This repository is the ThinkFlowLab RSI Team canonical home for OmniRSI and a copyable scaffold for new evidence-driven repos. The playbook lives in docs/repo-template.md; this tree is the concrete skeleton that implements it without baking campaign-specific revision pins into CI or skills.
- Copy the top-level layout (or use GitHub’s “Use this template” once enabled):
AGENTS.md,docs/,recipes/,benchmarks/,evidence/,src/,tests/,skills/,.github/workflows/,pyproject.toml, and thisREADME.md. - Rename the Python package under
src/rsi_scaffold/to your project name (e.g.src/my_inference_lab/). Updatepyproject.toml[project].nameand any workflowpathsfilters if you rename directories. - Read docs/using-the-scaffold.md and keep docs/repo-template.md as the chapter playbook (PM, Arch, RSI, CI, Ops).
- Use Research fill-ins: positioning, claims, downstream, demo gate. Open campaigns via acceptance checklist or the GitHub issue template (pinned SHAs in the issue only—not workflow defaults).
- Wire your four CI lanes (draft → ready → merge → nightly) as documented in docs/scaffold-architecture.md.
| Path | Purpose |
|---|---|
AGENTS.md |
Read first — portable contract for coding agents (CI lanes, L0–L4, evidence bar) |
docs/ |
Playbook + scaffold PRD / architecture / test design |
recipes/ |
Runnable scenario recipes (RunSpec-oriented) |
benchmarks/ |
Campaign and bench harness stubs (no pinned upstream SHAs here) |
evidence/ |
First-class evidence tree: plan/, results/, patch/, validator/ |
src/rsi_scaffold/ |
Minimal package; SIAS seams: harness/, model/, infra/ |
tests/ |
CPU smoke tests for CI |
skills/ |
Portable agent skills (generic RSI / evidence guidance) |
Verdicts are PASS, FAIL, or INCONCLUSIVE. A public PASS requires pinned baseline and candidate revisions, frozen validators, and a complete evidence tree—documented in the playbook, not hardcoded in this template.
python -m pip install -e ".[dev]"
pytest
ruff check src tests- Repository template playbook
- Issue #1 — design
- Campaign acceptance issues own SHAs and gates (see playbook § PM / Test design).