What problem does it solve?
Running a second AI coding agent alongside your main session often means juggling tmux panes, guessing whether the agent is ready, and accidentally typing into the wrong pane. This Skill provides a safe, ordered workflow for discovering, starting, prompting, waiting on, and answering a Sidecar-managed agent entirely from a shell, with structured refusal codes instead of silent failures.
Core Features & Use Cases
- Safe agent lifecycle: Discover shells with
sidecar shell list and sidecar agent list, create the layout separately with sidecar create shell, then start a provider with sidecar agent start that returns only when the agent is confirmed ready.
- Pinned prompt-and-wait:
sidecar agent prompt --wait submits and waits as one operation pinned to a specific session, pane, process, and provider, returning explicit refusal codes like agent_blocked, agent_replaced, or agent_prompt_stalled.
- Read before answering: Inspect a blocked agent with
sidecar agent read (visible, recent, unwrapped, detection, or transcript sources) before replying with logical keys via sidecar agent send-keys.
- Use Case: Ask a second agent to review the current diff in a new right-side pane, wait up to two minutes for it to finish, and read its findings — all without disturbing the user's existing layout.
Quick Start
Ask the agent to start a reviewer agent in a new right-side shell pane, prompt it to review the current diff, wait for it to finish, and report the findings.
