smart-fork

Fail

Audited by Gen Agent Trust Hub on Jun 24, 2026

Risk Level: HIGHCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [COMMAND_EXECUTION]: The skill configures an automated PostToolUse hook that executes a shell script located at ~/.claude/hooks/smart-memory-search.sh whenever the Task tool is used. This allows for background code execution without explicit user per-action approval.
  • [DATA_EXFILTRATION]: The skill searches across sensitive local directories, including the user's personal Obsidian vault (~/Documents/Obsidian/MiVault/), session handoffs (~/.ralph/handoffs/), and continuity ledgers (~/.ralph/ledgers/). Accessing these files containing personal notes and session history represents a high risk of sensitive data exposure.
  • [COMMAND_EXECUTION]: The skill's security notes acknowledge a command injection surface where user-provided keywords are passed to grep -E. This confirms that untrusted input flows directly into shell commands within the background scripts.
  • [COMMAND_EXECUTION]: Troubleshooting instructions guide the user to manually execute the hook script using bash and grant it execution permissions (chmod +x), which can bypass platform-level safety constraints for tool execution.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Jun 24, 2026, 08:21 PM
Security Audit — agent-trust-hub — smart-fork