ClawHub
Back to skill

Security audit

eval-integrity

Security checks for vulnerabilities and agentic risk

Overview

This skill performs a disclosed, read-only benchmark audit; its GitHub PR lookup broadens scope slightly but remains non-mutating and purpose-aligned.

Install this if you want a rigorous read-only audit of a benchmark repository. Be aware that the workflow may query GitHub PR metadata with your local gh authentication; avoid running it in environments where even read-only PR metadata access is not acceptable.

A.I.G

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs

SkillSpector

Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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
Findings (3)

Description-Behavior Mismatch

Medium
Confidence
91% confidence
Finding
The skill is presented as a read-only local repo audit, but it mandates querying GitHub PR state via `gh pr list`, which expands scope beyond the target repository. This is a capability mismatch: it can disclose or depend on remote metadata and network access not necessary for many local audits, though it does not itself execute code or modify state.

Context-Inappropriate Capability

Medium
Confidence
84% confidence
Finding
Requiring GitHub PR enumeration as a standing, non-optional step gives the skill broader access than its core benchmark-integrity function requires. In environments with sensitive remotes, private repository metadata, or restricted network policies, this unnecessary remote lookup can leak contextual information or violate least-privilege expectations.

Intent-Code Divergence

Low
Confidence
72% confidence
Finding
The documentation presents the skill as a read-only benchmark audit focused on the target repository, which suggests repo-contained inspection. Later instructions require checking open PRs via `gh pr list`, creating a mild contradiction between the narrow read-only repo-audit framing and actual behavior that reaches into external repository hosting state.

Static analysis

No suspicious patterns detected.