subagent-cli

What problem does it solve?

This Skill provides a safe, repeatable way to delegate complex, multi-file coding, review, and refactoring tasks to external AI agent CLIs via Bash while retaining the host agent's identity, memory, and decision logic.

Core Features & Use Cases

• Tool orchestration and fallback: Prefer codex exec, fall back to cursor-agent -p, then claude -p with cost limits and sandboxing flags.
• Workspace-targeted execution: Enforces explicit workspace directory, captures stdout/stderr or writes outputs to files, and recommends git-based workflows for change tracking.
• Background execution and resilience: Uses nohup, timeout, and PID checks to run long tasks asynchronously and avoid blocking the host agent.
• Safety and audit: Recommends ephemeral sessions, sandboxed workspace-write, prompt hygiene, and post-run git diff checks.
• Use case: Delegate a multi-file refactor, unit test generation, or a comprehensive code review that requires a dedicated external coding agent.

Quick Start

Run a background codex exec targeting the project workspace with --full-auto and --ephemeral, saving the agent output to a result file and using a timeout to prevent runaway execution.