other-llmlisted

# Other LLM: Second Opinion / Hands-On via the Other Coding CLI Delegate work to whichever coding CLI is backed by a DIFFERENT model family from the one currently running this session — a review, an exploration, or actually doing something (running commands, making changes). The other CLI always runs with full access; what it's allowed to do comes from the **prompt**, not a sandbox flag. ## Step 1: pick the runner 1. Identify the active model/provider for this session (in a multi-provider tool, an env var or session context field typically names it directly). 2. If the active model is Anthropic/Claude, the runner is `codex`. 3. If the active model is OpenAI Codex (or another OpenAI model), the runner is `claude`. 4. If the active model family can't be determined, ask the user which CLI to use. Do not guess. Everything below is written for `codex` as the runner, since that is the common case (this session IS Claude). If Step 1 selected `claude` instead, run the equivalent `claude` invocation — same detach/poll/digest shape, just `claude -p --no-session-persistence "<prompt>"` in place of the `codex exec ...` command below, with no `--sandbox`/`--output-format` flags (those are Codex-specific). These runs are LONG (minutes, sometimes over an hour). This skill spawns **one dedicated subagent** whose entire job is: launch the runner detached → block on a re-invoked poll loop until it exits → read the output → digest it → return a summary. The calling agent never bab