choose-agentlisted

# choose-agent Skill for picking WHICH coding agent to delegate a task to. The readiness MCP (`owner_coding_agents_check_readiness`) gives you the facts; this skill is the judgement. Keep the two separate: facts live in the MCP, the choice lives here, so the policy can change without touching the MCP. ## Always check before delegating Before every new delegation, call **`owner_coding_agents_check_readiness`** once. It returns a snapshot of all 5 coding agents in one shot: ``` agents: [{ platform, installed, logged_in, busy, last_worked_at, last_worked_page, last_worked_step }, ...] ``` Do not delegate blind. "Bridge online" does NOT mean the agent is usable — an agent can be alive but logged out, and the task would go into the void. ## Confirm what you hand off — BEFORE you dispatch (mandatory) Delegating is a state change (a task starts running on a real agent), so it obeys `confirm-before-mutation`. Before you call `owner_delegate_task_to_platform`, restate to the owner EXACTLY what you are about to hand off and wait for an explicit go: > Правильно ли я вас понимаю, что вы хотите, чтобы я передал агенту кодирования: > - задача: <одна-две строки — что именно должен сделать кодер> > - агент-получатель: <claude-code | codex | …> > - что он получит на вход: <страница/шаг/требования> > > Подтвердите — и я передаю. Never dispatch silently. The owner must see the exact payload and the target agent before any coder starts work. ## The two facts to report — "pre