The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to analyze the user's local shell command history and tool usage patterns found in ~/.claude/projects/ to identify frequent invocations.\n- [DATA_EXPOSURE]: The skill accesses recent session transcripts located in ~/.claude/projects/*.jsonl. While these files contain interaction history, the data is processed locally to generate a frequency list of commands and is not exfiltrated.\n- [PROMPT_INJECTION]: The skill provides strict logical guidelines for the agent, explicitly forbidding the allowlisting of commands that allow arbitrary code execution (e.g., shells, interpreters, and broad wildcards like npm run *). It also enforces a 'read-only' filter to exclude commands that write, delete, or push data.\n- [REMOTE_CODE_EXECUTION]: The skill contains no remote downloads or external script executions. All analysis and configuration changes (modifying .claude/settings.json) occur locally within the user's project environment.

fewer-permission-prompts