meta-cognition-parallel

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input from the $ARGUMENTS variable and interpolates it directly into prompts intended for sub-agents or internal analysis sections.
  • Ingestion points: User-provided Rust questions enter the agent context through the $ARGUMENTS parameter in SKILL.md.
  • Boundary markers: The skill uses Markdown headers (e.g., ## User Query) to separate user content from the analyzer's instructions, though it lacks strict enforcement or "ignore previous instructions" safety guardrails.
  • Capability inventory: The skill is limited to reading local agent configuration files and outputting text; it does not have capabilities for network operations, file writing, or system command execution.
  • Sanitization: No explicit filtering or sanitization is performed on the user input before it is used to construct sub-agent prompts.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 11:57 AM
Security Audit — agent-trust-hub — meta-cognition-parallel