skillsvault 0.1.0

Govern agent skills from any Python framework — detect, install, gate, audit.

skillsvault Python SDK

Govern agent skills from any Python framework — LangChain, LangGraph, the OpenAI Agents SDK, LlamaIndex, or plain code. These have no skills directory and no harness hook, so they connect through this SDK instead.

Same control plane as the CLI and the JS SDK: sync() the signed policy bundle once, then gate(skill) locally before each skill/tool use (no model, no network on the hot path). governed() wraps a tool so a banned skill raises before it runs and every use is audited.

pip install skillsvault

Only dependency is cryptography (Ed25519 bundle verification); HTTP uses the standard library.

Model

from skillsvault import create_client

sv = create_client(
    base_url="https://skillsvault.io",
    org="acme",
    key=os.environ["SKILLSVAULT_AGENT_KEY"],  # Settings → API keys
    harness="langchain",                       # free-form framework id
    machine="prod-agent-7",
)

sv.sync()                                # verify the Ed25519-signed bundle
skills = sv.list_skills()                # org-approved skills
content = sv.get_skill("acme/deploy")    # SKILL.md + files at runtime
decision = sv.gate("acme/deploy")        # Decision(decision="allow"|"warn"|"halt", reason, ...)

Framework adapters

governed(skill, fn) returns a wrapped fn — drop it in wherever a framework expects the tool's function. A halt raises SkillsvaultError before the tool runs; the model sees the org's reason and stops. Works for sync and async functions.

LangChain

from langchain_core.tools import StructuredTool

def _deploy(env: str) -> str:
    return deploy_to(env)

deploy = StructuredTool.from_function(
    name="deploy",
    description="Deploy the app",
    func=sv.governed("acme/deploy", _deploy),
)

LangGraph

from langgraph.prebuilt import ToolNode

tools = [StructuredTool.from_function(
    name="deploy", description="Deploy the app",
    func=sv.governed("acme/deploy", _deploy),
)]
tool_node = ToolNode(tools)   # a halted skill raises inside the graph node

OpenAI Agents SDK

from agents import function_tool

@function_tool
def deploy(env: str) -> str:
    ...

# wrap the underlying callable
deploy.on_invoke_tool = sv.governed("acme/deploy", deploy.on_invoke_tool)

Plain code

run = sv.governed("acme/deploy", lambda env: deploy_to(env))
run("prod")   # raises SkillsvaultError if the org halts acme/deploy

API

Call	Description
create_client(base_url, org, key, *, machine, harness, timeout, retries)	Construct the client.
sync()	Pull + verify the signed policy bundle. Call at startup and to refresh.
gate(skill) -> Decision	Local, deterministic allow/warn/halt.
governed(skill, fn)	Wrap a tool: gate + audit, raise on halt. Sync or async.
list_skills()	The org's approved skills.
get_skill(name)	Current-version content (SKILL.md + files), or None.
audit(event)	Append an audit event (best-effort).
evaluate(bundle, skill)	Pure gate function, if you hold a bundle yourself.

Decision is a frozen dataclass: decision, reason, policy_id, policy_name.

License

MIT