The Agent Skills Directory

[COMMAND_EXECUTION]: The skill installs and configures several bash scripts (e.g., ai-limits, sessions-gate.sh, quiet-hours.sh) to run as lifecycle hooks within the AI agent's environment. These scripts are used to enforce user-defined limits on tool calls and prompts.
[PERSISTENCE]: The skill modifies the agent's persistent configuration (specifically ~/.claude/settings.json) to register the aforementioned hooks. It also suggests adding a local directory to the user's system PATH to facilitate access to the management CLI.
[DYNAMIC_EXECUTION]: The provided management utility bin/ai-limits uses shell sourcing (. "$CONF") to dynamically load user preferences from a local configuration file. This is a standard practice for shell-based configuration management.
[INDIRECT_PROMPT_INJECTION]: During the setup interview, the skill ingests user-provided text (reasons for setting limits) and interpolates this content verbatim into a local markdown file (limits.md). While this text is not sanitized, its primary purpose is human documentation and it does not influence the skill's execution logic.

limits-setup