The Agent Skills Directory

[COMMAND_EXECUTION] (LOW): The skill provides instructions and examples for creating local shell scripts to be executed by the Claude Code environment's hook system. This is the primary, documented purpose of the skill.
[INDIRECT_PROMPT_INJECTION] (LOW):
Ingestion points: The TOOL_INPUT environment variable contains data passed to the hook scripts, which may include untrusted input from previous agent steps.
Boundary markers: Absent in the examples; scripts rely on simple string matching (grep).
Capability inventory: Scripts use bash, git, and jq to inspect and potentially block actions or provide warnings.
Sanitization: The provided examples use grep and jq for sanitization and validation specifically to prevent dangerous actions (e.g., blocking rm -rf / or git push --force).

hooks