Skip to content

kou2525-wow/nucleus-spec

Nucleus Specification

Agent Operating Core for Production AI Systems

Nucleus is a production-grade operating core for AI agents.
It replaces static instruction files with a five-layer model for:

  • purpose and hard boundaries
  • runtime control
  • environmental adaptation
  • auditable evolution
  • multi-agent coordination

When Do You Need Nucleus?

You probably already have prompt files, instruction docs, or a CLAUDE.md that tells your AI what to do. That works. Nucleus is not a replacement for that.

Nucleus is for when "what to do" is not enough, and you need to define "what must not happen."

Situation Prompt files / CLAUDE.md Nucleus
Solo developer, one AI assistant Sufficient Not needed
Defining tasks and workflows Good fit Not the right tool
Enforcing hard limits that must never be broken Best-effort Structurally enforced
Tracking why agent behavior changed over time Git diff Lineage with reason and approver
Multiple people operating the same agent Tribal knowledge Shared, typed constraints
Providing an agent as a product to others Fragile Portable specification
Multi-agent coordination Implicit Explicit Protocol
Compliance or audit requirements Unstructured Structured and immutable

The honest answer: If you are one person iterating on your AI setup through conversation, you do not need Nucleus yet. Nucleus becomes valuable when constraints need to survive beyond a single person's memory -- when other people, other agents, or future-you must respect the same boundaries without re-deriving them.

What Nucleus adds to what you already have

Prompt files and instruction docs define behavior: what the agent should do. Nucleus defines constraints: what the agent must not do, how far it can go, and who must approve what.

They are complementary, not competing.

CLAUDE.md / Skills  =  "Do this, this way"        (action)
Nucleus Charter     =  "Never do this"             (boundary)
Nucleus Runtime     =  "Use these tools, this budget" (constraint)
Nucleus Adaptation  =  "Change only within these rules" (governance)
Nucleus Lineage     =  "Here is why it changed"    (audit)

Core Thesis

Prompt as Document -> Prompt as Runtime

Nucleus does not eliminate prompts.
It governs prompt activation as part of a runtime control system.

One-Line Definition

Nucleus is the operating core that makes AI governable, auditable, and evolvable in production.

SwarmOps: The Operational Counterpart

Nucleus defines boundaries. SwarmOps defines the workflow.

SwarmOps is an operational methodology for AI-driven development built on three pillars:

  1. Branch + Cloud + DB synchronization -- every AI work unit gets an isolated environment
  2. AI-driven E2E testing -- AI writes and runs tests for its own output
  3. Vibe -> Spec -> E2E loop -- from human intent to verified implementation

Nucleus provides the constraints. SwarmOps provides the execution cycle. Together, they form a complete model for AI that ships safely.

Start Here

  1. Read docs/NUCLEUS_SPEC_v1.1.md
  2. Read docs/SWARMOPS.md
  3. Open ai/ONBOARDING_GUIDE.md
  4. Use the starter templates in /templates
  5. Customize your own Charter, Runtime, Adaptation, and Protocol
  6. Validate changes through CI and pull requests

About

A production-grade operating core for AI agents. Nucleus defines purpose, boundaries, runtime control, adaptation, lineage, and protocol for governed AI systems.

Topics

Resources

License

Code of conduct

Contributing

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors