ClawHub
Back to skill

Security audit

可追溯审计团

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a document-only audit workflow, but its own integrity metadata does not match the actual SKILL.md file and its local-only claim is inconsistent with required online checks.

Review this package before installing because the shipped integrity claims do not match the actual SKILL.md file. If you use it, treat audits as potentially involving sensitive identity data and network lookups, require explicit approval before submitting identifiers to external sites, and keep report redaction enabled by default.

SkillSpector

Vulnerability Patterns
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • 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
Findings (3)

Intent-Code Divergence

Medium
Confidence
93% confidence
Finding
The skill claims it contains no data collection and operates entirely locally, yet its SOP explicitly requires online verification of standards and credentials. This creates a misleading trust boundary: users may provide sensitive identity materials believing nothing leaves the local environment, while the workflow can trigger network access and external disclosure risks.

Intent-Code Divergence

Low
Confidence
78% confidence
Finding
The skill mandates redaction of sensitive information, but the reporting structure still instructs collection and inclusion of evidence, sources, and appendices without defining enforced masking behavior. In an identity-audit context, this can result in accidental exposure of names, IDs, certificate numbers, or account identifiers in generated reports.

Natural-Language Policy Violations

Medium
Confidence
93% confidence
Finding
The file explicitly requires all public outputs to carry a specific brand name and forbids use of real names, employer names, or internal paths. In an agent skill, this can improperly constrain downstream outputs and pressure the model to inject branding regardless of user intent or deployment policy, creating attribution manipulation and possible concealment of the real origin of content.

Static analysis

No suspicious patterns detected.