driving-claude-code-sessions
Fail
Audited by Gen Agent Trust Hub on May 14, 2026
Risk Level: HIGHCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONCREDENTIALS_UNSAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill relies on executing various shell scripts (e.g., launch-worker.sh, converse.sh, stop-worker.sh) to orchestrate processes and manage tmux sessions.
- [REMOTE_CODE_EXECUTION]: The skill launches worker sessions with the --dangerously-skip-permissions flag. This explicitly disables the platform's standard safety controls, allowing automated agents to execute arbitrary code and tools without user review or intervention.
- [CREDENTIALS_UNSAFE]: Instructions direct the agent to access the ~/.claude/projects/ directory to read session log files (JSONL). These logs frequently contain sensitive data, environment variables, and API tokens processed during previous AI sessions.
Recommendations
- AI detected serious security threats
Audit Metadata