Back to skill

Security audit

Generate Pediatric Dental Clinic Client Education Handout

Security checks for vulnerabilities and agentic risk

Overview

The skill appears to be a simple pediatric dental handout generator, but it requests broad shell and file-reading tools that are not explained by that purpose.

Review this skill before installing because its content-generation purpose does not explain why it needs shell command execution or broad local file reads. Prefer a version that removes Bash and Read or limits file access to explicit templates, and use human review for any patient-facing dental or health claims.

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:8
Finding

Excessive Shell and File-Read Permissions

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, line 8
Vulnerability Type: Excessive tool permissions and violation of least privilege
Risk Level: Medium

Vulnerable Code Snippet:

yaml
allowed-tools: Bash, Read

Technical Analysis

The skill is intended to generate a pediatric dental clinic education handout. Its documented workflow involves drafting content and potentially generating images, but it does not describe any legitimate requirement to execute shell commands or read arbitrary local files.

Declaring both Bash and Read therefore grants capabilities beyond the skill's stated purpose. In a host environment that enforces this declaration by making those tools available, Bash may permit command execution with the privileges of the agent process, while Read may expose files accessible to that process. If the skill's execution is subsequently influenced by malicious prompts, untrusted external content, or unsafe generated instructions, these excessive permissions could become an access and command-execution channel.

Attack Path

  1. A user installs and loads the skill.
  2. The host grants the declared Bash and Read tools to the skill.
  3. Malicious or otherwise untrusted instructions influence the skill during execution.
  4. The influenced workflow invokes Read to access local files or Bash to execute shell commands unrelated to handout generation.
  5. Commands and file reads run within the operating-system permissions and sandbox boundaries of the agent process.

No malicious invocation is present in the audited files themselves; exploitation depends on the host granting these declared permissions and on later instruction influence.

Impact Assessment

Potential impact includes unauthorized disclosure of files available to the agent process and execution of commands under that process's account. The precise scope depends on host sandboxing, tool-level policy enforcement ...[truncated 202 chars]

Remediation
View remediation

Remediation Suggestions

  • Remove Bash and Read because neither is required by the documented handout-generation workflow.
  • Allow only narrowly scoped content-generation and image-generation capabilities that are necessary for the skill's stated purpose.
  • If local templates are essential, replace general file access with a restricted reader limited to a fixed, read-only template directory.
  • If command execution later becomes necessary, expose specific validated operations rather than a general shell.
  • Enforce runtime sandboxing, deny access to credentials and sensitive filesystem locations, and require explicit user approval for privileged actions.
  • Add automated validation that rejects skill manifests requesting tools not justified by their documented workflows.
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
93% confidence
Finding

This trigger text is incomplete and overly broad, making it unclear exactly when the skill should or should not be used. Because it lacks specific activation constraints or negative examples, it could match a wide range of ordinary client-education requests and lead to unintended invocation.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The phrase 'Use when handling client education work for...' does not provide a specific trigger phrase or clear activation boundary. This ambiguity increases the chance of accidental invocation for general education or content-writing tasks outside the intended domain.

Content

No source excerpt is available for this finding.