Outcome-verified watchdog for scheduled agents
Watchdog that verifies scheduled agents' artifacts itself — never the claim.
我该使用它吗
质量与安全性
基于对工具定义和协议合规性的自动分析。
上下文开销
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": {
"scheduled-agent-outcome-verification": {
"url": "https://textforagents.com/products/scheduled-agent-outcome-verification/mcp"
}
}
}远程端点
https://textforagents.com/products/scheduled-agent-outcome-verification/mcpstreamable-http它能做什么
工具清单
工具（8）
⚪register_job(job_id, contract, stall_budget_s, due_in_s)
Register a scheduled job with an outcome contract (file / URL / record, present + non-empty + fresh, optional hash/marker — or a chain contract whose stages' artifacts are the next stage's required input). Args: job_id, contract, stall_budget_s?, due_in_s?, alert via account webhook.
输入模式
{
"type": "object",
"properties": {
"job_id": {
"type": "string"
},
"contract": {
"type": "object"
},
"stall_budget_s": {
"type": "number"
},
"due_in_s": {
"type": "number"
}
},
"required": [
"job_id",
"contract"
]
}⚪start_run(job_id)
Start a run for a job. Args: job_id.
输入模式
{
"type": "object",
"properties": {
"job_id": {
"type": "string"
}
},
"required": [
"job_id"
]
}⚪checkpoint(run_id, note)
Report run progress — the watchdog watches progress checkpoints against the job's own wall-clock budget, never process existence. Args: run_id, note?.
输入模式
{
"type": "object",
"properties": {
"run_id": {
"type": "string"
},
"note": {
"type": "string"
}
},
"required": [
"run_id"
]
}🟢complete_run(run_id, claimed)
Claim completion; the watchdog verifies the declared artifact itself (the claim is never trusted) and issues a signed, hash-chained verdict receipt. Metered check ($0.002 after the one-time 100-check trial / plan rules). Args: run_id, claimed?.
输入模式
{
"type": "object",
"properties": {
"run_id": {
"type": "string"
},
"claimed": {
"type": "string"
}
},
"required": [
"run_id"
]
}🟢verify_now(job_id)
Spot-verify a job's artifact now (independent fetch/hash/read-back by the watchdog). Metered check ($0.002 after trial / plan rules). Args: job_id.
输入模式
{
"type": "object",
"properties": {
"job_id": {
"type": "string"
}
},
"required": [
"job_id"
]
}⚪watchdog_tick
Run the watchdog sweep: STALLED for active runs with no progress checkpoint inside the job's wall-clock budget, MISSED once per due window with no verified run. Not metered.
输入模式
{
"type": "object",
"properties": {},
"required": []
}⚪last_verified_success(job_id)
When was this job's last VERIFIED success, and how old is it? Args: job_id.
输入模式
{
"type": "object",
"properties": {
"job_id": {
"type": "string"
}
},
"required": [
"job_id"
]
}🟢list_verdicts
List this operator's signed verdict receipts + receipt-chain re-verification result.
输入模式
{
"type": "object",
"properties": {},
"required": []
}社区
证据