The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill documents the elevated:true parameter for the bash tool, which enables running shell commands on the host system instead of within a restricted sandbox environment.
[COMMAND_EXECUTION]: The documentation encourages the use of the --yolo flag for the Codex CLI, which explicitly disables sandboxing and security approvals, allowing for unrestricted and potentially dangerous command execution.
[REMOTE_CODE_EXECUTION]: The skill provides workflows for cloning external, untrusted repositories (via git clone or gh pr checkout) and executing their contents using package managers like pnpm install and coding agents with safety flags disabled. This creates a high risk of executing malicious code embedded in Pull Requests or external projects.
[EXTERNAL_DOWNLOADS]: The instructions include the installation of third-party software from an unverified external source (npm install -g @mariozechner/pi-coding-agent).
[COMMAND_EXECUTION]: The "Auto-Notify on Completion" section describes a pattern of embedding system commands (openclaw system event) directly into the prompts processed by coding agents. This increases the surface area for command injection, especially when the agent is processing external code or data.
[PROMPT_INJECTION]: The skill instructs the agent to use specific flags (--yolo, --full-auto) and instruction patterns that are designed to bypass standard safety constraints and user approval steps.

coding-agent