The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a Node.js relay script (scripts/relay.mjs) to spawn the omp process. When running in write mode, the implementer executes with the --yolo flag, which bypasses approval for tool calls. As the skill explicitly notes, the implementer has no sandbox and can execute arbitrary shell commands and modify the filesystem with the user's host permissions.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of an external dependency, @oh-my-pi/pi-coding-agent, from a public registry. This package is not part of a verified or trusted vendor list, introducing a potential supply chain risk if the package were compromised.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the processing of project-local workspace files (including AGENTS.md and CLAUDE.md) that could be manipulated by an attacker to include malicious instructions. While the relay script mitigates this by disabling project-local extensions and rules by default, an user opting-in via the --approve flag increases the attack surface for indirect prompt injection.
[DYNAMIC_EXECUTION]: The relay script dynamically constructs and executes shell commands for the omp and git binaries using node:child_process. Although it validates certain inputs like model IDs and provider names against a strict character whitelist to prevent flag injection, the core functionality involves executing complex coding tasks passed via stdin to the implementer binary.

omp-delegate