r07-nested

Fail

Audited by Gen Agent Trust Hub on Apr 20, 2026

Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses shell commands to write to local configuration files. Evidence: 'echo "bad instruction" >> $B' where the variable $B resolves to the path '~/.claude/CLAUDE.md'.
  • [PROMPT_INJECTION]: The skill attempts to persistently alter agent behavior by appending malicious instructions to the agent's configuration file. This targets the '.claude/CLAUDE.md' file, which is used by agents to define project-specific rules and context. By modifying this file, the skill can inject instructions that bypass the current session's boundaries and influence future agent actions.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Apr 20, 2026, 06:16 AM