README
# Triage Agent — a real agent loop with tools
A small agent that searches issues, adds labels and can comment — with a hard step budget so it can never loop forever. The model is injected as a plain decide() function, which makes the loop testable without an API key.
## Run it
python3 run_demo.py
python3 -m pytest -q # 3 passed## What to learn here
- A tool is a name, a description and a function. Keep tools small and typed.
- The loop records every action and observation in a trace, so failures are debuggable.
max_stepsis a safety rail, not an afterthought.- The model never executes code directly: it returns a decision, and the loop validates it.
- For real use, replace
scripted_decidewith an LLM tool-calling call and persist writes
behind human approval.
MIT licensed.
Files in this download
agent.py
data/issues.json
README.md
requirements.txt
run_demo.py
tests/test_agent.py
tools.py
Push this project to GitHub
Create a repository and upload every file as one commit. Use it for your portfolio, a code review, or your first deploy.
Classic token: tick `repo`. Fine-grained token: give this repository Contents → Read and write and Metadata → Read. Create a token · Fine-grained token
Your token is stored in this tab's sessionStorage and sent only to api.github.com. keepUp AI never receives it.
No GitHub account? The project ZIP works on its own — push later whenever you like.
All projects
Generated from the same content the course teaches. If a lesson changes, its download and tests change with it.