Back to skill

Security audit

Generate Independent Insurance Agency Client Education Handout

Security checks for vulnerabilities and agentic risk

Overview

This handout generator shows no malicious payload, but it requests broad shell and file-reading access that is not justified by its stated purpose.

Review this skill carefully before installing. It appears to be a simple insurance handout generator and I found no malicious scripts or data-exfiltration instructions, but it should not need shell-command execution or broad local file access. Prefer installing only if the platform can restrict or confirm Bash and file-read operations, or if the publisher narrows the declared tools.

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:12
Finding

Excessive Bash and File-Read Permissions

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, line 12
Vulnerability Type: Excessive tool permissions that violate least privilege
Risk Level: Medium

Vulnerable Code Snippet:

yaml
allowed-tools: Bash, Read

Technical Analysis

The skill is documented as generating an insurance-agency client education handout. Its workflow consists of clarifying content requirements, drafting the handout, enriching supporting material, and refining the output. This documented purpose does not require arbitrary shell-command execution.

Granting access to Bash creates a general local command-execution capability, while Read may permit access to files unrelated to the requested handout. These permissions exceed the minimum capabilities required by the stated workflow. If attacker-controlled task content or referenced source material influences the agent, it could attempt to induce the agent to use these tools outside the intended scope.

The finding concerns excessive authorization in the skill configuration. The reviewed files do not contain an explicit malicious command or an instruction that directly exploits these permissions.

Attack Path

  1. A user or external content supplies adversarial instructions as part of a handout-generation request.
  2. The skill is loaded with both Bash and Read available.
  3. The adversarial content instructs the agent to inspect unrelated local files or execute shell commands under the pretext of collecting supporting information.
  4. If the agent follows those instructions without adequate runtime authorization controls, it invokes Read or Bash.
  5. The operation runs with the filesystem access and operating-system privileges of the host process.

Impact Assessment

Successful exploitation could allow unauthorized reading of files accessible to the agent process and execution of local commands with that process's privileges. Depending on the runtime environmen ...[truncated 545 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove Bash from allowed-tools, because arbitrary command execution is not necessary for generating an educational handout.
  2. Remove Read if the skill is expected to work exclusively from information supplied in the conversation.
  3. If source-document reading is required, restrict access to explicitly selected, user-approved files or a dedicated workspace directory.
  4. Require per-operation user confirmation for access outside the active project or uploaded-document directory.
  5. Apply runtime sandboxing, filesystem allowlists, command allowlists, and outbound-network restrictions as defense-in-depth controls.
  6. Document every required tool and the specific workflow step that needs it, rejecting permissions that cannot be justified by the skill's stated function.

A hardened declaration should omit unnecessary tools, for example:

yaml
allowed-tools: []

If the platform requires a different syntax for declaring no tools, use its supported empty or omitted configuration.

Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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 (2)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The 'Use When' guidance says to use the skill for 'client education work for insurance agency owners, account managers,' which is a broad class of everyday work rather than a narrow, clearly bounded trigger. It does not provide specific activation phrases, scope limits, or exclusion examples, increasing the chance of unintended invocation.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The suggested prompt '[Generate Independent Insurance Agency Client Education Handout] for my insurance agency' is still a broadly reusable request pattern without constraints on audience, handout type, or context. Because the manifest and markdown do not narrow invocation with negative examples or context restrictions, the trigger remains vague.

Content

No source excerpt is available for this finding.