The Agent Skills Directory

[SAFE]: The skill is entirely safety-focused, providing mechanical enforcement instructions to prevent destructive operations such as unintended file deletions or database drops.
[COMMAND_EXECUTION]: The skill utilizes a specialized CLI tool 'dcg' for command validation and safety checks. Static analysis detections regarding destructive commands (e.g., rm -rf /) are false positives, as these patterns appear exclusively within documentation examples and test cases that describe the tool's blocking capabilities.
[PROMPT_INJECTION]: The instructions contain strict anti-circumvention directives, explicitly forbidding the agent from bypassing the guardrail or requesting overrides without first identifying safer alternatives.
[EXTERNAL_DOWNLOADS]: The documentation references an update mechanism (dcg update) and installation scripts to ensure the safety tool remains current, pointing to official repository sources.

dcg