gemini-cli.md

DeepRefine Skill for Gemini CLI

DeepRefine-Skill can be installed as a Gemini CLI extension. This adds Gemini slash commands for the existing agent-native DeepRefine workflow while keeping Cursor support unchanged.

Prerequisites

pip install -e /path/to/DeepRefine-Skill
node -v   # Gemini CLI requires Node.js 20+
gemini --version

Recommended development install

From the DeepRefine-Skill repository root:

deeprefine gemini link
# equivalent to: gemini extensions link .

Restart Gemini CLI after linking:

gemini

Then verify inside Gemini CLI:

/extensions list
/commands list

You should see the deeprefine-skill extension and these commands:

/deeprefine
/deeprefine:review
/deeprefine:apply

Install a copied extension

If you do not want to link the working tree:

deeprefine gemini install
# equivalent to: gemini extensions install <bundled-extension-template>

Restart Gemini CLI afterwards.

Manual fallback

If the official Gemini extension manager is unavailable, copy the bundled template:

deeprefine gemini install --copy-only

This places files under:

~/.gemini/extensions/deeprefine-skill/

Prefer deeprefine gemini link or deeprefine gemini install because they use Gemini CLI's official extension manager and are more likely to appear in /extensions list.

Usage

Default pending-query workflow:

/deeprefine

Read-only review:

/deeprefine:review "How does the training pipeline work from data loading to model output?"

Validated apply workflow:

/deeprefine:apply "Apply the validated refinement for the missing train_epoch relation."

Notes

• Gemini CLI authentication is separate from DeepRefine installation.
• If /extensions list shows no extension, exit Gemini CLI, run deeprefine gemini link, and restart Gemini CLI.
• /deeprefine expects a project with graphify-out/graph.json; in this repository itself it may correctly report that Graphify outputs are missing.