aiskillstore 1.0.0

Python SDK for the Universal Skill Kit (USK) Skill Store API

Skill Store SDK

Python client for the Universal Skill Kit (USK) Skill Store API.

Zero external dependencies — uses Python stdlib only.

Install

pip install aiskillstore

Or copy the skillstore_sdk/ directory into your project.

Quick Start

from skillstore_sdk import SkillStore

store = SkillStore()  # defaults to https://aiskillstore.io

# Search for skills
results = store.search(capability="web_search", min_trust="community")
for skill in results:
    print(skill.name, skill.trust_level, skill.capabilities)

# Get full schema before installing
schema = store.schema(results[0].skill_id)
print(schema.call_example())      # {'query': 'Search query string'}
print(schema.permissions)         # {'network': True, 'filesystem': False, ...}
print(schema.interface)           # {'type': 'cli', 'call_pattern': 'stdin_stdout', ...}

# Download for your platform
path = store.download(results[0].skill_id, platform="OpenClaw", target_dir="./skills/")
print(f"Downloaded to {path}")

# One-liner: search + download
path = store.find_and_download("web_search", platform="ClaudeCode", target_dir="./skills/")

Authentication

Read operations (search, schema, download) require no authentication.

Upload requires an API key obtained from your Skill Store account:

store = SkillStore(api_key="sk_your_key_here")
# or via env var:
# export SKILLSTORE_API_KEY=sk_your_key_here

result = store.upload("./my-skill-1.0.0.skill")
print(result["vetting_report"]["status"])

Configuration

Parameter	Env var	Default
base_url	SKILLSTORE_URL	https://aiskillstore.io
api_key	SKILLSTORE_API_KEY	(none)
timeout	—	30 seconds

store = SkillStore(
    base_url="https://aiskillstore.io",
    api_key="sk_...",
    timeout=60,
)

API Reference

store.info() → dict

Returns service metadata including USK spec version and supported platforms.

store.search(...) → SearchResult

Search for skills by capability, keyword, platform, or trust level.

results = store.search(
    capability="translation",   # semantic capability tag
    q="korean",                 # keyword search
    platform="OpenClaw",        # filter by platform
    usk_v3=True,                # only auto-convertible skills
    trust="verified",           # exact trust level
    min_trust="community",      # minimum trust level
    limit=10,                   # max results (default 20, max 50)
)
# SearchResult is iterable
for skill in results:
    print(skill)  # <Skill my-skill v1.0.0 | verified [USK v3] ⚡>

store.schema(skill_id) → SkillSchema

Fetch the full USK schema for a skill. Use this to evaluate a skill before installing.

s = store.schema("abc123")
s.call_example()           # minimal valid input dict
s.interface                # {'type': 'cli', 'entry_point': 'main.py', ...}
s.input_schema             # JSON Schema dict
s.output_schema            # JSON Schema dict
s.permissions              # {'network': bool, 'filesystem': bool, ...}
s.platform_compatibility   # ['OpenClaw', 'ClaudeCode', 'CustomAgent']

store.download(skill_id, platform, target_dir, filename) → str

Download a skill package. Returns the absolute path to the saved file.

Supported platforms: "OpenClaw" | "ClaudeCode" | "CustomAgent" | "original"

store.upload(skill_file_path) → dict

Upload a .skill package. Requires api_key.

Returns {"status": ..., "skill_name": ..., "version_number": ..., "vetting_report": {...}}.

store.find_and_download(capability, platform, target_dir, trust) → str | None

Convenience: search by capability and download the top result. Returns None if no match.

Trust Levels

Level	Meaning
verified	Admin-approved, manually reviewed
community	Auto-scan passed, community-trusted
sandbox	Unchecked — use with caution

Supported Platforms

Platform	Package Type
OpenClaw	openclaw_wrapper.py + openclaw_manifest.json
ClaudeCode	Markdown slash command + runner.py
CustomAgent	HTTP adapter (/invoke) + openapi.json
original	Original uploaded package

USK v3 skills with can_auto_convert: true are automatically available for all platforms.

Agent Usage Pattern

Agents typically follow this discovery-evaluate-install flow:

from skillstore_sdk import SkillStore

store = SkillStore(base_url="https://aiskillstore.io")

# 1. Discover
results = store.search(capability="web_search", min_trust="community", usk_v3=True)
if not results:
    raise RuntimeError("No web_search skill available")

# 2. Evaluate — check permissions before installing
top = results[0]
schema = store.schema(top.skill_id)
if schema.permissions.get("filesystem"):
    raise RuntimeError("Skill requests filesystem access — not allowed")

# 3. Install
path = store.download(top.skill_id, platform="OpenClaw", target_dir="./skills/")
print(f"Installed: {path}")