← Back to Skills Marketplace
Mixture of Agents
by
John Scianna
· GitHub ↗
· v1.2.0
671
Downloads
0
Stars
1
Active Installs
1
Versions
Install in OpenClaw
/install moa
Description
Mixture of Agents: Make 3 frontier models argue, then synthesize their best insights into one superior answer. ~$0.03/query.
Usage Guidance
This skill is coherent and appears to be what it claims: it needs an OpenRouter API key and will send your prompts (and the model responses) to openrouter.ai. Before installing, (1) confirm you are comfortable sending query text and any sensitive context to OpenRouter, (2) set OPENROUTER_API_KEY only with a key scoped/rotatable for this use, (3) inspect scripts/moa-paid.js (it contains a hard-coded demo prompt and runs immediately if executed), and (4) avoid using this skill with secrets or highly sensitive data because prompts and responses are logged to stdout and sent to external models. Also correct or be aware of the metadata mismatch (some metadata claims no env vars while the manifest and SKILL.md require OPENROUTER_API_KEY). If you need stronger assurance, ask the author for a canonical repository URL and a checksumed release, and test the code in an isolated environment first.
Capability Analysis
Type: OpenClaw Skill
Name: moa
Version: 1.2.0
The OpenClaw skill 'moa' is designed to orchestrate multiple LLM calls via OpenRouter. Its code (scripts/moa.js, scripts/moa-paid.js) and documentation (SKILL.md, manifest.json) consistently show its purpose: taking a user prompt, sending it to various LLMs on openrouter.ai using a user-provided OPENROUTER_API_KEY, and then synthesizing the results. There is no evidence of malicious intent, such as unauthorized data exfiltration, arbitrary code execution, persistence mechanisms, or prompt injection attempts against the OpenClaw agent itself. All network calls are directed to the expected OpenRouter API endpoint, and the API key is used for legitimate authentication.
Capability Assessment
Purpose & Capability
The skill's stated purpose (mixing multiple LLMs and synthesizing their outputs) matches the code and instructions. However, registry metadata at the top of the provided package summary lists no required environment variables, while both SKILL.md and manifest.json require OPENROUTER_API_KEY — an inconsistency that should be corrected but is not evidence of malicious behavior.
Instruction Scope
SKILL.md and the code instruct the agent to call OpenRouter's chat completions API with the user's prompt, aggregate model responses, and synthesize them. The instructions do not ask for unrelated files or credentials. One minor operational note: scripts/moa-paid.js contains a hard-coded example prompt and calls runMoA(prompt) immediately (it runs when executed), which is a benign demo but may be surprising if someone runs that file expecting only a library. The scripts also print prompt snippets and model responses to stdout (logging), which may expose sensitive prompts/responses in logs.
Install Mechanism
There is no install spec (instruction-only skill with Node.js files). No remote downloads or archive extraction are used. The package relies on axios (a normal dependency) and runtime Node >=18 as declared in manifest.json.
Credentials
The only required secret is OPENROUTER_API_KEY, which is appropriate for a skill that calls OpenRouter. The earlier top-level summary incorrectly claimed 'no required env vars', which conflicts with the manifest/SKILL.md — the environment requirement itself is proportionate, but the metadata mismatch should be fixed.
Persistence & Privilege
The skill does not request always: true and does not modify other skills or global agent configuration. It behaves as a normal user-invocable skill that makes outbound API calls; autonomous invocation is allowed by default but not unusually privileged here.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install moa - After installation, invoke the skill by name or use
/moa - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.2.0
V1.2: Better docs, API reference, failure modes. Dogfooded with MoA itself.
Metadata
Frequently Asked Questions
What is Mixture of Agents?
Mixture of Agents: Make 3 frontier models argue, then synthesize their best insights into one superior answer. ~$0.03/query. It is an AI Agent Skill for Claude Code / OpenClaw, with 671 downloads so far.
How do I install Mixture of Agents?
Run "/install moa" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Mixture of Agents free?
Yes, Mixture of Agents is completely free (open-source). You can download, install and use it at no cost.
Which platforms does Mixture of Agents support?
Mixture of Agents is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Mixture of Agents?
It is built and maintained by John Scianna (@jscianna); the current version is v1.2.0.
More Skills