meta-cognition-parallel

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION] (LOW): Susceptibility to indirect prompt injection through raw interpolation of user arguments into the prompts of parallel sub-agents.\n
  • Ingestion points: $ARGUMENTS are parsed in Step 1 and used in Step 2 to build sub-agent prompts.\n
  • Boundary markers: Minimal delimiters (a markdown header '## User Query') are used, which do not sufficiently isolate untrusted data from the instruction context.\n
  • Capability inventory: The skill coordinates sub-agents; it lacks direct file-system, command execution, or network capabilities itself, which limits the potential impact of an injection.\n
  • Sanitization: No sanitization or escaping of the user-provided $ARGUMENTS is implemented.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 06:23 PM