Command Development

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFE
Full Analysis
  • [Indirect Prompt Injection] (SAFE): The skill provides documentation on creating commands that ingest external data through arguments and file references. \n
  • Ingestion Points: Positional arguments ($1, $2) and file inclusion (@path) are demonstrated in multiple examples in simple-commands.md. \n
  • Boundary Markers: While templates lack explicit runtime delimiters, the documentation emphasizes the use of argument-hint for structure. \n
  • Capability Inventory: Examples demonstrate restricted bash tool usage (git, npm, kubectl) rather than broad access. \n
  • Sanitization: The reference guide recommends using the most restrictive allowed-tools settings and provides a manual-approval flag (disable-model-invocation) for destructive or sensitive operations.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 04:37 PM