The Agent Skills Directory

[COMMAND_EXECUTION]: The skill defines a bash tool that permits the execution of arbitrary shell commands. It includes an elevated parameter specifically designed to run processes on the host system rather than within a restricted sandbox.
[REMOTE_CODE_EXECUTION]: The documentation actively promotes the --yolo flag for the Codex CLI, which is described as having "NO sandbox, NO approvals," creating a direct path for unverified, LLM-generated code to execute with full system access.
[EXTERNAL_DOWNLOADS]: The skill provides explicit instructions for downloading and installing external packages (e.g., npm install -g @mariozechner/pi-coding-agent) and executing dependency installations (pnpm install) from potentially untrusted remote repositories.
[PROMPT_INJECTION]: The skill contains instructions designed to bypass tool-specific security constraints, such as using mktemp and git init to circumvent Codex's refusal to run in untrusted directories.
[PROMPT_INJECTION]: The skill is highly vulnerable to indirect prompt injection due to its core functionality of processing external code and Pull Requests.
Ingestion points: External repositories and Pull Request diffs via git clone and gh pr checkout in SKILL.md.
Boundary markers: None are defined to separate untrusted code from the agent's instructions.
Capability inventory: Includes full bash execution, elevated host permissions, and interactive terminal control through process tool actions like write and submit.
Sanitization: No sanitization or validation of the content being processed is performed; the skill relies solely on advisory warnings to the user.

coding-agent