agent-core

The shared framework every agent depends on. Agents never copy this code — they install it as a versioned package (agent-core==x.y). Change it here, tag a release, and every agent picks it up on its next version bump. That is what keeps every agent identical.

What's inside

Module	Role
agent.py	the Agent assembly object (prompt + skills + tools)
tools.py	the Tool base class — code the LLM can call
skills.py	loads markdown skill docs into the system prompt
brain.py	the LLM tool-use loop (model orchestrates the tools)
runtime.py	starts the HTTP service (every agent is a server)
service.py	FastAPI factory + health/metrics endpoints
intake.py	request acceptance (validate incoming events)
manager.py	orchestrates a run (brain loop, or a custom handler)
monitoring.py	metrics / health (the "line graphs")
connections/	connectors — http ships in core; jira/slack/github/k8s are extras
llm/	the model client (Claude tool-use; keyless NoOp for local runs)
messaging.py + contracts/	dormant seam for future agent-to-agent comms

Develop

pip install uv            # if you don't have it: https://docs.astral.sh/uv/
uv sync --extra dev
uv run pytest

Release

Tag vX.Y.Z and push — .github/workflows/publish.yml builds and publishes the package. Agents then bump agent-core==X.Y.Z to adopt it.