SkillDock Python SDK and CLI

skilldock is an OpenAPI-driven Python client (and a simple CLI) for the SkillDock API.

It loads the OpenAPI spec at runtime and exposes:

• A Python SkilldockClient that can call any operationId
• A CLI that can list available operations and call them from your terminal

Package version: 2026.02.101309 (note: PyPI normalizes versions like 2026.02 to 2026.2 per PEP 440).

Install

pip install skilldock

Quickstart (CLI)

List operations from the OpenAPI spec:

skilldock ops

The first column is a python-friendly python_name you can use as client.ops.<python_name>(...).

Authenticate (browser login + polling):

skilldock auth login

This starts a CLI auth session on the API, prints an auth_url, opens it in your browser, then polls until it receives an app-issued access_token and saves it as the CLI token. Access tokens are short-lived; if you see auth errors later, run skilldock auth login again.

Create a long-lived personal API token (recommended for CI and to avoid short-lived JWT expiry):

# Prints the token (shown only once by the API) and saves it into the CLI config.
skilldock tokens create --save

# List your tokens
skilldock tokens list

Call an endpoint by operationId:

skilldock call SomeOperationId --param foo=bar --json '{"hello":"world"}'

Search skills:

skilldock skills search "docker"

Install a skill locally (default destination is ./skills, with recursive dependency resolution):

Public skills can be installed without auth. If a token is configured, the CLI sends it for skill discovery/install flows so private skills can be resolved when authorized.

# latest
skilldock install acme/my-skill

# exact version
skilldock i acme/my-skill --version 1.2.3

# custom local destination
skilldock install acme/my-skill --skills-dir /path/to/project/skills

Uninstall a direct skill and reconcile/remove no-longer-needed dependencies:

skilldock uninstall acme/my-skill

Verify a local skill folder (packages a zip and prints sha256/size):

skilldock skill verify .

Upload a new skill release:

skilldock skill upload --namespace myorg --slug my-skill --version 1.2.3 --path .

# Explicit private publish
skilldock skill upload --namespace myorg --slug my-skill --version 1.2.3 --path . --visibility private

This packages the folder into a zip and uploads it as multipart form field file. For this registry, tags are read by the backend from SKILL.md frontmatter inside the uploaded zip. There is no separate upload tags field (or CLI --tag flag) for publish.

---
name: my-skill
description: Does X
version: 1.2.0
tags:
  - productivity
  - automation
  - cli
---

Tag values should be strings; non-string values are ignored by the backend. The CLI packages upload archives with top-level folder name equal to --slug, so keep frontmatter name aligned with your slug.

If your API supports release dependencies, you can pass them from CLI too:

# Repeatable string form:
skilldock skill upload --namespace myorg --slug my-skill --path . \
  --dependency "core/base-utils@^1.2.0" \
  --dependency "tools/lint@>=2.0.0 <3.0.0"

# JSON form (array or map), inline or from file:
skilldock skill upload --namespace myorg --slug my-skill --path . \
  --dependencies-json @dependencies.json

If you haven't created the namespace yet:

skilldock namespaces create myorg
skilldock namespaces list

Low-level request (method + path, bypassing operationId):

skilldock request GET /health

Quickstart (Python)

from skilldock import SkilldockClient

client = SkilldockClient(
    # Optional: override if needed
    openapi_url="https://api.skilldock.io/openapi.json",
    # base_url="https://api.skilldock.io",
    token=None,  # set after `skilldock auth login`
)

ops = client.operation_ids()
print("operations:", len(ops))

# Call by operationId (params are split into path/query/header based on OpenAPI metadata)
result = client.call_operation("SomeOperationId", params={"id": "123"})
print(result)

# Or call by a generated python-friendly name:
# (see `skilldock ops` output and use the `python_name`-like identifier)
# result = client.ops.someoperationid(id="123")

client.close()

Configuration

The CLI stores config (including token) in a local JSON file:

skilldock config path
skilldock config show

You can set config values:

skilldock config set --base-url https://api.skilldock.io --openapi-url https://api.skilldock.io/openapi.json
skilldock config set --token "YOUR_TOKEN"

Environment variables (override config):

• SKILLDOCK_OPENAPI_URL
• SKILLDOCK_BASE_URL
• SKILLDOCK_TOKEN
• SKILLDOCK_TIMEOUT_S

Authentication Notes (Google)

This SDK assumes the API accepts a token in an HTTP header (usually Authorization: Bearer <token>). The exact details are derived from the OpenAPI securitySchemes when present.

The SkillDock API can accept (depending on server configuration):

• Google ID token (JWT)
• App-issued access token (JWT, returned by the CLI OAuth flow)
• Personal API token (opaque string, created via skilldock tokens create)

skilldock auth login works like this:

1. Creates a CLI auth session via POST /auth/cli/sessions
2. Prints the returned auth_url and opens it in your browser
3. After you complete Google login, the backend approves the session
4. The CLI polls GET /auth/cli/sessions/{session_id} until it receives an app-issued access_token, then saves it as the configured API token

If you want to set a token manually:

skilldock auth set-token "PASTE_TOKEN_HERE"

To create a personal API token (recommended for longer-lived auth):

skilldock tokens create --save

Development

Run the small unit test suite:

python -m unittest discover -s tests