Skip to content
Back to skills

Gemini Cli Runtime

ASecurity

Internal helper contract for calling the gemini-companion runtime from Claude Code

  • 24 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added May 26, 2026
developmentbashnode

Works with

  • claude code
  • cli

Security analysis

A100/100

Scanned May 27, 2026

npx -y skills add sakibsadmanshajib/gemini-plugin-cc --skill gemini-cli-runtime --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Gemini Cli Runtime?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Gemini Cli Runtime
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/sakibsadmanshajib-gemini-cli-runtime/badge)](https://www.skillsdirectory.com/skills/sakibsadmanshajib-gemini-cli-runtime)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: gemini-cli-runtime
description: Internal helper contract for calling the gemini-companion runtime from Claude Code
user-invocable: false
---

# Gemini Companion Runtime Contract

## Invocation

```bash
node "${CLAUDE_PLUGIN_ROOT}/scripts/gemini-companion.mjs" task [flags] -- <prompt>
```

Everything after `--` (or the first non-flag positional) is the task text sent to Gemini.

## Flags

| Flag | Type | Default | Description |
|------|------|---------|-------------|
| `--write` | boolean | false | Enable Gemini to make file changes (sets `--approval-mode auto_edit`) |
| `--model <name>` | string | `auto-gemini-3` | Model override — see model reference below |
| `--thinking-budget <n>` | integer | (unset) | Thinking token budget for the model |
| `--approval-mode <mode>` | string | `default` | One of: `default`, `auto_edit`, `yolo`, `plan` |
| `--resume-last` | boolean | false | Resume the most recent task thread in this repository |
| `--background` | boolean | false | Run as a detached background job |
| `--wait` | boolean | true | Run in the foreground (default) |
| `--cwd <path>` | string | `$CLAUDE_PROJECT_DIR` | Working directory override |
| `--json` | boolean | false | Emit structured JSON instead of rendered markdown |

## Model Reference

The default model is **`auto-gemini-3`**, which routes to the best available Gemini 3.x model.

### Auto-routing aliases (recommended)

| Alias | Routes to |
|-------|-----------|
| `auto-gemini-3` | Gemini 3.1 or 3 (best available) — **default** |
| `auto-gemini-2.5` | Gemini 2.5 (stable, production) |
| `pro` | `gemini-3.1-pro-preview` (highest capability) |
| `flash` | `gemini-3-flash-preview` (speed-optimised) |
| `flash-lite` | `gemini-3.1-flash-lite-preview` (fastest) |

### Concrete model IDs (pin to specific preview)

| Model ID | Generation |
|----------|-----------|
| `gemini-3.1-pro-preview` | Gemini 3.1 |
| `gemini-3.1-flash-lite-preview` | Gemini 3.1 |
| `gemini-3-pro-preview` | Gemini 3 |
| `gemini-3-flash-preview` | Gemini 3 |
| `gemini-2.5-pro` | Gemini 2.5 |
| `gemini-2.5-flash` | Gemini 2.5 |
| `gemini-2.5-flash-lite` | Gemini 2.5 |

Unlisted model IDs are forwarded as-is to Gemini CLI.

## Safety Rules

- Exactly one Bash call per rescue invocation.
- Never chain additional tool calls after the companion returns.
- Never inspect, modify, or second-guess Gemini's output.
- If the companion exits non-zero, return nothing — do not fabricate a response.
- Do not include `--thinking-budget`, `--model`, `--resume-last`, `--background`, or `--wait` in the task text itself. They are runtime controls.

## Output

- `stdout`: Rendered markdown (default) or JSON (`--json`).
- `stderr`: Progress updates during execution.
- Exit code 0 = success, non-zero = failure.

## Job Health Labels

`/gemini:status` reports a conservative `Health` label for each active job.
Interpret each label as follows when deciding what to do next:

| Label | Interpretation | Recommend to the user |
|-------|----------------|-----------------------|
| `active` | Gemini emitted progress recently. | Wait for completion; do not cancel. |
| `quiet` | Heartbeat is recent but no new progress. | Re-check `/gemini:status` shortly. |
| `possibly_stalled` | No recent heartbeat or progress. | Re-check `/gemini:status`, fetch `/gemini:result`, or retry if the job does not recover. |
| `rate_limited` | Explicit quota or 429-class diagnostic. | Wait, switch models (`--model flash` / `--model auto-gemini-2.5`), or cancel with `/gemini:cancel`. |
| `auth_required` | Explicit auth/credential diagnostic. | Point the user to `/gemini:setup` to re-authenticate before retrying. |
| `broker_unhealthy` | The ACP broker reported busy/disconnected. | Re-check status shortly; a restart may be needed if it persists. |
| `worker_missing` | Worker PID is no longer alive. | Fetch `/gemini:result`; retry if the output is incomplete. |
| `failed` | Worker or Gemini ended with an error. | Fetch `/gemini:result` for the diagnostic and retry only after understanding it. |
| `completed` | Gemini finished successfully. | Fetch `/gemini:result` and present output; do not retry. |
| `cancelled` | Job was cancelled by user or system. | Fetch `/gemini:result` for final diagnostics; consider retrying if needed. |

Never claim a job is dead or useless based on `quiet` or `possibly_stalled`
alone. Those labels mean "check again", not "give up".

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.