguards against recommended NIST SP 800-53 and CIS v8.1 controls.
- Management Prioritization produces a qualitative risk list (High, Medium, Low) for executive decision-making.
How it works
- Supply an inventory of your AI systems, including model types, tools, autonomy levels, and data sensitivity.
- The agent identifies information gaps and marks assumptions for organizational validation.
- It generates a full risk report with six mandatory sections, including OWASP mappings and ATT&CK technique IDs.
- Review the management priority list to identify where human approval gates or kill switches are required.
Frameworks & tools
This skill utilizes the CyberRiskGuardian risk method, OWASP Top 10 for LLM (2025) and Agentic AI (2026), NIST SP 800-53, MITRE ATLAS, MITRE ATT&CK, and CIS Controls v8.1.
Why this beats prompting it yourself
Generic prompts fail to maintain the rigorous causal chain required for formal risk assessments. This skill enforces specific taxonomy from the CRGO ontology, ensuring that every risk is mapped to industry-standard IDs rather than vague descriptions.
Use cases
- Conduct a pre-deployment security review for an autonomous CRM-linked agent.
- Draft a formal AI risk register for compliance with ISO/IEC 42001 or 23894.
- Analyze the supply chain risks of using third-party MCP servers or plugins.
- Brief management on the impact of prompt injection on internal RAG assistants.
Known limitations
Does not perform quantitative risk analysis, financial modeling, or CVSS scoring. It excludes model bias, fairness assessments, and penetration testing instructions.
Requires a manual AI inventory input. Qualitative tiers only; no probability or dollar values. Does not provide ISO/IEC 42001 clause text or conformity statements.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 7 days ago