codex-dispatch is a Codex skill for making a visible Direct/Single/Parallel/Sequential decision before non-trivial Codex work and before any subagent delegation.
It is intentionally separate from evidence-first: codex-dispatch decides how work should be organized, while evidence-first keeps execution and delivery evidence-backed.
- Emit a visible dispatch decision before substantial work and before the first child-work call.
- Block
spawn_agent,followup_task, or scope-expanding child work until the dispatch preflight is done. - Audit existing children instead of restarting conforming work if the preflight was missed.
- Start from Direct and spawn subagents only when a hard delegation gate is met.
- Treat strong root models and subagent-heavy environments as reasons to be stricter, not looser.
- Keep root ownership of requirements, shared context, conflict control, synthesis, and final user-facing delivery.
List the available skill:
npx skills add yinguangyao/codex-dispatch-skill --listInstall globally for Codex:
npx skills add yinguangyao/codex-dispatch-skill --skill codex-dispatch --agent codex --globalInstall non-interactively:
npx skills add yinguangyao/codex-dispatch-skill --skill codex-dispatch --agent codex --global --yesUse with evidence-first for disciplined Codex work:
npx skills add yinguangyao/evidence-first-skill --skill evidence-first --agent codex --global
npx skills add yinguangyao/codex-dispatch-skill --skill codex-dispatch --agent codex --globalRecommended order for non-trivial Codex work:
codex-dispatch: choose Direct, Single, Parallel, or Sequential before delegating.evidence-first: execute and report with verified claims.
mkdir -p ~/.codex/skills
git clone git@github.com:yinguangyao/codex-dispatch-skill.git ~/.codex/skills/codex-dispatchSKILL.md: Codex skill entry point.agents/openai.yaml: Codex UI metadata.