The Agent Skills Directory

[PERSISTENCE]: The skill configures the environment to execute a custom script for every subsequent tool use. Specifically, it instructs the agent to create an executable script in ~/.claude/hooks/ or .claude/hooks/ and modify settings.json to register this script as a PreToolUse hook. While this creates a persistent change to the agent's execution flow, it is the primary intended function of the skill to provide ongoing git guardrails.
[COMMAND_EXECUTION]: The skill involves the creation and execution of a bash script (scripts/block-dangerous-git.sh). The script itself uses jq and grep to analyze command strings. It does not execute the input it receives; rather, it exits with a non-zero status code to block execution by the primary agent if a dangerous pattern is detected.

git-guardrails-claude-code