A small, independent starter for AI-assisted Python/ML development, informed by LibreYOLO's public repository practices. It is not a fork, an official LibreYOLO product, or a reconstruction of private prompts.
Install uv first, then:
uv sync --frozen --dev
make doctor
make check
uv run ai-project
make smokeThe working stack is uv + pytest/xdist/timeout + Ruff + ty + GitHub Actions, with Claude Code/Codex skill-discovery adapters. See software stack and daily loop for core, optional external and hardware tiers.
Without Make, use the exact commands shown in Makefile with uv run --no-sync.
The GPU workflow and cloud/account integrations are not activated.
- How the repository works: roles, files, daily agent loop, review, ML experiments and release handoff.
- Connected software stack: what is implemented versus optional or not activated.
- Free-first tool choices: licenses, popularity evidence, service-cost boundaries and alternatives.
- Task, experiment and review templates.
- Rename the distribution, package directory and CLI in pyproject.toml, src/ and tests/.
- Replace the demo CLI with your first acceptance-tested feature.
- Fill docs/architecture.md and record major decisions in docs/adr/.
- Keep one canonical AGENTS.md; CLAUDE.md forwards to it.
- Use the task-specific skills/ guides and PR evidence template.
- Configure branch protection manually. CI files do not enable required checks.
- Add real model/data/hardware tests when relevant; none are claimed by this starter.
- Agent and reviewer contracts; plan/review/release skills.
- Python packaging demo, hermetic unit checks and installed-wheel smoke test.
- Least-privilege GitHub CI and PR provenance-presence gate.
- Source evidence, test tiers, provenance, and adoption checklist.
No paid tools, hosted GPUs, model downloads, auto-publishing or external agent services are enabled. CI has not run until you push to your own repository.