Skip to content
Scion

GCP Skill Registry

The GCP Skill Registry lets you store and serve skills through Google Cloud’s Vertex AI Agent Platform. Once imported, skills can be referenced from any scion-agent.yaml using the gcp-skill:// URI scheme — no Scion Hub required.

This tutorial walks through setting up the registry, importing skills from a GitHub repository, and verifying the results using the helper scripts in scripts/skill-bank/gcp/.

Prerequisites

Section titled “Prerequisites”
  • A GCP project with the Agent Platform API (aiplatform.googleapis.com) enabled
  • Python 3.8+
  • Application Default Credentials configured
  • The scripts from scripts/skill-bank/gcp/ in your local checkout

Enable the API

Section titled “Enable the API”
Terminal window
gcloud services enable aiplatform.googleapis.com --project=YOUR_PROJECT_ID

Authenticate

Section titled “Authenticate”
Terminal window
gcloud auth application-default login

Or, for a service account:

Terminal window
export GOOGLE_APPLICATION_CREDENTIALS=/path/to/key.json

Your account needs the roles/aiplatform.user IAM role on the project.

Install dependencies

Section titled “Install dependencies”

The scripts require a few Python packages:

Terminal window
pip install -r scripts/skill-bank/gcp/requirements.txt

This installs:

  • google-auth — GCP authentication
  • google-auth-httplib2 — HTTP transport for auth
  • requests — HTTP client
  • pyyaml — YAML parsing for SKILL.md frontmatter

Configure the registry

Section titled “Configure the registry”

Edit scripts/skill-bank/gcp/config.py to point at your project and region:

PROJECT_ID = "your-gcp-project-id"
LOCATION = "us-central1"
API_BASE = f"https://{LOCATION}-aiplatform.googleapis.com/v1beta1"
SKILLS_PARENT = f"projects/{PROJECT_ID}/locations/{LOCATION}"
SKILLS_URL = f"{API_BASE}/{SKILLS_PARENT}/skills"

Change PROJECT_ID to your GCP project and LOCATION to your preferred region (e.g. us-central1, europe-west1).

Verify registry access

Section titled “Verify registry access”

The Skill Registry is a project-level service — there is no separate creation step. Run create_registry.py to confirm the API is enabled and accessible:

Terminal window
python3 scripts/skill-bank/gcp/create_registry.py

Expected output:

Project:  your-gcp-project-id
Location: us-central1
Endpoint: https://us-central1-aiplatform.googleapis.com/v1beta1/projects/your-gcp-project-id/locations/us-central1/skills


Registry is accessible. 0 skill(s) found.

If you see a 403 or 404 error, double-check that the API is enabled and your credentials have the roles/aiplatform.user role.

Import skills from a GitHub repository

Section titled “Import skills from a GitHub repository”

The import_skills.py script clones a skill repository, discovers all valid skill directories, and uploads each one to the registry:

Terminal window
python3 scripts/skill-bank/gcp/import_skills.py

Example output:

Cloning https://github.com/mattpocock/skills.git...


Found 24 skills to import.


  Importing engineering/api-design... OK
  Importing engineering/code-review... OK
  Importing engineering/debugging... OK
  Importing misc/markdown-formatting... OK
  Importing productivity/time-management... ALREADY EXISTS (skipped)
  ...


Results: 22 imported, 2 already existed, 0 failed
Total skills in repo: 24

How skill discovery works

Section titled “How skill discovery works”

The script walks the cloned repository looking for directories that contain a SKILL.md file. Each SKILL.md must have YAML frontmatter with name and description fields:

---
name: code-review
description: Guidelines for thorough code reviews.
---


# Code Review


Review checklist and best practices...

The entire skill directory (including any supporting files) is zipped and uploaded to the registry. Skills that already exist in the registry are skipped (HTTP 409).

Verify the import

Section titled “Verify the import”

List all skills in the registry to confirm the import succeeded:

Terminal window
python3 scripts/skill-bank/gcp/list_skills.py

Example output:

Listing skills in your-gcp-project-id / us-central1


Name                                               Display Name                   State
------------------------------------------------------------------------------------------
api-design                                         api-design                     ACTIVE
code-review                                        code-review                    ACTIVE
debugging                                          debugging                      ACTIVE
markdown-formatting                                markdown-formatting            ACTIVE
time-management                                    time-management                ACTIVE


Total: 5 skill(s)

Using imported skills in templates

Section titled “Using imported skills in templates”

Once skills are in the GCP registry, reference them from a scion-agent.yaml using the gcp-skill:// URI scheme:

scion-agent.yaml
schema_version: "1"
name: my-agent


skills:
  - uri: gcp-skill://my-registry/code-review
  - uri: gcp-skill://my-registry/api-design@1.0.0
    optional: true

The gcp-skill:// scheme follows the format gcp-skill://<alias>/<skillId>@<version>, where:

  • alias — the registry alias configured in your skill federation settings
  • skillId — the skill name as registered in GCP
  • version — an optional version specifier (defaults to latest)

For the federation setup that connects your Scion Hub to a GCP registry, see Skill Registry & Federation.

See also

Section titled “See also”