ClawHub
Back to skill

Security audit

Dispatching Parallel Agents

Security checks for vulnerabilities and agentic risk

Overview

This is a documentation-review skill whose broad repo reading and optional agent-instruction edits are disclosed and aligned with its purpose.

Install if you want an agent to audit or improve repository documentation and agent-governance files. Expect it to read broadly across docs, README files, AGENTS/CONTRIBUTING files, and related aliases; use report-only mode or a narrow file scope if you do not want it to edit documentation or compatibility pointers.

SkillSpector

Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (1)

Missing User Warnings

Medium
Confidence
86% confidence
Finding
The skill encourages parallel dispatch and mentions conflicts only after agents return, but it does not prominently warn users before dispatching about concurrent edits to the same files, shared resources, or other interference risks. In practice, this can lead to race conditions in code changes, inconsistent workspace state, flaky test results, or accidental overwrites when multiple agents operate simultaneously.

Static analysis

No suspicious patterns detected.