---
user-invocable: true
name: ai-agent-ux-review
description: Reviews UX of AI agents, coding assistants, and autonomous systems, focusing on trust, transparency, and controllability.
---

# Purpose

Evaluate UX of AI agents like Claude Code, Codex, Cursor, or agentic systems.

---

# Evaluate:

## 1. Agent transparency
Can the user see:

- what agent is doing
- what agent plans to do
- what agent already did

Hidden actions are CRITICAL UX failures.

---

## 2. Predictability
Does the agent behave consistently?

Unpredictable behavior destroys trust.

---

## 3. Controllability
Can the user:

- interrupt
- redirect
- undo
- constrain agent actions

---

## 4. Feedback clarity

Agent must clearly show:

- reasoning
- actions
- changes made
- failures

---

## 5. Trust safety

Evaluate:

- destructive action safeguards
- confirmation requirements
- preview mechanisms

---

# Critical Failure Conditions

Flag CRITICAL if:

- agent acts without visibility
- agent cannot be controlled
- agent makes silent changes
- agent failures are unclear

---

# Output

Include:

Trust score: 1–10  
Controllability score: 1–10  
Transparency score: 1–10  

List risks and fixes.