What problem does it solve? Delegating work to peer AI CLIs often fails silently: workers answer without reading the repository, models get refused by outdated CLIs, parallel runs collide, and nobody verifies the result. This Skill provides a disciplined workflow for launching codex and agy workers with scoped prompts, capability probes, run-state tracking, and mandatory post-run review. ## Core Features & Use Cases - Capability probing: probe.sh checks the installed codex version, flag support, web-search mode, and whether sandboxed file reads actually work before a launch. - Managed launches: launch.sh runs codex in the foreground with unit locks, run directories, status snapshots, expected-artifact checks, and classified exit codes for sandbox blocks and model-version gaps. - Run inspection: status.sh reports heartbeat age, activity age, artifact sizes, and exit codes by unit or status file, with a --wait mode for shadowing another run. - Use Case: Fan out two independent code-review units to codex with disjoint write scopes, monitor them via status.sh, then re-run the acceptance tests and inspect every touched file before accepting the result. ## Quick Start Ask the assistant to delegate a bounded implementation task to codex by writing a file-backed prompt with acceptance commands and launching it through the skill's launch.sh script.