Back to skill

Security audit

postey-voice

Security checks for vulnerabilities and agentic risk

Overview

The voice-learning workflow can create drafts and upload media to Postey without a clear request to save content there.

Review before installing. Require an explicit request before saving drafts or uploading media to Postey; confirming sample hooks should only confirm the voice. Choose the target account, local writing folder, profile storage, and output filename deliberately, since existing output files can be overwritten. Profiles and ledgers persist across sessions and may contain exact draft text and feedback. All bundled files were inspected, including the scanner's script coverage gaps; the separate postey dependency and missing referenced templates were not available for inspection.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • System Prompt LeakageDirect Leakage, Indirect Extraction, Tool-Based Exfiltration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (8)

Ae3

High
Category
analysis-evasion
Confidence
90% confidence
Finding

Text artifact contains embedded NUL bytes

Content

No source excerpt is available for this finding.

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · scripts/voiceRules.js (reported line 80)May include surrounding context.

js
a.scope.localeCompare(b.scope)
  );

  return rules;
}

/** Only `active` rules constrain drafting. Candidates are questions, not law. */

Sharing any website or social handle can activate the brand-voice workflow without an explicit branding request.

Medium
Category
Trigger Abuse
Confidence
94% confidence
Finding

Merely sharing a website or handle is treated as authorization to start a workflow with persistence and remote-write side effects. A user sharing a link for an unrelated question has not necessarily requested brand profiling or draft creation.

Content

No source excerpt is available for this finding.

The workflow retains a researched brand profile across sessions without requiring user-facing disclosure.

Medium
Category
Output Handling
Confidence
92% confidence
Finding

The workflow mandates cross-session retention in an unspecified persistence mechanism, even though its broad activation conditions do not establish that the user requested lasting profiling. Persistent brand profiles are appropriate for this skill's purpose, but undisclosed storage can retain inferred information beyond the intended task or account context.

Content

No source excerpt is available for this finding.

The voice-learning flow also creates and modifies remote content drafts without requiring a separate drafting request.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

A request to learn the user's voice can lead directly to creating, modifying, uploading, and tagging content in a remote account without a request to create drafts. Keeping posts in DRAFT limits public exposure but does not authorize these remote writes or associated transfers.

Content

No source excerpt is available for this finding.

Draft content and media are sent to Postey without a required disclosure before transfer.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

This flow sends generated content and potentially media to Postey even when the initiating request may only concern learning a voice. Postey is the intended integration, not an apparent attacker endpoint, but connecting an account does not by itself authorize every subsequent content transfer.

Content

No source excerpt is available for this finding.

The stated working-directory containment for --out is not enforced.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

L114-L116 describe an earlier unrestricted output-path behavior and state that output should be contained to the working directory. The implementation merely emits a TTY-only note for an outside path, then uses the unrestricted path for the write at L177; L117-L119 also explicitly contradict the containment claim.

Content

No source excerpt is available for this finding.

The CLI writes files directly despite claiming that every write goes through MCP.

Medium
Category
Excessive Agency
Confidence
100% confidence
Finding

The module documentation at L005-L008 says the CLI only reads files, creates nothing, and leaves every write to the agent through MCP. However, L176-L178 directly creates or overwrites a local profile file with fs.writeFileSync whenever --out or --account selects an output filename.

Content

No source excerpt is available for this finding.