tool-guard

Pass

Audited by Gen Agent Trust Hub on Apr 7, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill generates and wires shell (.sh), PowerShell (.ps1), and TypeScript (.ts) scripts to act as pre-execution hooks that intercept and validate agent terminal commands.\n- [COMMAND_EXECUTION]: Uses the chmod +x command to grant execution permissions to the generated shell scripts on the host system.\n- [COMMAND_EXECUTION]: Modifies repository configuration files, such as .claude/settings.json and .github/hooks/tool-guard.json, to register these scripts for automated execution during agent activity.\n- [PROMPT_INJECTION]: Creates a surface for indirect prompt injection where a repository's policy.json file can include instructions within the 'reason' fields. These strings are returned to the agent context when a command is blocked or triggers a warning, potentially influencing the agent's behavior.\n
  • Ingestion points: hooks/tool-guard/policy.json (read by hooks at runtime).\n
  • Boundary markers: Absent; the reason string is interpolated directly into the platform's JSON response to the agent.\n
  • Capability inventory: Terminal command monitoring and blocking capabilities across multiple platforms.\n
  • Sanitization: None; feedback messages are passed to the agent as raw strings.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 7, 2026, 09:28 AM
Security Audit — agent-trust-hub — tool-guard