meta-prompting

Pass

Audited by Gen Agent Trust Hub on Sep 20, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill functions as a meta-reasoning layer that processes user input to apply various analysis patterns. This creates a surface where malicious instructions embedded in the user-provided data could attempt to influence the reasoning output.
  • Ingestion points: The skill processes user queries through structured commands (/adversarial, /verify, etc.) and auto-triggers on specific keywords.
  • Boundary markers: The skill instructions specify the use of --- to separate combined pattern outputs, providing some structural delimitation.
  • Capability inventory: The runtime instructions in SKILL.md do not define any dangerous capabilities such as file writes, network requests, or shell command execution.
  • Sanitization: No explicit content sanitization or instruction-ignoring delimiters are defined for the data processed by the patterns.
  • [COMMAND_EXECUTION]: The SPEC.md file contains documentation for running evaluation scripts (e.g., python3 distillery/scripts/distiller.py validate-plugin). These are standard maintenance procedures for developers and are not invoked by the agent during the execution of the skill's primary functions.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 20, 2026, 08:34 AM
Security Audit — agent-trust-hub — meta-prompting