roast-my-agents-md

Warn

Audited by Gen Agent Trust Hub on Mar 14, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes shell commands to perform its primary auditing and A/B testing functions.\n
  • Evidence: SKILL.md Phase 2 Step 7 contains instructions for the agent to move files in sensitive global configuration locations such as ~/.claude/, ~/.codex/, and ~/.cursor/ to create clean environments for testing.\n
  • Evidence: It uses the find command with -exec to locate and rename all "memory" directories across the user's projects.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted, user-controlled configuration files.\n
  • Ingestion points: Reads instructions from AGENTS.md, CLAUDE.md, and other AI configuration files located via recursive search in the project directory (SKILL.md Phase 1 Step 0).\n
  • Boundary markers: Absent; audited content is directly analyzed and utilized in the creation of evaluation prompts for the A/B testing phase.\n
  • Capability inventory: The skill can execute Node.js scripts, perform significant filesystem operations, and spawn new agent instances.\n
  • Sanitization: No sanitization of the audited content is performed before processing or prompt interpolation.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 14, 2026, 02:02 PM