Summary

Safety Guard is a protective skill that intercepts and gates destructive operations through three modes: Careful (warns on dangerous commands like rm -rf and git push --force), Freeze (locks writes to a specific directory), and Guard (combines both). It works via PreToolUse hooks to inspect Bash, Write, and Edit tool calls before execution, making it useful for autonomous agents and production environments.

Patterns detected by deterministic static analysis before AI scoring. Hover over any finding code for detailed information and remediation guidance.

Detected Capabilities

Trigger Keywords

Phrases that agents use to match this skill to user intent.

Risk Signals

Use Cases

• Restrict autonomous agent edits to a single directory during development
• Prevent accidental destructive commands (rm -rf, git force push) in production environments
• Gate dangerous operations during sensitive migrations or deployments
• Add confirmation workflows for high-risk git and database operations
• Lock down test environments to prevent unintended cross-project changes

Quality Notes

• Skill is defensive in nature: it describes *blocking* dangerous patterns, not executing them — this is the correct approach
• Clear three-mode design with concrete examples of activation syntax (/safety-guard freeze, /safety-guard guard)
• Well-defined scope: operates via PreToolUse hooks on Bash, Write, Edit, MultiEdit tools only
• Good practical use cases (production, autonomous agents, directory locking)
• Missing: no documentation of what 'confirmation' actually looks like or how user provides it
• Missing: no edge cases documented (e.g., what happens if a tool call matches multiple patterns, or if user ignores confirmation)
• Missing: no examples of blocked vs. allowed commands for clarity
• Missing: error handling and fallback behavior if hooks fail or are misconfigured