# Flight Risk Assessment Prover MCP for AI Agents AI Agent Connect

> Flight Risk Assessment Prover forces ICAO SMS-level rigor onto flight risk assessments. It moves your AI agent past vague 'weather risk' descriptions to require specific METAR/TAF data, ICAO 5x5 risk matrices, and Swiss Cheese barrier modeling. It eliminates sycophantic go-bias by demanding explicit, defensible GO/NO-GO decisions based on quantified threat indices and human factors like SHELL and IMSAFE checklists.

## Overview
- **Category:** aviation
- **Price:** Free
- **Endpoint:** https://edge.vinkius.com/vk_preview_yVQ8SH5a7Wg2PDaxUdAoa4oiGD5CpUjO8dF2jCt1/ai-agent-connect
- **Tags:** aviation, flight-risk, icao, sms, safety, tem, swiss-cheese, shell, imsafe, go-no-go, risk-assessment, prover

## Description

Flight Risk Assessment Prover forces ICAO SMS-level rigor onto flight risk assessments. Imagine a situation where a dispatch office clears a flight into known thunderstorm activity with a simple 'proceed with caution.' That's how accidents happen. This Connector changes that by forcing your AI client to think like a high-level aviation safety officer. Instead of letting the agent give you a lazy summary, it demands specific data points like crosswind components, RVR versus minimums, and exact icing levels. It forces the AI to calculate a Risk Index using the ICAO 5x5 matrix, where any score over 15 is an automatic no-go. You'll see the agent map out the Swiss Cheese layers. It looks at Organizational, Supervisory, Preconditions, and Acts layers to find where the safety barriers are failing. It even digs into human factors, looking at fatigue hours and the SHELL model rather than just blaming pilot error. It forces the AI to consider the IMSAFE checklist for crew stress and alcohol. By using this Connector through the Vinkius catalog, you're moving from feelings-based risk to a defensible, audit-ready safety assessment that could actually stand up in a CVR transcript review. It turns a vague 'maybe' into a hard, data-backed decision.

## Tools

### validate_flight_risk
Forces the AI to identify specific threats, calculate ICAO 5x5 risk indices, and map Swiss Cheese barriers. It ensures the final output is a defensible GO/NO-GO decision based on human factors and METAR data.

## Prompt Examples

**Prompt:** 
```
Check if the crew can fly to LFLB tonight. PIC is tired and the FO is new. METAR shows mountains and snow.
```

**Response:** 
```
**RISK_ELEVATED**
The current conditions and crew status present significant risks:
* **Weather:** Mountainous terrain + snow + night conditions.
* **Human Factors:** PIC fatigue (high hours awake) + FO inexperience.
* **Verdict:** NO-GO. The combination of high-fatigue and non-precision mountain approach in snow exceeds safety margins.
```

**Prompt:** 
```
We have three MEL items: APU off, one pack off, and no autobrakes. Is it safe to fly into fog?
```

**Response:** 
```
**RISK_ELEVATED**
Analysis of MEL items and environment:
* **MEL Impact:** Single-pack and APU off limits may restrict altitude/ascent.
* **Environment:** Low visibility (fog) increases risk of runway excursion without autobrakes.
* **Compound Index:** >15 (Intolerable).
* **Verdict:** NO-GO. Cross-check dispatch minima for single-pack operation immediately.
```

**Prompt:** 
```
Analyze this flight plan: B737 from EDDM to EGLL. METAR shows CB activity. Is there a risk of hitting a ceiling limit?
```

**Response:** 
```
**RISK_ELEVATED**
Analysis of flight risk:
* **Threat:** CB activity + potential ceiling limits.
* **ICAO 5x5:** Weather (C3) + Fuel/Time (C3).
* **Verdict:** NO-GO. The risk of encountering CB tops exceeds the DA and requires a detour that may exceed fuel reserves.
```

## Capabilities

### Quantify threat levels using ICAO 5x5 matrices
The AI calculates a specific Risk Index to replace vague adjectives like 'medium' or 'high' with hard numbers.

### Map safety barriers using the Swiss Cheese model
It identifies holes in organizational, supervisory, and operational layers to find where accidents happen.

### Analyze human factors via SHELL and IMSAFE checklists
The agent evaluates crew fatigue, stress, and medication rather than just attributing risks to pilot error.

### Extract specific METAR and TAF weather parameters
It forces the AI to cite exact values like crosswind components and RVR instead of generic weather summaries.

### Enforce binary GO/NO-GO decisions to eliminate go-bias
The agent must provide a clear, defensible decision based on pre-defined criteria to avoid sycophantic safety gaps.

## Use Cases

### Complex Weather Navigation
A dispatcher wants to know if a flight can proceed with a degraded radar and CB activity. The agent uses validate_flight_risk to calculate the compound risk index and issue a NO-GO.

### Crew Fatigue Management
A crew is approaching their limits. The agent uses validate_flight_risk to check FDP position and hours since sleep against IMSAFE criteria.

### MEL Item Analysis
A plane has multiple inoperative items. The agent uses validate_flight_risk to check if these deferrals align across Swiss Cheese layers.

### Audit Preparation
A safety officer needs to justify a recent flight plan. The agent uses validate_flight_risk to create a CVR-defensible risk assessment.

## Benefits

- Eliminate go-bias by forcing the agent to provide a binary GO/NO-GO verdict rather than 'proceed with caution' advice.
- Get precise risk quantification by using the ICAO 5x5 matrix to turn feelings into a measurable Risk Index.
- Identify hidden safety gaps by mapping Organizational, Supervisory, Preconditions, and Acts layers in the Swiss Cheese model.
- Move past vague weather reports by requiring specific METAR/TAF parameters like crosswind components and RVR values.
- Improve human factors analysis by forcing the agent to use the SHELL model and IMSAFE checklists for crew fatigue and stress.

## How It Works

The bottom line is that it replaces vague 'proceed with caution' advice with a rigorous, quantifiable safety audit.

1. Input flight details, METAR/TAF data, and crew status into your AI client.
2. The Connector forces the agent to calculate risk indices and map defense layers.
3. You receive a defensible, audit-ready risk verdict with specific GO/NO-GO criteria.

## Frequently Asked Questions

**What does the Flight Risk Assessment Prover do for aviation safety?**
It forces your AI agent to conduct rigorous, ICAO-standard safety audits. It moves past vague weather summaries to provide quantifiable risk indices and defensible go/no-go decisions.

**How does it help with ICAO SMS compliance?**
It automates the rigor required for Safety Management Systems by forcing the AI to map safety barriers across organizational and supervisory layers.

**Can it help me avoid go-bias in flight planning?**
Yes. It eliminates the tendency to simply 'proceed with caution' by requiring a binary decision based on specific threat parameters and risk scoring.

**Does it look at crew fatigue and human factors?**
It does. The Connector forces the AI to use the SHELL model and IMSAFE checklists to analyze fatigue, stress, and other human factors.

**How does it handle complex weather like thunderstorms?**
It requires the AI to cite specific METAR/TAF data, including crosswind components and RVR, to calculate a compound risk index.

**Can it analyze MEL items and safety barriers?**
Yes. It checks for alignment across the Swiss Cheese layers to see if multiple mechanical issues combined with environmental factors create a critical safety gap.

**What makes this different from a standard flight risk assessment tool?**
Standard tools accept 'medium risk' as an answer. This Prover rejects anything that is not quantified on the ICAO 5×5 matrix with probability (A-E) and severity (1-5). It catches 5 specific failure modes: generic threats without METAR data, adjective-based risk without numerical scoring, missing Swiss Cheese barrier analysis, human factors reduced to 'pilot error' instead of SHELL/IMSAFE, and sycophantic go-bias where the AI says 'proceed with caution' instead of committing NO-GO when the data demands it.

**How does the go-bias detection work?**
The engine maintains a semantic trap list of go-bias phrases: 'proceed with caution,' 'acceptable to proceed,' 'can proceed,' 'within acceptable limits.' If the LLM uses any of these instead of an explicit GO or NO-GO decision with pre-defined criteria, the assessment is rejected with GO_BIAS verdict. The LLM must define NO-GO criteria BEFORE the assessment and then commit to a binary decision defensible on a CVR transcript.

**What aviation frameworks does this enforce?**
Five industry-standard frameworks: (1) ICAO Annex 19 Safety Management System — proactive hazard identification and risk assessment. (2) ICAO 5×5 Risk Matrix — probability × severity quantification. (3) Reason's Swiss Cheese Model — multi-layer defense analysis with hole alignment detection. (4) SHELL Model — Software-Hardware-Environment-Liveware interaction analysis for human factors. (5) Threat and Error Management (TEM) — threat categorization into environmental, airline, and crew factors per FAA AC 120-92B.