feat: Phase 5 — GitHub Actions for lint + profile re-run

[Jammy2211]

Overview

Phase 5 of the autolens_profiling z_feature roadmap — wire up GitHub Actions for the new repo. Two distinct workflows shipped in the same PR but kept in separate .yml files:

1. Workflow 1 (PR + push to main) — cheap, fast (<5 min), CPU-only: ruff lint + format check, markdown link-rot check, smoke import / dry-run of one script per section.
2. Workflow 2 (manual + on release tag) — re-run the real profile scripts, regenerate JSON+PNG artifacts under results/, call scripts/build_readme.py from Phase 4 to refresh the tables, commit the diff back to main.

Depends on Phase 0 (repo exists, shipped) and benefits from Phase 4 ([issue to-be-created], dashboard build script). If Phase 4 hasn't shipped when Phase 5 runs, Workflow 2's "refresh README" step can stub to just committing the new JSONs; flip on once build_readme.py lands.

Plan

• Add .github/workflows/lint.yml for the PR / push gate.
• Add .github/workflows/profile.yml for the release / manual profile re-run + README refresh.
• Add pyproject.toml with the ruff config copied from PyAutoLens/pyproject.toml for parity across PyAutoLabs repos.
• Add .github/workflows/README.md documenting which workflow does what and how to trigger Workflow 2 manually.
• Wire AUTOLENS_PROFILING_SMOKE=1 env var into every profile-able script so Workflow 1's smoke step short-circuits to a tiny problem size (no real result artifacts produced).
• Document the open design decisions (runner type, cadence, bot identity, failure handling) inline in the workflow YAML comments so future maintainers see the reasoning.

Detailed implementation plan

Affected Repositories

• PyAutoLabs/autolens_profiling (primary, target).
• PyAutoLabs/PyAutoLens (read-only reference — copy .github/workflows/ patterns + pyproject.toml ruff config).

Work Classification

CI / infrastructure (workspace-style — no Python library code changes). Manual worktree setup.

Branch Survey

Repository	Current Branch	Dirty?
./autolens_profiling	main	clean

Suggested branch: feature/ci-actions
Worktree root: ~/Code/PyAutoLabs-wt/ci-actions/

Workflow 1 — lint.yml (PR + push to main)

Should run in <5 min on a GitHub-hosted ubuntu runner with no JAX install. Checks:

• ruff check (Python lint).
• ruff format --check (formatting parity with PyAutoLens / sister repos).
• Markdown link-rot via lychee (cheap, runs offline-friendly).
• Smoke import / dry-run of:
  • The smallest pixelization profile script (probably likelihood/imaging/mge.py with AUTOLENS_PROFILING_SMOKE=1).
  • The cheapest simulator (probably simulators/imaging.py at default instrument with smoke flag).
  • The smallest Nautilus run (searches/nautilus/simple.py with n_live=10).
• The smoke step must not produce real artifacts — short-circuit to tiny problem size and skip the JSON+PNG write, or write to a temp dir that's discarded.

Workflow 2 — profile.yml (manual + release-triggered)

workflow_dispatch (manual) plus on: release: types: [published]. Steps:

1. Install PyAutoLens + dependencies (full install, not smoke).
2. Run each script under likelihood/, simulators/, searches/ — capture stdout to logs, write JSON+PNG to results/.
3. Run python scripts/build_readme.py (from Phase 4) to refresh markdown tables across all READMEs.
4. Commit the diff back to main with author github-actions[bot] and [skip ci] in the subject to avoid loops.
5. If any single profile script fails, mark that cell ERR in the dashboard tables and continue (don't fail the whole workflow on one regression).

Files to produce

• .github/workflows/lint.yml (Workflow 1).
• .github/workflows/profile.yml (Workflow 2).
• .github/workflows/README.md documenting both.
• pyproject.toml carrying the ruff config (copy from PyAutoLens/pyproject.toml).
• Optional .lychee.toml for markdown link-rot config if defaults aren't enough.

Open design decisions

1. Runners: CPU-only on GitHub-hosted runners, or self-hosted GPU runners? Recommend: GitHub-hosted CPU-only first; structure Workflow 2 so a future self-hosted GPU job can append GPU numbers without touching the workflow shape. (HPC-GPU results would land manually for now via a results/*_a100_v*.json upload from the user.)
2. Cadence: every release? weekly cron? manual only? Recommend: manual + on release tag so we don't burn CI minutes on noise.
3. Bot identity: github-actions[bot] with [skip ci] in subject; needs contents: write permission on the workflow.
4. Failure handling: continue-on-error per script + cell marked ERR in the dashboard. Recommend this over "fail whole workflow on one regression".
5. Smoke env var name: AUTOLENS_PROFILING_SMOKE=1 (proposed). Open to a different name if there's a convention in sister repos worth following.

Implementation Steps

1. Set up worktree manually:

   source admin_jammy/software/worktree.sh
   worktree_create ci-actions autolens_profiling
   cd ~/Code/PyAutoLabs-wt/ci-actions/autolens_profiling
   git checkout -b feature/ci-actions

2. Copy pyproject.toml ruff config from PyAutoLens/pyproject.toml. Keep only the [tool.ruff] section — this repo isn't a Python package, so no [project] table.

3. Author .github/workflows/lint.yml (Workflow 1) modelled on PyAutoLens/.github/workflows/.

4. Thread AUTOLENS_PROFILING_SMOKE=1 through every profile script: short-circuit to tiny problem size (e.g. n_live=10 for Nautilus, smallest mask for imaging, single channel for datacube), skip JSON+PNG writes. Touch all 10+ profile scripts.

5. Author .github/workflows/profile.yml (Workflow 2). If Phase 4's scripts/build_readme.py exists, call it; if not, stub the README-refresh step with a TODO comment.

6. Author .github/workflows/README.md documenting both workflows.

7. Smoke locally by running act or gh workflow run against the lint workflow.

8. Open PR against main of PyAutoLabs/autolens_profiling. PR body: link this issue + summary of design choices made + which open decisions were resolved.

Key Files

• New: .github/workflows/lint.yml, .github/workflows/profile.yml, .github/workflows/README.md.
• New: pyproject.toml (or setup.cfg) for ruff config.
• Edit: every profile script (likelihood/**/*.py, simulators/*.py, searches/nautilus/*.py) — add AUTOLENS_PROFILING_SMOKE short-circuit.
• Reference (read-only): PyAutoLens/pyproject.toml, PyAutoLens/.github/workflows/*.yml.

F1 lesson applied

Workflow 2's profile-rerun step regenerates the JSON+PNG artifacts the dashboard reads. Make sure it copies the input datasets from this repo (not from _developer) so the CI run reflects what an end user sees. Phase 1's re-mirror (PR #3) ensures these are clean origin/main snapshots.

Out of scope

• Cross-platform matrix (macOS, Windows) — lens-modeling profiling is Linux-only in practice.
• PyPI / conda packaging — the repo isn't a Python package.
• Coverage reporting — there are no unit tests in this repo, by design.
• Self-hosted GPU runner provisioning — design Workflow 2 so it can be added later without restructuring.

Original Prompt

Click to expand starting prompt

The full prompt content is preserved at PyAutoPrompt/issued/ci_actions.md after this issue is created.