Ejecuta cualquier Skill en Manus
con un clic

Ejecuta cualquier Skill en Manus con un clic

configure-codex-with-ollama

Estrellas1

Forks0

Actualizado16 de junio de 2026 a las 04:58

This skill should be used when the user asks to "open the Ollama control panel", "set up Codex App with Ollama", "configure Codex CLI with Ollama", "use Ollama with Codex", "switch Codex App to a Codex model", "switch Codex App to an Ollama model", "run Codex with a local model", "run Codex with an Ollama Cloud model", "restore Codex after Ollama", "check Ollama Codex status", or mentions `ollama launch codex-app`, `ollama launch codex`, `codex --oss`, or `codex --profile ollama-launch`.

Instalación

Instalar con Codex o Claude Copia este prompt, pégalo en Codex, Claude u otro asistente, y deja que revise la página de la skill y la instale por ti.

Ejecutar en Manus

Fuente

daniel-p-green

daniel-p-green/ollama-codex-plugin

Abrir repositorio de GitHub Ver repositorios del creador

Descarga

Ejecutar en Manus

Ocupaciones relacionadasSOC

Basado en la clasificación ocupacional SOC

Desarrolladores de softwareOcupaciones informáticas y matemáticas·SOC 15-1252

SKILL.md

readonly

name	configure-codex-with-ollama
description	This skill should be used when the user asks to "open the Ollama control panel", "set up Codex App with Ollama", "configure Codex CLI with Ollama", "use Ollama with Codex", "switch Codex App to a Codex model", "switch Codex App to an Ollama model", "run Codex with a local model", "run Codex with an Ollama Cloud model", "restore Codex after Ollama", "check Ollama Codex status", or mentions `ollama launch codex-app`, `ollama launch codex`, `codex --oss`, or `codex --profile ollama-launch`.
version	0.4.0

Configure Codex With Ollama

Use this skill to route Codex App and Codex CLI setup through the Ollama for Codex plugin. Keep the integration thin: verify local readiness, run the official Ollama or documented Codex config path, and report exactly what changed or what still needs attention.

Core Rule

Delegate Ollama-side Codex App configuration to Ollama whenever an official ollama launch codex-app command exists. For switching back to a native Codex/OpenAI model from the panel, use the wrapper's app-use-codex-model <model> command, which backs up Codex App config and edits the documented top-level model key.

For Codex CLI profile setup, use the wrapper's deterministic cli-config <model> command. Live acceptance testing on Ollama 0.30.8 showed that ollama launch codex --config --model <model> writes the expected profile/catalog and then can still launch an interactive Codex TUI. The plugin avoids that nested-session surprise by writing the documented ollama-launch CLI profile and model catalog directly.

Use the bundled wrapper for deterministic checks, dry-run support, and consistent command names:

bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh status

Current User Surface

When the user asks for the GUI, visual panel, control panel, or easiest way to manage Ollama in Codex, be precise about the current supported surface:

The accepted GUI today is Codex's plugin GUI: plugin card, starter prompts, enable/disable controls, slash-command discovery, and command output in the Codex conversation.
A native in-chat visual model picker is not currently a verified public plugin surface for this repo-local plugin.
The /ollama and /ollama-codex-panel commands are experimental widget probes retained for future host support and diagnostics.

Start with the supported command surface:

bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh doctor
bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh status

If the user explicitly wants to try the experimental widget probe, use the render_ollama_codex_panel tool from the ollama_codex MCP server. In Codex tool notation this is expected to appear as mcp__ollama_codex__render_ollama_codex_panel.

Do not call the widget probe accepted unless Codex renders it visibly inside a fresh Codex Mac app thread. If Codex returns JSON, structured tool output, or a plain text result, report that the command layer is working but the host visual surface is not available in that build/session.

If the render_ollama_codex_panel tool is not available in the current Codex thread, or if the MCP call fails with Transport closed, say so plainly and run the plugin doctor if the command layer is available:

bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh doctor

Then ask the user to open a fresh Codex thread after installing or updating the plugin. Do not fall back to calling a localhost panel the plugin GUI.

Status And Discovery

Run status first when the user's request includes "check", "verify", "status", "doctor", "why is this not working", or troubleshooting language:

bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh status

Report:

Whether ollama is installed.
Whether Ollama is at least v0.24.0.
Whether the Ollama HTTP API is reachable.
A brief local model summary when available.
Whether codex is installed.
Whether Codex CLI Ollama profile/catalog files exist.
Whether Codex App backup files exist.

For model discovery:

bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh list-models
bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh pull-model <model>

Preserve model names exactly, including tags such as :cloud.

Codex App Workflows

Use these commands for OpenAI's desktop Codex App on macOS or Windows:

bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh app-setup
bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh app-use-codex-model gpt-5.4
bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh app-use-model gemma4:31b
bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh app-use-model kimi-k2.6:cloud
bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh app-restore

Backward-compatible aliases remain available:

bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh setup
bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh use-model gemma4:31b
bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh restore

Use app-setup for "Set up Codex App with Ollama". Use app-use-codex-model <model> for "Switch Codex App to a native Codex/OpenAI model". Use app-use-model <model> for "Switch Codex App to an Ollama model". Use app-restore for "Restore Codex App's previous profile".

The wrapper passes Ollama's --yes flag for Ollama App setup, Ollama model switching, and restore so the command can complete the restart/profile flow. app-use-codex-model backs up Codex App config, restores away from an active Ollama profile when possible, and sets the native Codex model. Report that Codex may restart or require a fresh task. Do not claim the switch or restore completed unless the command completed.

Codex CLI Workflows

Use these commands for the codex terminal CLI:

bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh cli-setup
bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh cli-config gpt-oss:20b
bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh cli-restore
bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh cli-run
bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh cli-run-model gpt-oss:120b
bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh cli-run-model gpt-oss:120b-cloud
bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh cli-run-profile

Use cli-setup for ollama launch codex, which refreshes the model catalog and launches a dedicated interactive Codex profile for that session. Use cli-config <model> to write the documented ollama-launch profile and model catalog without launching a nested Codex session. Use cli-restore for ollama launch codex --restore, which removes the Ollama launch profile and generated model catalog. Use cli-run for codex --oss. Use cli-run-model <model> for codex --oss -m <model>. Use cli-run-profile for codex --profile ollama-launch.

For Codex CLI model selection, prefer models with at least a 64k context window, per Ollama's guidance.

cli-setup, cli-run, cli-run-model, and cli-run-profile launch interactive Codex CLI sessions. In normal Codex plugin use, run them with --dry-run and tell the user to run the printed command in a real terminal. Only run them directly when the user explicitly asks for nested Codex and OLLAMA_CODEX_ALLOW_INTERACTIVE=1 is set.

Dry Runs

Use --dry-run before mutating App or CLI state when verifying command routing:

bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh --dry-run app-setup
bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh --dry-run app-use-codex-model gpt-5.4
bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh --dry-run app-use-model gemma4:31b
bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh --dry-run app-restore
bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh --dry-run cli-setup
bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh --dry-run cli-config gpt-oss:20b
bash ${CLAUDE_PLUGIN_ROOT}/scripts/ollama-codex.sh --dry-run cli-restore

Restore Boundaries

Treat plugin disablement, Codex App restore, and Codex CLI restore as separate actions:

Disabling or removing this plugin stops the plugin from being available in Codex.
Restoring Codex App requires ollama launch codex-app --restore or app-restore.
Restoring Codex CLI requires ollama launch codex --restore or cli-restore.

Do not imply one restore command affects both profiles.

References

Read references/ollama-codex-docs.md when exact official command behavior, minimum version, model examples, backups, troubleshooting, or Codex CLI versus Codex App distinctions matter.