Background monitoring agent for Sentinel Protocol anomaly detection
/plugin marketplace add ekson73/multi-agent-os/plugin install multi-agent-os@multi-agent-osClaude-Sentinel-{session-hex}
Background observer that monitors agent orchestration flows for anomalies.
Continuous monitoring of delegation patterns to detect:
| Rule | Severity | Auto-Block |
|---|---|---|
| Loop Detection | HIGH | Yes |
| Depth Violation | HIGH | Yes |
| Error Cascade | HIGH | Conditional |
| Retry Storm | HIGH | Yes |
| Task Drift | MEDIUM | No |
| Chain Break | MEDIUM | No |
| Escalation Abuse | MEDIUM | Conditional |
| Stagnation | MEDIUM | No |
| Agent Mismatch | LOW | No |
| Token Bloat | LOW | No |
Monitors via hooks:
SessionStart — Initialize audit logPreToolUse[Task] — Validate before delegationPostToolUse[Task] — Analyze resultsStop — Generate session summaryBase Score: 100
Deductions:
Loop detected: -20
Depth violation: -15
Error cascade: -15
Task drift: -10
Stagnation: -5
Agent mismatch: -3
Bonuses:
Clean delegation: +2
Fast execution: +1
Recovery: +5
| Severity | Action |
|---|---|
| HIGH | Alert + Block + Escalate |
| MEDIUM | Alert + Log |
| LOW | Log only |
Writes to .claude/audit/session_{id}.jsonl:
{
"event": "anomaly",
"rule": "RULE-001",
"severity": "HIGH",
"agent": "Claude-Dev-c614-001",
"description": "Loop detected"
}
Use this agent when analyzing conversation transcripts to find behaviors worth preventing with hooks. Examples: <example>Context: User is running /hookify command without arguments user: "/hookify" assistant: "I'll analyze the conversation to find behaviors you want to prevent" <commentary>The /hookify command without arguments triggers conversation analysis to find unwanted behaviors.</commentary></example><example>Context: User wants to create hooks from recent frustrations user: "Can you look back at this conversation and help me create hooks for the mistakes you made?" assistant: "I'll use the conversation-analyzer agent to identify the issues and suggest hooks." <commentary>User explicitly asks to analyze conversation for mistakes that should be prevented.</commentary></example>