my-agent

Starter repo for a new agent. Clone it, then change a few details — that's the whole workflow. No special tooling required.

An agent is: a prompt (its role), skills (markdown capabilities the LLM follows), and tools (code the LLM can call). The framework (agent-core) runs an LLM tool-use loop over them.

Create a new agent

1. Clone/copy this repo under your agent's name; set the name in pyproject.toml, Makefile, and deploy/helm/values.yaml (and AGENT_NAME).
2. Give it a brain: set AGENT_MODEL (e.g. claude-opus-5) and provide ANTHROPIC_API_KEY.
3. Edit the three things that define what it does:
   • src/agent/prompts/system.md — its role
   • src/agent/skills/*.md — its capabilities (plain language)
   • src/agent/tools/tools.py — the code it can call

Everything else (wiring, server, Dockerfile, Helm, CI) is inherited from agent-core and rarely touched.

Run locally

uv sync
cp .env.example .env          # set AGENT_MODEL + ANTHROPIC_API_KEY for a real brain
uv run agent                  # serves on http://localhost:8080

curl -sX POST localhost:8080/run -H 'content-type: application/json' \
  -d '{"input":"echo hello"}'

Ship (Kubernetes)

make image
make deploy