# Gemini CLI Runtime > Internal helper contract for calling the gemini-companion runtime from Claude Code Use when this capability is needed. - Skill: `tomevault-io/gemini-cli-runtime-13` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add tomevault-io/gemini-cli-runtime-13` - Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/gemini-cli-runtime-13/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: tomevault-io (https://skillmd.com/u/tomevault-io) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/tomevault-io/gemini-cli-runtime-13 --- # Gemini Runtime Use this skill only inside the `gemini:gemini-rescue` subagent. Primary helper: - `node "${CLAUDE_PLUGIN_ROOT}/scripts/gemini-companion.mjs" task ""` Execution rules: - The rescue subagent is a forwarder, not an orchestrator. Its only job is to invoke `task` once and return that stdout unchanged. - Prefer the helper over hand-rolled `git`, direct Gemini CLI strings, or any other Bash activity. - Do not call `setup`, `review`, `adversarial-review`, `status`, `result`, or `cancel` from `gemini:gemini-rescue`. - Use `task` for every rescue request, including diagnosis, planning, research, and explicit fix requests. - You may use the `gemini-prompting` skill to rewrite the user's request into a tighter Gemini prompt before the single `task` call. - That prompt drafting is the only Claude-side work allowed. Do not inspect the repo, solve the task yourself, or add independent analysis outside the forwarded prompt text. - Leave `--thinking` unset unless the user explicitly requests a specific level. The flag is parsed but currently behaves as a one-shot/one-time hint; the runtime does not yet ship a per-invocation thinking override. For persistent control, set the default in `settings.json` via `/gemini:setup`. - The default model is `auto-gemini-3`. Add `--model` only when the user explicitly asks for one. Aliases include `pro`, `auto-gemini-2.5`, `flash`, `flash-lite`. - If the user asks for `flash`, map that to `--model gemini-3-flash-preview`. - Default to a write-capable Gemini run by adding `--write` unless the user explicitly asks for read-only behavior or only wants review, diagnosis, or research without edits. Command selection: - Use exactly one `task` invocation per rescue handoff. - If the forwarded request includes `--background` or `--wait`, treat that as Claude-side execution control only. Strip it before calling `task`, and do not treat it as part of the natural-language task text. - If the forwarded request includes `--model`, pass it through to `task`. - If the forwarded request includes `--thinking`, pass it through to `task`. Accepted values are `off`, `low`, `medium`, `high`. - If the forwarded request includes `--stream-output`, pass it through. Default (no flag) uses compact stderr markers. - If the forwarded request includes `--resume`, strip that token from the task text and add `--resume-last`. - If the forwarded request includes `--fresh`, strip that token from the task text and do not add `--resume-last`. - `--resume`: always use `task --resume-last`, even if the request text is ambiguous. - `--fresh`: always use a fresh `task` run, even if the request sounds like a follow-up. - `task --resume-last`: internal helper for "keep going", "resume", "apply the top fix", or "dig deeper" after a previous rescue run. Safety rules: - Default to write-capable Gemini work in `gemini:gemini-rescue` unless the user explicitly asks for read-only behavior. - Preserve the user's task text as-is apart from stripping routing flags. - Do not inspect the repository, read files, grep, monitor progress, poll status, fetch results, cancel jobs, summarize output, or do any follow-up work of your own. - Return the stdout of the `task` command exactly as-is. - If the Bash call fails or Gemini cannot be invoked, return nothing. ## Job health labels `/gemini:status` and the companion record a `health` field on every tracked job. When you see a job in any of these states, follow the recommended action: - `active` — Gemini is producing output recently. Recommended action: wait; do not cancel or poll aggressively. - `quiet` — No recent progress chunk but the worker is alive and emitting heartbeats. Recommended action: let it run; check again with `/gemini:status` after a minute or two. - `possibly_stalled` — Both progress and heartbeat have aged out. Recommended action: run `/gemini:status ` to inspect recent events; if stuck, `/gemini:cancel ` and retry. - `rate_limited` — Gemini reported a rate-limit diagnostic. Recommended action: back off, retry later, or switch model with `--model`. - `auth_required` — Gemini is missing credentials or the session expired. Recommended action: re-run `/gemini:setup` and authenticate (`!gemini` or `GEMINI_API_KEY`). - `broker_unhealthy` — The ACP broker process is misbehaving. Recommended action: cancel the job and start a new one; if it persists, restart the Claude session to recycle the broker. - `worker_missing` — A job is marked running but the worker process is gone. Recommended action: `/gemini:cancel ` to clear it, then start a fresh task. - `failed` — Job terminated with an error. Recommended action: read `/gemini:result ` for the failure event and decide whether to retry. - `completed` — Job finished successfully. Recommended action: fetch the result with `/gemini:result `. - `cancelled` — Job was cancelled by the user or the system. Recommended action: start a new task if needed; do not fabricate output for a cancelled run. --- > Source: [seungmanchoi/claude-gemini-plugin](https://github.com/seungmanchoi/claude-gemini-plugin) — distributed by [TomeVault](https://tomevault.io).