reviewing-command-definitions

Pass

Audited by Gen Agent Trust Hub on Sep 7, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze user-authored command files which may contain malicious instructions.\n
  • Ingestion points: Reads files from commands/**/*.md and .claude/prompts/**/*.md using Read, Grep, and Glob tools.\n
  • Boundary markers: Employs explicit instructions directing the agent to treat input as data and ignore embedded directives or instructions within the analyzed files.\n
  • Capability inventory: Access is restricted to read-only tools (Read, Grep, Glob), preventing file modification or direct shell execution.\n
  • Sanitization: Relies on strict meta-instruction safety constraints to separate analyzed content from agent instructions.\n- [COMMAND_EXECUTION]: The skill contains example shell command strings like rm -rf ~ for educational purposes to help identify injection vulnerabilities. These strings are static parts of the documentation and are not invoked by the skill itself.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 7, 2026, 06:11 PM
Security Audit — agent-trust-hub — reviewing-command-definitions