Skip to content

AGENTS.md

Latest commit

 

History

History
51 lines (40 loc) · 2.6 KB

AGENTS.md

File metadata and controls

51 lines (40 loc) · 2.6 KB
Raw

PyAutoCTI Workspace Test — Agent Instructions

This is the integration-test suite for PyAutoCTI, run in CI to verify the core library works end-to-end. It is not a user-facing workspace — see ../autocti_workspace for examples and tutorials. These are the canonical, agent-agnostic instructions for this repo.

Dependencies: autocti, autofit, autoarray, and arcticpy (source-only C++ sdist — install with pip install arcticpy==2.6 --no-build-isolation --no-deps after numpy+cython, with libgsl-dev present; see PyAutoCTI/AGENTS.md).

Repository Structure

scripts/                     Integration-test scripts run in CI
  dataset_1d/model_fit.py    1D calibration: simulate -> factor-graph fit -> aggregator round-trip
  imaging_ci/model_fit.py    2D charge injection: simulate -> factor-graph fit -> result inspection
  plot/subplots.py           Drives the autocti.plot function surface to files
legacy/                      Euclid VIS heritage (2022-2023, pre-resurrection API — not runnable;
                             see legacy/README.md)
config/                      Current PyAutoCTI config (mirrors autocti_workspace)
config/build/env_vars.yaml   Per-script env for smoke runs (PYAUTO_TEST_MODE=2 defaults)
smoke_tests.txt              The curated smoke list (small on purpose)

Running

python .github/scripts/run_smoke.py     # the smoke list, with env_vars.yaml applied
python scripts/dataset_1d/model_fit.py  # one script, real search (no env applied)

CI runs the smoke list through PyAutoHeart's reusable smoke workflow (thin caller in .github/workflows/smoke_tests.yml, chain PyAutoNerves PyAutoFit PyAutoArray PyAutoCTI; the arcticpy build lives in .github/scripts/smoke_install.sh).

Conventions

  • Keep smoke_tests.txt a small curated subset — do not mass-promote scripts.
  • Integration scripts are self-contained (simulate their own data, small shapes) and single-trap (identical-prior ordered-trap models tie at prior medians under the PYAUTO_TEST_MODE=2 bypass and raise their own assertion — a filed autofit issue; avoid the pattern here).
  • The test-mode knob is PYAUTO_TEST_MODE (2 bypasses sampling); PYAUTOFIT_TEST_MODE does not exist.
  • Never edit legacy/ — it is preserved Euclid VIS history.

Never rewrite history

Never rewrite pushed history on any repo with a remote — no git init over a tracked repo, no force-push to main, no fresh-start "Initial commit", no filter-repo / filter-branch / rebase -i on pushed branches. To get a clean tree: git fetch origin && git reset --hard origin/main && git clean -fd.