메뉴
Bot-to-bot emergency recovery protocol via NATS. Each agent runs a CPR daemon listening on agents.cpr.<agent_name> that can ping, status, restart-gateway, kill-session, read-config, and check-mcp for the other agent. No SSH required — pure NATS.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Remotely maintains and upgrades Raspberry Pi systems running Pi-hole via SSH. Handles apt updates, upgrades, Pi-hole software updates, kernel patches, and safe reboots. Use when you need to synchronize OS versions or apply updates to a fleet of Raspberry Pi systems running Pi-hole.
Remediate specific CVEs across the fleet — given a cve-report.py or CVE ID, determine which hosts need what fixes and execute them without re-scanning anything.
Unified multi-agent coordination protocol for Rune and Clomp. NATS + JetStream for real-time signalling, NFS for durable task files and history log.
Automates rolling upgrades and maintenance for HA MicroK8s clusters on Ubuntu via SSH. Supports node discovery, channel selection, pre-flight checks, and stateful recovery.
Use when integrating Infisical secret management — debugging CLI errors (401, "not yet connected this project"), choosing auth methods (Service Tokens vs Machine Identities), wiring secrets into systemd/OpenClaw, installing the Infisical Operator on Kubernetes, writing InfisicalSecret CRDs, rotating DB passwords behind auto-reload, or hardening core infra (Postgres, Infisical itself) against plaintext credentials.
Remotely maintains and upgrades Ubuntu systems via SSH. Handles apt updates, upgrades, kernel patches, and safe reboots across multiple nodes. Use when you need to synchronize OS versions or apply security patches to a fleet of Ubuntu servers.
| name | agent-cpr |
| description | Bot-to-bot emergency recovery protocol via NATS. Each agent runs a CPR daemon listening on agents.cpr.<agent_name> that can ping, status, restart-gateway, kill-session, read-config, and check-mcp for the other agent. No SSH required — pure NATS. |
Each Hermes agent runs a lightweight CPR daemon (agent-cpr.py) as a systemd --user service. The daemon listens on agents.cpr.<agent_name> for JSON commands from the peer agent.
Rune (diffuser) Clomp (mink)
┌──────────────────────┐ ┌──────────────────────┐
│ agent-cpr.py │ NATS │ agent-cpr.py │
│ agents.cpr.rune ◄────┼──────────┼► agents.cpr.clomp │
│ │ │ │
│ hermes gateway │ │ hermes gateway │
└──────────────────────┘ └──────────────────────┘
| Command | What it does | Risk | Requires |
|---|---|---|---|
ping | Heartbeat: is the daemon alive? | None | — |
status | Gateway PID, uptime, platforms, MCP counts | Read-only | — |
restart-gateway | hermes gateway restart | Medium | "confirm": true |
kill-session | Delete a corrupted session from state.db | Low | session_id |
read-config | Return config section (redacted) | Low | section (optional) |
check-mcp | Test MCP endpoints, report which respond | Low | — |
{
"command": "status",
"params": {},
"source": "rune",
"reply_to": "agents.cpr.reply.rune.status.uuid"
}
The script lives at ~/.hermes/scripts/agent-cpr.py on each agent.
Source of truth: /shared/agents/common/scripts/agent-cpr.py
~/.config/systemd/user/agent-cpr.service:
[Unit]
Description=Agent CPR Daemon — bot-to-bot emergency recovery via NATS
After=network-online.target
Wants=network-online.target
[Service]
Type=simple
ExecStart=/usr/bin/python3 /home/%u/.hermes/scripts/agent-cpr.py %u --nats nats://10.3.10.55:4222
Restart=always
RestartSec=5
StandardOutput=journal
StandardError=journal
[Install]
WantedBy=default.target
systemctl --user daemon-reload
systemctl --user enable agent-cpr
systemctl --user start agent-cpr
For user clomp on a remote host where XDG_RUNTIME_DIR isn't set automatically:
sudo -u clomp XDG_RUNTIME_DIR=/run/user/1001 systemctl --user start agent-cpr
Send a ping from the command line:
from nats import connect
import asyncio, json
async def ping(agent: str):
nc = await connect("nats://10.3.10.55:4222")
reply = f"agents.cpr.reply.{agent}.ping.test"
future = asyncio.get_event_loop().create_future()
sub = await nc.subscribe(reply, cb=lambda m: future.set_result(json.loads(m.data.decode())))
await nc.publish(f"agents.cpr.{agent}", json.dumps({
"command": "ping", "params": {}, "source": "test", "reply_to": reply
}).encode())
result = await asyncio.wait_for(future, timeout=5)
await sub.unsubscribe()
await nc.close()
return result
print(asyncio.run(ping("rune")))
print(asyncio.run(ping("clomp")))
ping their CPR daemonstatus to diagnose, restart-gateway to bounce itkill-session {session_id}