The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Bash commands to resolve the plugin root, perform status checks on configuration files, and manage the installation of hook assets. These operations are restricted to the skill's own assets and the agent's local configuration directories (~/.claude or project-specific .claude folders).
[DYNAMIC_EXECUTION]: The installer uses node -e to execute small, static JavaScript snippets for JSON manipulation. These snippets facilitate safe reading and merging of configuration values (like time budgets and agent-type overrides) by passing them via environment variables rather than direct string concatenation in the shell.
[PERSISTENCE]: The skill modifies settings.json to register PreToolUse and SubagentStop hooks. This is the intended and documented primary function of the skill, allowing it to monitor and enforce budgets across agent sessions.
[DATA_EXPOSURE]: The hooks store runtime state (start times and status flags) in the system's temporary directory. The implementation uses security best practices such as lstat to prevent symlink-follow attacks and verifies directory ownership and permissions (0700) to ensure state data is not accessible to other users on shared systems.

agent-deadline-setup