ak-predict

name	ak:predict
description	5 expert personas debate proposed changes before implementation. Catches architectural, security, performance, and UX issues early. Use before major features or risky changes.
user-invocable	true
when_to_use	Invoke before high-risk changes that need persona debate.
category	utilities
keywords	["prediction","debate","review","risk"]
argument-hint	<feature description or change proposal> [--files <glob>] [--chain reason|probe]
metadata	{"author":"agentkit","attribution":"Multi-persona prediction pattern adapted from autoresearch by Udit Goenka (MIT)","license":"MIT","version":"1.0.0"}

ak:predict — Multi-Persona Pre-Analysis

Five expert personas independently analyze a proposed change, then debate conflicts to produce a consensus verdict before a single line of code is written.

When to Use

• Before implementing a major or high-risk feature
• Before a significant refactor or architecture change
• Evaluating competing technical approaches
• Stress-testing assumptions in a proposed design

When NOT to Use

• Trivial or low-risk changes (use ak:debug for bugs, ak:plan for already-decided tasks)
• Already-approved work with no open design questions
• Pure dependency upgrades with no API changes

---

The 5 Personas

Persona	Focus	Core Questions
Architect	System design, scalability, coupling	Does this fit the architecture? Will it scale? What new coupling does it introduce?
Security	Attack surface, data protection, auth	What can be abused? Where is data exposed? Are auth boundaries respected?
Performance	Latency, memory, queries, bundle size	What is the latency impact? N+1 queries? Memory leaks? Bundle bloat?
UX	User experience, accessibility, error states	Is this intuitive? What does the error state look like? Accessible on mobile?
Devil's Advocate	Hidden assumptions, simpler alternatives	Why not do nothing? What is the simplest alternative? Which assumption could be wrong?

---

Debate Protocol

1. Read the proposed change/feature description from the argument
2. Read relevant code if file paths are provided (grep for affected areas)
3. Each persona analyzes independently — do not let personas influence each other during this phase
4. Identify agreements — points where all (or 4+) personas align
5. Identify conflicts — points where personas meaningfully disagree
6. Weigh tradeoffs — for each conflict, evaluate which concern has higher impact
7. Produce verdict — GO / CAUTION / STOP with actionable recommendations