Skip to content

refactor: AI_POLICY/CONTRIBUTING back to root (PyAutoMind#253) (#244) #81

refactor: AI_POLICY/CONTRIBUTING back to root (PyAutoMind#253) (#244)

refactor: AI_POLICY/CONTRIBUTING back to root (PyAutoMind#253) (#244) #81

Workflow file for this run

name: Brain Tests
# Brain's own unit suite — the reasoning layer's self-test. Until this existed,
# PyAutoBrain PRs carried ZERO check runs (`docs.yml` is path-filtered to
# `docs/**`, `nightly-release.yml` is a cron release scheduler), so ~190 tests
# covering the conductors, faculties, worktree helpers, skill install and policy
# seams ran in no CI at all and a Brain PR's only gate was whatever the
# authoring session happened to run locally.
#
# pytest plus ONE drift leg. The suite must not run `pyauto-brain <agent>`
# against live repos or reach the network — those need the full workspace and
# belong to the scheduled health/nightly drivers, not a PR gate. The suite is
# stdlib + PyYAML only, so it stays fast (~30s) and flake-free. The one
# non-pytest step is the tenant-firewall gate (PyAutoMind#198): every firewall
# finding to date merged through a green PR because no PR CI ran the check, so
# the leg a Brain PR can cause is verified here, in the PR that authors it.
# `--only` keeps Mind-side legs (map blocks, registries) from reddening Brain
# PRs on drift Brain cannot cause; organs not checked out (Heart, Hands) are
# skipped by the check itself, so this gates exactly Brain.
#
# TWO repos are checked out on purpose. PyAutoBrain alone cannot even COLLECT
# the suite: `agents/faculties/sizing/_sizing.py` reads the body map
# (`BRAIN_HOME.parent / "PyAutoMind" / "repos.yaml"`) at import time and is
# deliberately strict, so `test_policy_seams.py` and `test_sizing_paths.py`
# error out before a single test runs. Checking PyAutoMind out as a sibling
# reproduces the workspace layout the code assumes — the same layout the
# firewall gate's `--root` needs. Both repos are public, so the default
# GITHUB_TOKEN suffices — no PAT. No other sibling repo is needed (verified by
# running the suite against exactly this two-repo layout).
# One run per commit: PR events carry the CI; pushes only build main.
# Superseded runs are cancelled on PR refs only — a cancelled main run would
# read as red CI (cancelled is in Heart's FAILURE_CONCLUSIONS).
on:
push:
branches: [main]
pull_request:
concurrency:
group: brain-tests-${{ github.ref }}
cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
jobs:
pytest:
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
python-version: ["3.12", "3.13"]
steps:
- name: Checkout PyAutoBrain
uses: actions/checkout@v4
with:
path: PyAutoBrain
# The body map `_sizing.py` imports. Pinned to main: this gate tests
# Brain against the CURRENT organism identity, the same pairing a local
# workspace has.
- name: Checkout PyAutoMind (the body map — required to import)
uses: actions/checkout@v4
with:
repository: PyAutoLabs/PyAutoMind
path: PyAutoMind
- name: Set up Python ${{ matrix.python-version }}
uses: actions/setup-python@v5
with:
python-version: ${{ matrix.python-version }}
- name: Install (pytest + PyYAML — the whole dependency set)
run: pip install pytest PyYAML
- name: Run tests
working-directory: PyAutoBrain
run: pytest tests/ -q
- name: Tenant firewall (instance facts stay in declared config surfaces)
run: >
python3 PyAutoMind/scripts/repos_sync.py --check
--only "tenant firewall (organ code)"
--root "$GITHUB_WORKSPACE"