ClawHub
Back to skill

Security audit

Multi Agent

Security checks for vulnerabilities and agentic risk

Overview

This is a documentation-only multi-agent coordination guide; its examples can send project context to agent CLIs and create coordination files, but those behaviors are disclosed and fit the stated purpose.

Install this if you want design guidance for multi-agent coding workflows. When applying the examples, use a trusted version-controlled workspace, limit worker tool permissions, inspect `.coordination`, `.claims`, `.worktrees`, and any mailbox files, and redact secrets or proprietary material before passing synthesis documents, task files, or diffs to external agent/model CLIs.

SkillSpector

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Rogue AgentSelf-Modification, Session Persistence
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The example passes the entire contents of `.coordination/synthesis.md` directly into a `claude -p` prompt, which can transmit sensitive project data, secrets, internal findings, or proprietary code context to an external model endpoint without any warning, minimization, or redaction step. In this multi-agent coordination context, the synthesis document is specifically designed to aggregate research results, which makes it more likely to contain concentrated sensitive information than an ordinary prompt.

Session Persistence

Medium
Category
Rogue Agent
Content
}
```

四状态：`pending → claimed → done/failed`。领取任务通过原子锁 `mkdir ".coordination/tasks.lock"`（POSIX 上 mkdir 是原子操作），10 次重试，指数退避 50-500ms，释放用 `rmdir`。

Claude Code Swarm 使用 per-agent inbox `~/.claude/teams/{team}/inboxes/{agent}.json` 替代共享 tasks.json，减少锁竞争。 → [详见](references/task-coordination.md)
Confidence
80% confidence
Finding
The skill recommends per-agent inbox files under '~/.claude/teams/{team}/inboxes/{agent}.json', which introduces persistent state outside the project workspace. Persistent cross-session artifacts can leak task contents, coordination data, or stale instructions into later runs, especially on shared machines or when team/agent names are reused.

Static analysis

No suspicious patterns detected.