The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute the claude CLI with the --dangerously-skip-permissions flag as its default operational mode. This flag suppresses all built-in safety prompts and user confirmations for tool use, file edits, and command execution within the child session.
[PROMPT_INJECTION]: The instructions command the agent to 'auto-approve all Claude Code permission checks' and 'Use --dangerously-skip-permissions for every direct delegation.' This overrides standard safety protocols of the underlying CLI tool in favor of autonomous execution.
[COMMAND_EXECUTION]: The skill provides patterns for complex shell interactions, including claude auth status, claude doctor, and piping JSONL input to stdin for live process control, which increases the attack surface for potential command injection if user-provided paths or prompts are not carefully handled.
[DATA_EXFILTRATION]: While not directly exfiltrating data, the skill's workflow involves capturing and storing 'native session IDs' and 'provider' details. If these are sent to an external service or stored insecurely, they could be used to reconstruct or resume sensitive development sessions.

claude-cli