Back to skill

Security audit

Generate Home Remodeling Company Client Education Handout

Security checks for vulnerabilities and agentic risk

Overview

This skill appears to be a simple handout generator, but it requests broad shell and file-read access that does not fit that purpose.

Install only if you are comfortable with this skill being granted broad local command and file-read capability. The visible instructions look benign, but the requested tools are broader than needed for making a handout; a safer version would remove Bash and Read or strictly limit file access to explicit user-provided inputs.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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
Findings (1)

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:11
Finding

Unnecessary Shell Execution and File-Read Permissions

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, line 11
Vulnerability Type: Excessive tool permissions violating least privilege
Risk Level: Medium

Vulnerable Code

yaml
description: "Create a polished explainer handout with visuals, FAQs, and clear next steps for a home remodeling company. Use when handling client education work for..."

> 📖 **Complete setup guide**: https://skillboss.co/skill.md
allowed-tools: Bash, Read

Technical Analysis

The skill is intended to generate a client education handout, and its declared APIs are chat and image_generation. This content-generation purpose does not inherently require arbitrary shell command execution or access to local files.

Declaring Bash and Read therefore expands the skill's authority beyond its documented functional requirements. If the host platform enforces allowed-tools as an authorization boundary, the skill may be able to execute operating-system commands and inspect files accessible to the agent process. This increases the consequences of prompt injection, malicious user input, or unsafe content incorporated into the generation workflow.

No instruction in the reviewed files directly attempts to exploit these permissions. The vulnerability is the unnecessary exposure of privileged tools, rather than evidence of active command execution or data theft.

Attack Path

  1. A user or an untrusted content source supplies crafted instructions while the skill is active.
  2. The injected instructions induce the agent to invoke Read or Bash, despite those tools not being required for handout generation.
  3. Read is used to inspect files available to the agent process, or Bash is used to execute local commands.
  4. Data obtained through those tools could be incorporated into generated output or otherwise exposed. Shell commands could also modify files or invoke other locally available programs, subject to the host process's operating-system permissions and sandbo ...[truncated 840 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove Bash and Read from allowed-tools.
  2. Grant only the content-generation capabilities required by the documented workflow, such as chat and image_generation.
  3. If file input becomes necessary, replace unrestricted file access with a narrowly scoped mechanism limited to explicit user-selected files and approved directories.
  4. If command execution later becomes essential, expose a purpose-built tool with fixed arguments instead of a general shell. Require explicit user confirmation and execute it in a restricted sandbox.
  5. Add runtime controls that deny access to secrets, home directories, credentials, and unrelated project files.
  6. Validate tool use against the skill's documented purpose and reject tool requests originating from untrusted handout content or embedded instructions.

A minimized configuration should resemble:

yaml
allowed-tools: []

Alternatively, list only the exact generation tools required by the host platform, without enabling general shell or filesystem access.

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