Back to skill

Security audit

Generate Social Media Content Calendar

Security checks for vulnerabilities and agentic risk

Overview

The skill's social-media purpose is clear, but it asks for broad shell and file-reading tools that are not needed for creating captions, hashtags, calendars, or images.

Review this skill before installing because it enables local file reads and shell commands despite only needing content-generation capabilities. There is no evidence of automatic misuse, but install it only in a sandboxed environment or after removing those permissions.

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

Unnecessary Shell Execution and Local File Read Permissions

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:8
Vulnerability Type: Excessive tool permissions violating least privilege
Risk Level: Medium

Complete Code Snippet:

yaml
allowed-tools: Bash, Read

Technical Analysis

The skill is designed to generate social-media captions, hashtags, calendars, and images. Its documented workflow does not require arbitrary shell command execution or unrestricted access to local files. Nevertheless, the configuration grants access to both Bash and Read.

These tools exceed the permissions needed for the declared functionality. If an agent follows attacker-controlled content embedded in a user request or referenced material, Read could be misused to access files available to the agent process, while Bash could execute unrelated operating-system commands. The resulting access remains subject to the operating-system privileges and sandbox restrictions of the process hosting the agent.

Attack Path

  1. A user or untrusted content source supplies a crafted social-media generation request containing instructions to inspect local files or run shell commands.
  2. The skill is loaded with Bash and Read enabled.
  3. The agent is induced to treat the malicious instructions as necessary steps in content generation.
  4. The agent invokes Read to access local data or Bash to execute commands unrelated to the skill's purpose.
  5. Retrieved information may be incorporated into generated output or shell commands may modify resources accessible to the agent process.

Impact Assessment

Successful exploitation could provide access to files readable by the agent process and permit command execution with that process's operating-system privileges. Depending on the runtime sandbox, accessible resources could include project files, configuration data, environment-derived credentials, or other local user data. Shell access may also allow file modification, network operati ...[truncated 367 chars]

Remediation
View remediation

Remediation Suggestions

  • Remove Bash and Read from allowed-tools; permit only the chat and image-generation capabilities required by the documented workflow.
  • If users must provide source files, replace unrestricted reading with an explicit user-selected file interface restricted to approved paths and file types.
  • If shell execution becomes necessary for a future feature, expose narrowly scoped operations rather than arbitrary Bash access. Enforce command allowlists, fixed arguments, path validation, timeouts, and an isolated low-privilege sandbox.
  • Require explicit user confirmation before accessing local data or performing side effects.
  • Prevent generated content and untrusted source material from being interpreted as tool-use instructions.
  • Add tests verifying that ordinary content-calendar requests cannot trigger local file access or operating-system commands.
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