Agent Development Overview
Developer Reference
This page covers internal implementation details. It is not included in the User Guide.
The ADK (Agent Development Kit) is the authoring and registration toolchain for domain teams. It validates and registers agent.yaml packages into the Context Engine.
Two roles
Role 1 — Author → domain expert uses the Authoring Agent → produces agent package
Role 2 — Register → ADK CLI reads package → validates → places in CE collectionsThe handoff between them is a file on disk: the assembled agent package.
Domain Expert ←→ Gateway → Authoring Agent (POST /api/v1/author, port 8085)
↓
assembled package (agent.yaml + skill.yamls + playbook.yamls + tools)
↓
python -m adk register PATH [--dry-run]
↓
Context Engine collections (via /register/* CRUD)
↓
Platform Framework (runtime)What the ADK does NOT do
- It does not execute skills — that is the Platform Framework's job
- It does not manage knowledge retrieval at runtime — that is the Context Engine's job
- It does not manage client-side estate data — that is the Context Manager's job
- It does not write to any database directly — all writes go through the CE CRUD API
Getting started
bash
# Install the ADK
cd v4-adk
python3 -m venv venv
source venv/bin/activate
pip install -e .
# Scaffold a new agent
adk scaffold --domain security --name my-agent
# Validate offline (no CE required)
adk validate ./my-agent/
# Register into Context Engine
adk register ./my-agent/Next steps
- agent.yaml Reference — The full canonical spec
- Authoring Workflow — The Author Agent conversation
- ADK Reference — All CLI commands