Collusion Sentinel for Agent Markets
Emergent collusion leaves no agreement to find — it lives in the history.
我該用這個嗎
品質與安全性
A
根據工具定義與協定合規性的自動化分析。
上下文成本
~266Token(工具定義)
~436 B典型回應大小
極小的注意力影響(128k 上下文的 0.21%)
這是每次將伺服器的工具載入模型上下文時所消耗的約略 token 數量。數量越高,可用於其他工作的注意力就越少。
安裝
一鍵安裝
將以下內容加入你的 `claude_desktop_config.json` 檔案:
{
"mcpServers": {
"collusion-sentinel-agent-markets": {
"url": "https://textforagents.com/products/collusion-sentinel-agent-markets/mcp"
}
}
}遠端端點
https://textforagents.com/products/collusion-sentinel-agent-markets/mcpstreamable-http它能做什麼
工具清單
工具(3)
🟢 唯讀🟡 寫入🔴 刪除⚪ 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.
輸入結構描述
{
"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.
輸入結構描述
{
"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).
輸入結構描述
{
"type": "object",
"properties": {
"audit_id": {
"type": "string"
},
"pair": {
"type": "string"
}
},
"required": [
"audit_id"
]
}社群
證據
近期觀測
已驗證未記錄版本3 個工具