safety-guard
Pass
Audited by Gen Agent Trust Hub on Apr 6, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill implements protective measures such as command interception and file system freezing to ensure safe autonomous agent behavior.
- [COMMAND_EXECUTION]: Destructive command patterns like 'sudo rm' and 'chmod 777' are listed solely as monitored behaviors to be intercepted, not for execution by the skill itself.
- [DATA_EXFILTRATION]: The skill logs blocked operations to a local file path ('~/.claude/safety-guard.log') for auditing purposes; no network exfiltration of sensitive data was found.
- [PROMPT_INJECTION]: While the skill processes tool inputs (Bash, Write, Edit) which could contain indirect prompt injections, its primary purpose is defensive validation. Mandatory Evidence Chain: 1. Ingestion points: Intercepts 'Bash', 'Write', 'Edit', and 'MultiEdit' tool calls. 2. Boundary markers: Not specified in the instructions. 3. Capability inventory: Monitoring and logic-based blocking of system tool calls. 4. Sanitization: Employs pattern matching and confirmation prompts to mitigate the risk of destructive actions.
Audit Metadata