advanced-skill-creator

Warn

Audited by Gen Agent Trust Hub on Mar 31, 2026

Risk Level: MEDIUMPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it generates new executable agent instructions by interpolating unvalidated user input into templates. This allows a malicious user to inject instructions that could override the behavior of the generated skill.
  • Ingestion points: Phase 1 (Requirements Analysis) and Phase 3 (Template Instantiation) capture user input for domain and parameters.
  • Boundary markers: The output template lacks delimiters to separate user content from system instructions.
  • Capability inventory: Phase 5 utilizes file-writing capabilities to save skills to the local filesystem.
  • Sanitization: No sanitization of user-provided strings is mentioned.
  • [COMMAND_EXECUTION]: The skill instructs the agent to create directories and write files using paths derived from user input (target-plugin and skill-name), creating a path traversal risk if the platform does not enforce strict directory constraints.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 31, 2026, 09:09 AM
Security Audit — agent-trust-hub — advanced-skill-creator