The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill provides explicit instructions to bypass platform-level security restrictions on the ~/.claude/ directory. It directs the agent to use shell commands (cp, mv, node) to circumvent protections that normally block the modification of global configuration files like settings.json.- [DYNAMIC_EXECUTION]: The installation and configuration merging processes rely on executing dynamic JavaScript code via node -e shell one-liners. This allows for the execution of string-embedded logic directly in the host environment to manipulate configuration data.- [PERSISTENCE_MECHANISMS]: By modifying the settings.json file, the skill installs permanent event hooks (SessionStart, UserPromptSubmit, SubagentStart). These hooks ensure that the skill's custom logic and prompt injections are automatically executed in every future agent session and subagent spawn.- [PROMPT_INJECTION]: The skill's core functionality is to inject specific directives ("terse mode") into the AI agent's context and all spawned subagents, overriding the default behavior of the model.- [DATA_EXPOSURE_AND_EXFILTRATION]: The skill reads the ~/.claude/settings.json file, which is a sensitive configuration file containing agent settings and potentially environment variables or other sensitive data.

think-short-setup