
from ask-codex9
Routes heavy repo exploration, implementation, and verification work from Claude Code to the Codex CLI with clear review loops and verification requirements.
This skill defines a collaboration framework between a decision-making agent (Claude Code) and an autonomous implementer (Codex). It provides routing rules, invocation patterns, and review-loop policies so that heavy repository exploration, data inspection, multi-file impact analysis, and implementation tasks are delegated to Codex while Claude Code retains final judgment and verification. The skill includes mode definitions (scout, single delegation, parallel, iterative review, brokered collaboration) and concrete invocation examples and exit-code handling.
Trigger this skill when a task requires broad codebase search, log/data inspection, multi-file refactors, long-running verification commands, or when the user explicitly asks to 'use Codex'. Do NOT use it for user-facing product decisions, PR/deployment operations, or tasks that need extensive multi-turn clarification before execution.
ask_codex.sh and codex_broker.sh (invocation patterns described, scripts referenced though not bundled)Best suited for setups using Claude Code as the coordinator and a Codex-compatible CLI/executor for autonomous exploration (Codex CLI, similar executor tools). It works where agents support shell/script invocation and session-based review loops.
This skill has not been reviewed by our automated audit pipeline yet.