audit-instructions

Pass

Audited by Gen Agent Trust Hub on Jul 30, 2026

Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes local helper scripts instruction-scan.sh and conflict-scan.sh using bash. These scripts utilize standard POSIX tools (grep, awk, sort) to perform pattern matching on instruction files. All command execution is local, deterministic, and used for the primary purpose of auditing file content.
  • [SAFE]: The skill implements a 'Read-only contract', explicitly stating it never auto-applies changes or performs write operations on audited files. All findings are presented as proposals gated by the user.
  • [SAFE]: The skill includes a 'Verify pass' which uses subagents to self-grade its own proposals, reducing the risk of over-aggressive or incorrect instruction removal.
  • [SAFE]: The skill follows least-privilege principles by restricting the metadata it extracts from configuration files (e.g., extracting only prompt text and explicitly avoiding secrets or tokens in settings.json).
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 30, 2026, 07:54 PM
Security Audit — agent-trust-hub — audit-instructions