Collusion Sentinel for Agent Markets
Emergent collusion leaves no agreement to find — it lives in the history.
我该使用它吗
质量与安全性
A
基于对工具定义和协议合规性的自动分析。
上下文开销
~266token 数（工具定义）
~436 B典型响应大小
对注意力的影响极小（占 128k 上下文窗口的 0.21%）
这是每次将服务器的工具加载到模型上下文窗口时所消耗的大致 token 数。数值越高，可用于其他任务的注意力就越少。
安装
一键安装
将以下内容添加到你的 `claude_desktop_config.json` 文件中：
{
"mcpServers": {
"collusion-sentinel-agent-markets": {
"url": "https://textforagents.com/products/collusion-sentinel-agent-markets/mcp"
}
}
}远程端点
https://textforagents.com/products/collusion-sentinel-agent-markets/mcpstreamable-http它能做什么
工具清单
工具（3）
🟢 只读🟡 写入🔴 删除⚪ 未知
⚪sentinel_audit(mode, rounds)
Score a pair/population joint price history (mode=prices) or a verifier pair's acceptance history (mode=verifier) for collusion signatures. $0.05/audit x402 after the one-time 100-audit grant; degenerate histories refused, never charged.
输入模式
{
"type": "object",
"properties": {
"mode": {
"type": "string"
},
"rounds": {
"type": "array",
"items": {
"type": "object"
}
}
},
"required": [
"rounds"
]
}⚪sentinel_certify(subject, kind, model, prompt_hash)
Run the tempt-to-violate certification battery for a subject ($25/certificate x402; battery model inference subject-pays / BYOK). model and prompt_hash are required — the certificate hash binds them.
输入模式
{
"type": "object",
"properties": {
"subject": {
"type": "string"
},
"kind": {
"type": "string"
},
"model": {
"type": "string"
},
"prompt_hash": {
"type": "string"
}
},
"required": [
"subject",
"kind",
"model",
"prompt_hash"
]
}⚪sentinel_control_plan(audit_id, pair)
Score-keyed counterparty rotation + memory hygiene plan (included).
输入模式
{
"type": "object",
"properties": {
"audit_id": {
"type": "string"
},
"pair": {
"type": "string"
}
},
"required": [
"audit_id"
]
}社区
证据
最近观测
已验证未记录版本3 个工具