Back to skill

Security audit

Yummy Gen Voice

Security checks for vulnerabilities and agentic risk

Overview

This TTS skill is mostly coherent, but it needs review because its shell command examples can mishandle user-provided text and it installs an unpinned executable dependency.

Install only if you trust the yummycli npm package and are comfortable providing Gemini credentials. Agents using this skill should pass text, speaker names, voices, and output paths as literal process arguments rather than constructing shell strings, and should choose an explicit output path for sensitive audio.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (2)

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:49
Finding

Shell Command Injection Through User-Controlled Command Arguments

Content
View full analysis
" ``` ```bash yummycli gemini speak \ --text "" \ --voice Kore \ --output narration.wav ``` ### Technical Analysis The documented command templates place user-controlled narration text directly inside shell commands. Although the placeholder is enclosed in double quotes, double-quoted Bash strings still process command substitutions such as `$(command)` and backtick expressions when those expressions become part of the command text parsed by the shell. The output path may also be user-controlled. If an agent constructs a shell command as a string by replacing these placeholders, malicious quotation marks, command substitutions, redirections, or shell metacharacters could alter the intended command. This issue applies when the agent generates or executes a shell command from the documented template. It can be avoided by invoking the executable directly with a structured argument array and without a shell. ### Attack Path 1. An attacker requests speech synthesis using malicious text, such as text containing `$(malicious-command)` or a quotation mark followed by shell syntax. 2. The agent substitutes the supplied text into the documented Bash template. 3. The resulting command string is passed to a shell for execution. 4. The shell interprets the injected command substitution or metacharacters before or alongside execution of `yummycli`. 5. The injected command runs with the operating-system privileges and environment available to the agent. ### Impact Assessment Successful exploitation could allow arbitrary local command execution with the agent's privileges. Depending on the execution environment, this could permit: - Reading or modifying files accessible to the agent. - Accessing environment vari ...[truncated 389 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
SKILL.md:18
Finding

Unpinned Executable npm Dependency

Content
View full analysis
Remediation
View remediation
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)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill explicitly describes generating .wav files and notes that, when --output is omitted, a timestamped audio file is written to the current working directory, but it does not prominently warn that this creates a persistent disk artifact. In agentic contexts, this can lead to unexpected local file creation, leakage of sensitive synthesized content into shared workspaces, or storage of files in directories the user did not intend.

Content

No source excerpt is available for this finding.

Vague Triggers

Low
Category
Not specified by scanner
Confidence
81% confidence
Finding

The 'When to Use' line lists generic requests such as 'convert text to audio' and 'generate a voiceover' as activation cues, but it does not include exclusion conditions or negative examples. In a markdown skill description, this can make invocation boundaries less precise and increase the chance of unintended activation for adjacent audio-editing or non-Gemini workflows.

Content

No source excerpt is available for this finding.