NATIVEHELIX / OPEN-SOURCE AI-NATIVE TOOLKIT
NativeHelix
Reusable Skills and Capabilities for governed agent systems.
NativeHelix turns practical AI-native product and architecture methods into installable Agent Skills, capability contracts, patterns, checklists, templates, and examples—while keeping project context, permissions, validation, and consequential authority with the adopting team.
01 / TOOLKIT BOUNDARY
Reusable material outside the project's authority boundary.
NativeHelix supplies design material. The adopting project remains authoritative for its users, outcomes, architecture, permissions, tools, credentials, data, costs, and release decisions. Copying a Skill does not grant access or execution authority.
Outcome and local constraints
Users, architecture, permissions, tools, data, cost, and release decisions.
Select reusable material
4 Skills15 Domains
Validate in the real environment
State, recovery, evidence, permissions, and product value.
Project outcome and local constraints lead to selecting reusable NativeHelix material, followed by project-local adaptation and validation; consequential authority remains with the project and identifiable humans.
02 / FOUR ENTRY-POINT SKILLS
Start from the work—not from the entire library.
Public Core v0.1.0 includes four self-contained Agent Skill packages. Each has a defined entry point and bounded purpose; a project can choose one without importing every Capability asset.
Four installable Agent Skill packages
- 01design-ai-native-architecture
Define system layers, boundaries, interfaces, permissions, evidence, recovery, and evolution.
- 02design-agent-workflow
Turn a frozen product outcome into tasks, states, gates, ownership, and stop conditions.
- 03review-agent-harness
Review execution control, evaluation, cost, safety boundaries, and readiness evidence.
- 04ingest-project-capabilities
Extract reusable practices while preserving source, maturity, ownership, and evidence boundaries.
The four separate entry points cover architecture design, workflow design, harness review, and capability ingestion.
03 / ADOPTION SEQUENCE
Select narrowly. Adapt locally. Validate before delegation.
The public guidance describes a bounded adoption sequence: define the outcome, select only what is needed, adapt it to real constraints, separate evidence from assumptions, validate it in context, and retain human approval for consequential actions.
From project need to human-approved use
- 01Define the outcome
Name the user outcome and final artifact.
- 02Select narrowly
Choose one Skill or a small set of Capability assets.
- 03Adapt locally
Fit the material to actual architecture, permissions, tools, data, and cost.
- 04Separate evidence
Keep observed facts distinct from assumptions and recommendations.
- 05Validate in context
Test state, recovery, evidence, permissions, and product value where the work will run.
- 06Retain authority
Keep human approval for consequential or external actions.
The adoption sequence defines an outcome, selects narrowly, adapts locally, separates evidence, validates in context, and retains human authority.
PUBLIC CORE BOUNDARY
A documentation and toolkit release—not a runtime.
Public Core v0.1.0 has passed bounded package and reference checks. The repository does not claim that every Skill has been independently forward-tested across multiple hosts.
WHAT IT PROVIDES
- Four installable Agent Skill packages
- Fifteen Capability domains with contracts, patterns, checklists, and templates
- Public governance and role-boundary guidance
- An AI-native reference architecture and a synthetic example
WHAT IT DOES NOT CLAIM
- Not an SDK, runtime, or hosted service
- Not a certification, security control, or compliance method
- Not a production-ready reference implementation
- No automatic filesystem, network, credential, tool, or external-write permission
START WITH ONE BOUNDED NEED
Choose one Skill. Keep project authority local.
The README and Getting Started guide explain the public boundary and adoption path. Use only the assets your project needs, then validate them inside the environment where they will run.
OPEN A CONVERSATION
Build something worth testing.
For AI-native products, global GTM, or independent projects, choose a channel below.