Back to skill

Security audit

ai-staff-agent

Security checks for vulnerabilities and agentic risk

Overview

This skill mostly does what it claims, but it gives broad file-writing and research instructions with weak path and consent controls.

Install only if you are comfortable with a broad production assistant that may create project files, cleaned datasets, research documents, and reusable scripts. Review target paths before use, especially the hard-coded /home/z/my-project/download/ locations, and ask it to confirm before writing or overwriting files.

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
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (6)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The activation guidance is overly broad and repeatedly says to use this skill for many common task classes, including 'any task' involving production-ready utility output. In an agentic system, loose trigger boundaries can cause inappropriate auto-selection, leading the agent to perform file-writing, data transformation, or web research actions in situations where a narrower or safer skill should be used.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The 'When to Use' section is framed broadly enough that many ordinary requests could match, without strong boundaries or approval gates. This increases the chance of misrouting user intent to a powerful production agent that can create files and perform multi-step actions the user did not explicitly authorize.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill instructs the agent to save generated code and scripts to project and skill directories, creating or modifying files without an explicit warning or consent checkpoint. In practice, this can lead to unintended workspace changes, overwriting files, or persistence of artifacts in sensitive environments.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The research workflow directs the agent to gather web sources and save synthesized documents to the user's workspace without privacy or data-handling safeguards. This is risky because research requests may contain sensitive topics, and storing results locally can create persistent records or mix external content into a workspace without the user's awareness.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The output conventions hard-code write locations for cleaned data and research documents, including fixed filesystem paths, without warning the user. Fixed-path writes are dangerous because they can cause unintended data persistence, collisions with existing files, and silent modification of local state beyond what the user expected.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The HTML embeds absolute local filesystem paths such as '/home/z/my-project/...' inside user-visible evaluation output. Exposing internal path structure leaks environmental details that can aid reconnaissance, reveal usernames/project layout, and sometimes disclose sensitive mount points or workspace conventions that make follow-on attacks easier. In this skill context, the page is meant to review agent outputs, so rendering raw eval artifacts increases the chance these paths are shown to users or shared externally.

Content

No source excerpt is available for this finding.