CO

Primary entrypoint for Codex Orchestrator usage: route tasks to the right pipeline, mode, and supporting skills with minimal, auditable steps.

Manage non-trivial tasks via focused collab subagents to save context and improve throughput. Use when work spans multiple files/components, can be split into independent streams, needs separate validation/review, or risks context bloat. Favor direct execution for trivial one-shot tasks.

Ship a signed tag + GitHub Release + npm publish for @kbediako/codex-orchestrator with low-friction, agent-first steps (PR -> watch-merge -> tag -> watch publish -> downstream smoke).

Use when operating the Codex delegation MCP server and tools (delegate.spawn, delegate.question.*, delegate.cancel, github.merge confirmation flow), or when configuring delegation mode/tool_profile with delegation MCP enabled by default.

Use for required periodic cross-check reviews during implementation and before handoff using `codex review`.

Control and inspect Chrome via the Chrome DevTools MCP server (navigate, interact, screenshots, console, network, perf).

Run collab/multi-agent eval scenarios (symbolic RLM, large-context, pause/resume, multi-hour checkpoints) and capture manifest-backed evidence.

Agent-first Codex Orchestrator usage playbook for downstream users: docs-first execution, delegation-first operations, non-coercive capability steering, and autonomy-preserving review guidance.

Structure multi-agent brainstorming and deliberation (options, tradeoffs, decision framing) without drifting into implementation.

Compatibility alias for delegation-first workflows. Prefer `delegation-usage` as the canonical skill.

Use when a task requires a spec-driven workflow: draft/refresh PRD + TECH_SPEC + ACTION_PLAN + tasks, link TECH_SPEC in tasks/index.json, and run docs-review before implementation.

Run an explicit post-implementation elegance/minimality pass to keep the smallest correct solution and remove avoidable complexity before handoff.

Wait for long-running operations (including multi-hour/multi-day runs) by polling to terminal status with checkpointed updates, explicit stall/intervention policy, and patience-first continuity.