claude-exec
Claude Exec — Cross-Agent Delegation
Delegate review and exploration tasks to an independent Claude sub-agent for a second opinion. The sub-agent acts as a strict counter-reviewer that catches blind spots you might miss.
When Not to Use
Skip this skill when you are confident in the output and the user hasn't asked for a second opinion. Reserve it for quality gates and deliberate review steps.
How It Works
In Claude Code, ALWAYS use the Agent tool. Do not shell out to
claude -p. The Bash-claude -ppath is the cross-agent fallback for environments without the Agent tool (Codex CLI, Cursor, OpenCode, …) — it is NOT a shortcut from inside Claude Code itself.
If the Agent tool is available (Claude Code): use it. The Agent tool streams output natively (the user sees progress as the sub-agent works, with no buffering), has no artificial --max-turns ceiling, and inherits the running session's environment without spawning a new claude process. The CLI fallback has none of those properties.
If the Agent tool is NOT available (Codex CLI, other non-Claude environments): fall back to claude -p via the Bash tool. See the CLI Fallback section for the correct invocation pattern.