The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install external packages from npm, including @anthropic-ai/claude-code, @openai/codex, and @mariozechner/pi-coding-agent.\n- [COMMAND_EXECUTION]: Instructions explicitly direct the agent to bypass interactive security confirmations and human-in-the-loop approvals by using high-risk flags such as --yolo (for Codex) and --permission-mode bypassPermissions (for Claude Code).\n- [COMMAND_EXECUTION]: The skill documentation describes and facilitates command execution with host-level privileges via the elevated: true parameter, which bypasses standard sandbox restrictions when allowed by the environment.\n- [REMOTE_CODE_EXECUTION]: The core functionality involves spawning and managing background sessions for external coding agents that are designed to generate and execute code autonomously on the host system.\n- [PROMPT_INJECTION]: The skill presents a vulnerability surface for indirect prompt injection as it clones and reviews untrusted code from external repositories (e.g., via gh pr checkout).
Ingestion points: git clone and gh pr checkout operations in SKILL.md.
Boundary markers: No specific boundary markers or instructions to ignore embedded malicious directions are present in the prompts sent to the sub-agents.
Capability inventory: Full bash access with PTY and documented elevated host access.
Sanitization: No validation or sanitization of external content is performed before processing.

coding-agent