from codex-delegate48
Supervise large, implementation-heavy coding tasks by delegating mechanical code generation to Codex while Claude plans, constrains scope, and verifies results.
Codex Delegate provides a supervisor–delegate workflow: Claude (supervisor) writes a precise task brief and acceptance criteria, then delegates repetitive, token-heavy code generation to a Codex CLI wrapper. The skill ensures machine-readable run contracts and sentinel files so Claude can reliably accept or reject outputs after automated execution. Typical uses include multi-file boilerplate generation, bulk refactors, and automated test scaffolding.
Use this skill when the work is implementation-heavy but judgment-light: batch edits across many files, generating consistent scaffolding, or adding structured changes (logging, docstrings, types) at scale. Do NOT use it for architecture decisions, root-cause debugging, security judgments, or ambiguous product choices.
.ai/codex_task_<name>.md, launch the Codex wrapper, check <log>.result.json, and perform Claude acceptance checks.Best fit for agents that support Codex CLI wrappers and supervisor workflows (Claude as planner/reviewer, Codex as executor).
This skill has not been reviewed by our automated audit pipeline yet.
