multi-agent-orchestrator

Warn

Audited by Socket on Nov 21, 2026

2 alerts found:

Anomalyx2
AnomalyLOW
SKILL.md

SUSPICIOUS: the skill's observer function coherently explains transcript review, model calls, and hook corrections, and there is no evident malware, hidden execution, or untrusted binary installer. However, it handles sensitive session evidence and credentials, permits an arbitrary API endpoint that could receive both, and can publish correction data to a configurable feed. Use only with trusted configuration and explicit consent to transmit session data.

Confidence: 92%Severity: 66%
AnomalyLOW
references/fleet-babysitter.md

This readable agent configuration directs an assistant to inspect scoped agent sessions and send corrective messages, and sets permissionMode to bypassPermissions. It exposes a meaningful authorization and control-plane risk if used without appropriate host-level safeguards, but this fragment contains no executable malware, credential theft, or external exfiltration destination. The referenced ~/.claude/agents/sami-proxy.md, ~/.claude/agents/autonomous-closer.md, references/guardian-policy.md, and scripts/intent_guardian.py require separate review to assess their behavior.

Confidence: 96%Severity: 57%
Audit Metadata
Analyzed At
Nov 21, 2026, 01:03 PM
Package URL
pkg:socket/skills-sh/samihalawa%2Fmulti-agent-orchestrator-skill%2Fmulti-agent-orchestrator%2F@8a99c481de59892b502fd96625816c22c6230d61