AI Second-Opinion Bridge

What it does

Uses a second, independent AI model as a sanity check on a plan before executing it: the reviewer iterates with the primary agent to reach consensus, with a stricter model called in only for high-risk approvals. Execution is then dispatched across risk tiers, running low-risk items in parallel while high-risk ones wait for explicit human approval.

When to use it

• You want an independent check on a risky plan before building or deploying it
• You're running several execution tracks at once and need risk-based routing instead of one-size-fits-all approval
• You want to reduce blind spots from relying on a single model's judgment

Steps

1. Draft the plan with the primary agent and list each action item's risk level
2. Send the plan to the second model for review, keeping that reviewer strictly read-only
3. Iterate between the two until they reach consensus on the plan
4. Escalate any high-risk or disputed item to a stricter review pass before approval
5. Dispatch low-risk approved items for parallel execution
6. Hold all high-risk items for explicit human sign-off before running them

The skill

---
name: ai-second-opinion-bridge
description: Route a plan through a second AI model for independent review and risk-tiered execution, keeping the reviewer read-only.
---

# AI Second-Opinion Bridge

## 1. Overview
Uses a second, independent AI model as a sanity check on a plan before executing it: the reviewer iterates with the primary agent to reach consensus, with a stricter model called in only for high-risk approvals. Execution is then dispatched across risk tiers, running low-risk items in parallel while high-risk ones wait for explicit human approval.

## 2. When to Use
- You want an independent check on a risky plan before building or deploying it
- You're running several execution tracks at once and need risk-based routing instead of one-size-fits-all approval
- You want to reduce blind spots from relying on a single model's judgment

## 3. Steps
1. Draft the plan with the primary agent and list each action item's risk level
2. Send the plan to the second model for review, keeping that reviewer strictly read-only
3. Iterate between the two until they reach consensus on the plan
4. Escalate any high-risk or disputed item to a stricter review pass before approval
5. Dispatch low-risk approved items for parallel execution
6. Hold all high-risk items for explicit human sign-off before running them

## 4. GoHighLevel Angle
Apply this inside GoHighLevel workflows (funnels, automations, or client reporting) wherever the same task recurs for an agency or client account.

## 5. Safety Rules
- Always produce a draft first; never send, publish, or spend money without explicit human approval.
- Never invent data, metrics, or results — use only what is confirmed or clearly label estimates.
- Never paste API keys, tokens, or passwords into chat; reference secrets by name from a secure store.

For Claude Code, save it as ~/.claude/skills/ai-second-opinion-bridge/SKILL.md. In Cursor, ChatGPT or another assistant, paste it as project instructions.