Multi Model Review Router is a skill that acts as an orchestration layer for AI agents that need independent verification from external models and tools. Rather than letting a single model validate its own output — a pattern that can produce repeated errors — this skill routes tasks to secondary systems and consolidates their findings into an adjudicated report with a final decision-making summary.

The skill is designed for three broad scenarios. First, adversarial validation: the skill explicitly directs an external reviewer to seek flaws in a proposed plan or code change, rather than confirming it. Second, rescue operations: when an agent has failed a task repeatedly, the skill can automatically trigger a second-opinion review from a different model. Third, specialized routing: tasks involving large repositories, video, or audio can be sent to models better suited to handle that content, such as Gemini or local LLMs.

It integrates with CLI-based reviewers, MCP servers, and secondary APIs already present in the agent’s environment. The output is a structured report combining raw findings from those external tools with a synthesized conclusion. This skill is appropriate for high-stakes workflows — code review pipelines, security audits, or multi-model consensus checks — where independent verification matters more than speed.

Use cases

Route a proposed code change to an adversarial reviewer that specifically looks for flaws before merging

Trigger an automatic second-opinion review when an agent has failed the same task multiple times

Send a large repository to a long-context model for analysis the primary model cannot handle

Delegate audio or video content to a multimodal model and receive findings back in a unified report

Coordinate multi-model consensus on a high-stakes decision and produce an adjudicated summary

Integrate CLI-based code review tools into an agent workflow and combine their output with model findings

When to use it

The agent is reviewing or generating code where self-validation is insufficient

A task has failed repeatedly and a second model's perspective is needed

The input involves modalities such as video, audio, or very large repos the primary model handles poorly

A security or adversarial audit requires deliberate fault-finding rather than confirmation

Multiple independent model opinions must be combined into one structured report

When not to use it

No external models, CLI tools, or secondary APIs are available in the environment to route tasks to

The task is low-stakes and the overhead of multi-model orchestration is not justified

Real-time latency is critical, as routing to external tools adds round-trip time

A simple single-model completion is all that is needed