Run this before publishing a package. The goal is to experience AgentForge like a first-time user.
python3 scripts/release_smoke.pyThis removes old dist/ artifacts, runs tests, runs doctor checks, builds a fresh wheel/sdist, and runs twine check.
Use a clean environment:
python3 -m venv /tmp/agentforge-v1-test
source /tmp/agentforge-v1-test/bin/activate
pip install /path/to/Agentforge/dist/agentforge_harness-0.1.0-py3-none-any.whl
agentforge --versionmkdir -p /tmp/agentforge-demo
cd /tmp/agentforge-demo
agentforge init
agentforge doctorConfirm:
- config is created
- API key guidance is clear
- hosted providers do not ask for a base URL
- custom provider asks for base URL before API key
- doctor output is understandable
- no private machine paths leak into generated config
Start:
agentforgeThen test:
/help
/doctor
/tools
/skills
List the files in this directory.
Create a file called notes.txt with one sentence about AgentForge.
Read notes.txt.
Edit notes.txt to add one more sentence.
/checkpoint
/checkpoints
/export html
/stats
/exit
Confirm:
- approval prompts are readable
- file tools show useful summaries
- failed commands give recovery hints
- TUI text does not overlap badly
/export htmlcreates a readable export
Create a local skill:
.agentforge/skills/demo/SKILL.md
With content:
---
name: demo
description: Use when the user asks for a tiny AgentForge demo response.
---
# Demo
Answer in two short bullet points.Then run:
/reload
/skills
use demo skill and explain AgentForge
/unskill demo
Confirm:
- the skill appears in
/skills - the TUI shows activation reason and file path
- the skill body affects the response
- inactive skills are not all dumped into context
At least one real model call should work for the provider you intend to document as the default.
Test:
Read README.MD if it exists, otherwise explain this empty project.
Confirm:
- streaming works
- tool calls work
- final answer appears
- no provider-specific crash occurs
You are ready to publish an alpha package when:
- local wheel install works
agentforge initworks in a clean directoryagentforge doctorgives actionable output- one real model call works
- read/write/edit tool flow works
- approval prompt is understandable
- skills list and manual activation work
- session export works
- release smoke passes
If any step feels confusing, fix docs or UX before publishing.