Noether Integration

AgentSpec operations are registered as Noether stages — content-addressed, type-safe, composable units of computation.

Why Noether?

Noether is a verified composition platform. Every stage is:

• Content-addressed — same input bytes → same hash, forever
• Type-checked — input/output types validated before execution
• Composable — stages chain into graphs, graphs into bigger graphs
• Trace-recorded — every execution leaves a trace

By making AgentSpec operations Noether stages, you get:

• All AgentSpec ops are reproducible
• They compose with the 370+ other Noether stages (data, AI, web, infra)
• They can be served as HTTP APIs via noether serve
• They can be compiled to standalone binaries via noether build

The 9 AgentSpec stages

noether stage search "agentspec"

Stage	ID	What
agentspec_validate	a61021f8	Schema validation
agentspec_resolve	d979bb0b	Environment negotiation
agentspec_hash	dd86f6e2	Content-addressable hashing
agentspec_merge	9df4c9dc	Inheritance with trust check
agentspec_evolve	2a068663	Retro-based evolution
agentspec_schema	7feb167c	JSON Schema export
agentspec_profile_create	7d22e469	Create profile from manifest
agentspec_profile_retro	d8995d46	Process retro into signed memories
agentspec_profile_export	99be6346	Export profile for registry

Use them locally:

echo '{"manifest": {"name": "test", "apiVersion": "agent/v1"}}' | \
  python3 stages/validate.py

Or via Noether:

noether run validate-graph.json --input '{"manifest": ...}'

Composing stages

Build a graph that validates → resolves → creates a profile:

{
  "name": "agent_pipeline",
  "operator": "Sequential",
  "stages": [
    {"stage": "agentspec_validate", "input": "$manifest"},
    {"stage": "agentspec_resolve", "input": "$.previous"},
    {"stage": "agentspec_profile_create", "input": "$.previous"}
  ]
}

Run as HTTP:

noether serve agent_pipeline.json --port 8080
curl -X POST http://localhost:8080/run \
  -H "Content-Type: application/json" \
  -d '{"manifest": {"name": "test", ...}}'

Or compile to binary:

noether build agent_pipeline.json --output ./agent-validator
./agent-validator --input <manifest.json>

Noether-flavored bases

examples/bases/*-noether.agent — base templates with Noether composition skills baked in:

base: ./claude.agent
merge:
  skills: append

skills:
  - noether-compose
  - noether-run
  - noether-search
  - noether-serve

behavior:
  traits:
    - noether-first      # Prefer Noether compositions over writing code
    - noether-verify     # Lint .nth files before running

The noether-first trait expands to:

Prefer Noether compositions over writing code from scratch. Use noether decompose to break problems into verified stages, noether search to find reusable stages in the library, and noether run to execute compositions. Noether stages are content-addressed and type-safe — reuse over reinvention.

In the registry

The noether-cloud registry stores AgentSpec stages alongside the other 370+:

NOETHER_REGISTRY=https://registry.agentspec.dev \
  noether stage search "agentspec"

When you push a .agent manifest, it's wrapped as a Noether stage:

• name: agent:my-researcher
• tags: [agentspec, agent-manifest, ...user_tags]
• implementation: <full manifest as JSON>

agentspec push and agentspec pull use this convention transparently.

Why this architecture wins

You don't need a separate AgentSpec registry, server, or infrastructure. You don't need a separate marketplace. Everything runs on the existing Noether infrastructure:

• Registry → Noether stage store (Postgres + semantic search)
• API server → noether serve
• Distribution → noether build to binaries/WASM
• Composition → mix AgentSpec ops with data/AI/web stages
• Verification → content-addressing + signing

One platform, one set of infrastructure, one ecosystem.