---
name: ollama-cli-runtime
description: Internal helper contract for calling the ollama-companion runtime from Claude Code
user-invocable: false
---

# Ollama Runtime

Use this skill only inside the `ollama:ollama-rescue` subagent.

## Invoking the Companion Script

Primary helper:
- `node "${CLAUDE_PLUGIN_ROOT}/scripts/ollama-companion.mjs" task "<raw arguments>"`

The companion script talks directly to Ollama's HTTP API. There is no broker process and no app-server to start or stop — Ollama itself must be running at `OLLAMA_HOST` before the companion is invoked.

## Environment Variables

| Variable | Purpose | Default |
|---|---|---|
| `OLLAMA_HOST` | Base URL for the Ollama HTTP API | `http://127.0.0.1:11434` |
| `OLLAMA_PLUGIN_DEFAULT_MODEL` | Default model when `--model` is not supplied | none (required to be set or passed explicitly) |
| `OLLAMA_PLUGIN_JOB_DIR` | Override for background job storage directory | `.ollama/companion-jobs/` inside workspace root |
| `OLLAMA_PLUGIN_LOG_LEVEL` | Verbosity: `silent`, `error`, `info`, `debug` | `info` |

## Selecting a Model

Pass `--model <name>` to override the default:
- `node "${CLAUDE_PLUGIN_ROOT}/scripts/ollama-companion.mjs" task --model qwen2.5-coder:14b "<task>"`

If `--model` is omitted and `OLLAMA_PLUGIN_DEFAULT_MODEL` is not set, the companion exits with a non-zero status and a message directing the user to `/ollama:setup`.

See the `ollama-model-prompting` skill for model selection guidance.

## 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 Ollama API calls, or any other Bash activity.
- Do not call `setup`, `review`, `adversarial-review`, `status`, `result`, or `cancel` from `ollama:ollama-rescue`.
- Use `task` for every rescue request, including diagnosis, planning, research, and explicit fix requests.
- Leave model unset by default. Add `--model` only when the user explicitly asks for one.
- Default to a write-capable Ollama 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 `--effort`, pass it through to `task`. (The flag is accepted but currently inert — Ollama exposes no equivalent of Codex's reasoning-effort knob.)
- If the forwarded request includes `--emit-patch`, pass it through. This forces the legacy patch-emit flow (the model returns a unified diff) instead of the default agentic tool-calling loop.
- 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.
- `--effort`: accepted values are `none`, `minimal`, `low`, `medium`, `high`, `xhigh`.
- `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 Ollama work in `ollama:ollama-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 Ollama cannot be invoked, return nothing.