From gemini
Internal helper contract for calling the gemini-companion runtime from Claude Code
How this skill is triggered — by the user, by Claude, or both
Slash command
/gemini:gemini-cli-runtimeThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Use this skill only inside the `gemini:gemini-rescue` subagent.
Use this skill only inside the gemini:gemini-rescue subagent.
Primary helper:
node "${CLAUDE_PLUGIN_ROOT}/scripts/gemini-companion.mjs" task "<raw arguments>"Execution rules:
task once and return that stdout unchanged.git, direct Gemini CLI strings, or any other Bash activity.setup, review, adversarial-review, status, result, or cancel from gemini:gemini-rescue.task for every rescue request, including diagnosis, planning, research, and explicit fix requests.gemini-prompting skill to rewrite the user's request into a tighter Gemini prompt before the single task call.--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.auto-gemini-3. Add --model only when the user explicitly asks for one. Aliases include pro, auto-gemini-2.5, flash, flash-lite.flash, map that to --model gemini-3-flash-preview.--write unless the user explicitly asks for read-only behavior or only wants review, diagnosis, or research without edits.Command selection:
task invocation per rescue handoff.--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.--model, pass it through to task.--thinking, pass it through to task. Accepted values are off, low, medium, high.--stream-output, pass it through. Default (no flag) uses compact stderr markers.--resume, strip that token from the task text and add --resume-last.--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:
gemini:gemini-rescue unless the user explicitly asks for read-only behavior.task command exactly as-is./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 <job-id> to inspect recent events; if stuck, /gemini:cancel <job-id> 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 <job-id> to clear it, then start a fresh task.failed — Job terminated with an error. Recommended action: read /gemini:result <job-id> for the failure event and decide whether to retry.completed — Job finished successfully. Recommended action: fetch the result with /gemini:result <job-id>.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.Provides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub seungmanchoi/claude-gemini-plugin --plugin gemini