feat: euclid mode — pipeline-paired skills + euclid literature wiki

[Jammy2211]

Overview

Extend autolens_assistant with a euclid mode: a domain layer that pairs the public euclid_strong_lens_modeling_pipeline (the collaboration-facing repo of standard Euclid modeling pipelines) to the assistant, plus a dedicated Euclid literature wiki covering both the strong-lensing Euclid papers and the mission/instrument/data papers. euclid_strong_lens_modeling_pipeline is being consolidated as the repo that reproduces the Euclid paper's results; once euclid mode is in place, real lenses will be modeled through it and the layer iterated against the science paper's needs. The euclid_assistant paper type-setting/editing tools are explicitly out of scope.

Plan

• Phase 0 — pipeline cleanup: remove the no-longer-required profiling/ and skills/ folders from euclid_strong_lens_modeling_pipeline, sweeping any references. Small standalone PR.
• Phase 1 — euclid skills layer: a euclid_*.md skills family in autolens_assistant pairing the pipeline's surfaces (setup/install, data preparation, the staged modeling scripts, workflow products, HPC runs) to the assistant, registered in skills/README.md with an AGENTS.md routing note.
• Phase 2 — dedicated euclid literature wiki: autolens_assistant/wiki/euclid/ following the wiki/literature/ schema, populated with the strong-lensing, mission/instrument, EXT-survey and photo-z/ZPC papers; bib entries mined verbatim from euclid_assistant's euclid.bib.
• Later (out of this task): model a small fraction of lenses through euclid mode and iterate on features the science paper needs.

Approved decisions: "mode" is implemented as skills + wiki (NOT a modes/ interaction preset — those stay teacher/assistant only). Euclid_DR1_impact_image_processing.pdf is ingested as a sources page + bib entry only; the PDF itself stays out of the public repo (collaboration-internal draft).

Detailed implementation plan

Affected Repositories

• autolens_assistant (primary)
• euclid_strong_lens_modeling_pipeline (Phase 0 cleanup; Phases 1–2 read-only pairing target)
• euclid_assistant (read-only source: euclid.bib, 296 entries; also knowledge/sources/euclid_assets/euclid.bib)

Branch Survey

Repository	Current Branch	Dirty?
./autolens_assistant	main	clean
./euclid_strong_lens_modeling_pipeline	main	clean

Suggested branch: feature/assistant-euclid-mode
Work Classification: Workspace
Worktree root: ~/Code/PyAutoLabs-wt/assistant-euclid-mode/ (created by /start_workspace)

Implementation Steps

Phase 0 — euclid_strong_lens_modeling_pipeline cleanup (own PR)

1. git rm -r profiling/ skills/ (profiling/ holds delaunay/simulator profiling scripts; skills/ holds a stale start-new-project copy).
2. Sweep references: grep README.md, start_here.py, smoke_tests.txt, tests/, config/, __init__.py, activate.sh for profiling/skills mentions and update.
3. Smoke-check nothing imports the removed modules.

Phase 1 — euclid skills layer in autolens_assistant (own PR)

1. Read the pipeline end-to-end first: scripts/ (initial_lens_model.py, sersic_lens_model.py, mge_lens_only.py, lens_model_waveband.py, full_model.py), preprocess/, workflow/ (csv_make.py, fits_make.py, png_make.py), hpc/, dataset/, start_here.py, util.py — the skill inventory is finalized from what the scripts actually do.
2. Author skills/euclid_*.md following skills/_style.md and _bootstrap_skill.md (Orient → Ask → Branch → Combine), expected shape:
   • euclid_setup_pipeline.md — clone/install the pipeline, activate.sh, dataset layout, start_here.py.
   • euclid_prepare_data.md — preprocess/ → modeling-ready dataset; cutouts/PSF/noise conventions (per-tile, sky-coordinate-dependent PSFs).
   • euclid_model_lens.md — the staged modeling progression (initial_lens_model → sersic_lens_model → mge_lens_only → lens_model_waveband → full_model); when to use which stage.
   • euclid_workflow_products.md — workflow/ csv/fits/png result products.
   • euclid_hpc_runs.md — hpc/sync CLI, SLURM submission, pulling results.
     (Merge/split as the reading dictates; each skill cites pipeline paths as Project:path.)
3. Register each skill in skills/README.md; add .claude/skills/ symlinks; add an AGENTS.md routing note so Euclid requests load the euclid layer.
4. Run python autoassistant/audit_skill_apis.py over any PyAuto* code blocks in the new skills.

Phase 2 — wiki/euclid/ literature wiki (own PR)

1. Mirror the wiki/literature/ schema (AGENTS.md there is canonical): index.md, concepts/, entities/, sources/, bibliography/.
2. bibliography/euclid.bib: copy entries verbatim from euclid_assistant/euclid.bib for the papers below; resolve any missing ones from public ADS/arXiv metadata — never fabricate; keep bibkey_aliases.yaml mapping prompt-style names → canonical keys.
3. Pages (paper list from the prompt):
   • entities/: euclid-mission (Mellier+25 = EuclidSkyOverview), vis (Cropper+25), nisp (Jahnke+25), euclid-wide-survey (Scaramella+22 Scaramella-EP1), q1/dr1 releases, ext-surveys (HSC Miyazaki+18, CFIS Ibata+17, Pan-STARRS Chambers+16, DES Abbott+21), ou-phz (Desprez+20 Desprez-EP10).
   • concepts/: euclid-psf (tile/sky-coordinate dependence — McCracken+25/26, Polenta+25/26), zero-point-corrections, psf-homogenisation / aperture photometry (Romelli+25, Boucaud+16 Wiener-filter kernels), euclid-photo-z (Tucci+25).
   • sources/: euclid-strong-lensing.md (Walmsley+25a, Rojas+25, Lines+25, Li+25, Holloway+25, Ecker+26 Q1-Ecker, Acevedo Barroso+25 AcevedoBarroso24, O'Riordan+25 ORiordan25), euclid-forecasts.md (Amara & Réfrégier 07, Ma+06, Huterer+06, Kitching+08), euclid-mission-data.md (Kümmel+26, Romelli+25, McCracken, Polenta, Desprez), euclid-dr1-image-processing.md (ingested from Euclid_DR1_impact_image_processing.pdf — content only, PDF NOT committed).
4. Cross-link with [[slug]]; append provenance to log.md; link the euclid wiki from the Phase-1 skills.

Key Files

• euclid_strong_lens_modeling_pipeline/profiling/, skills/ — removed (Phase 0)
• autolens_assistant/skills/euclid_*.md — new skills family (Phase 1)
• autolens_assistant/skills/README.md, AGENTS.md — registration + routing (Phase 1)
• autolens_assistant/wiki/euclid/** — new sub-wiki (Phase 2)
• euclid_assistant/euclid.bib — read-only bib source (296 entries)
• autolens_assistant/wiki/literature/AGENTS.md — schema to mirror

Privacy notes

• autolens_assistant is public: no PyAutoMemory references, no collaboration-internal PDFs, no personal content.
• 2025/2026 Euclid Collaboration bib entries ride verbatim from the bib (public preprint metadata).

Original Prompt

Click to expand starting prompt

Extend autolens_assistant with a euclid mode

Type: feature
Target: workspaces
Repos:

• autolens_assistant
• euclid_assistant
• euclid_strong_lens_modeling_pipeline
  Difficulty: too-large
  Autonomy: supervised
  Priority: high
  Status: formalised

In the project euclid_strong_lens_modeling_pipeline, we have all the examples used for modeling euclid data.
This is the github project I put standard pipelines out there to the collaobration so they can perform the
modeling themselves.

However, science was done at the path /mnt/c/Users/Jammy/Science/euclid, but this accumulated a lot of extra
scripts and code which is not really important for general Eulcid science.

I want to consoliate euclid_strong_lens_modeling_pipeline as a repo which has all the necessary code and infrastructure
to reproduce the resutls of the euclid paper. For now, lets assume euclid_strong_lens_modeling_pipeline is complete,
what I will do is I will gradually do a series of tasks in order to validate I get the results I want.

First remove the "profiling" and "skills" folders, which are no longer required.

The real task, is to extend the autolens_assistant with a "euclid" mode, which has the following aims:

1. It uses the euclid_strong_lens_modeling_pipeline to model euclid data, which means it will need its own wiki of skills which
   pair the euclid modeling scripts to the assistant.

2. For the literature have a dediciaged euclid_wiki which contains all the strong lensing euclid papers for reference
   but also non lensing euclid papers describing the instrument and other parts of the data.

Look in euclid_assistant/*/euclid.bib and use these papers:

Amara & Réfrégier 2007;
Euclid Collaboration: Mellier et al. 2025;
Ma et al. 2006; Huterer et al. 2006;
Kitching et al. 2008;
Euclid Collaboration: Tucci et al. 2025;
Euclid Collaboration: Cropper et al. 2025;
Euclid Collaboration: Jahnke et al. 2025;
Euclid Collaboration: Walmsley et al. 2025a;
Euclid Collaboration: Rojas et al. 2025;
Euclid Collaboration: Lines et al. 2025;
Euclid Collaboration: Li et al. 2025;
Euclid Collaboration: Holloway et al. 2025;
Euclid Collaboration: Ecker et al. 2026;
Euclid Collaboration: Romelli et al. 2025;
Euclid Collaboration: Kümmel et al. 2026

These papers describe EXT data which is part of Euclid Data:

Combining data from the Subaru Hyper Suprime Camera (HSC; Miyazaki et al. 2018) for the g and z bands; the
Canada–France Imaging Survey (CFIS; Ibata et al. 2017) for the u and r bands; and the Panoramic Survey
Telescope and Rapid Response System (Chambers et al. 2016, Pan-STARRS) for the i band. In the Southern
Hemisphere, we use imaging from the Dark Energy Survey (DES; Abbott et al. 2021) in the griz bands.

The PSF of a particular band is unique to the target depending on its tile and sky coordinates
(Euclid Collaboration: McCracken et al. 2025, 2026; Euclid Collaboration: Polenta et al. 2025, 2026).

The ZPCs were calculated by the Euclid Photometric-Redshift Organisation Unit (OU-PHZ; Euclid Collaboration:
Desprez et al. 2020).

Also put PyAutoLabs/Euclid_DR1_impact_image_processing.pdf in there.

The standard approach to calculate aperture photometry across multiple wavebands is to first homogenise the
PSFs by generating convolution kernels that match higher-resolution images to the lowest-resolution band.
For example, Euclid Collaboration: Romelli et al. (2025) employed the kernel creation algorithm of
Boucaud et al. (2016), which builds convolution kernels based on Wiener filtering with a tunable
regularisation parameter.

The Euclid satellite will detect 1.5 billion galaxies over the Euclid Wide Survey (EWS, Euclid Collaboration:
Mellier et al. 2025; Euclid Collaboration: Scaramella et al. 2022). With an area of 14 000 deg2 (Euclid
Collaboration: Mellier et al. 2025), a IE PSF of 0.16" (Euclid Collaboration: McCracken et al. 2026; Euclid
Collaboration: Cropper et al. 2025), as well as three near-infrared bands providing crucial colour
information (Euclid Collaboration: Jahnke et al. 2025), the survey will revolutionise strong lensing.

(Acevedo Barroso et al. 2025; O'Riordan et al. 2025)

There are lots of papers above with key context on euclid strong lensing but also the instruments, data,
photo-zs etc.

Look at the euclid_assistant but note that, for now, the goal is not to have its type setting and editing
tools for papers to make it into autolens_assistant — this is just to help euclid strong lens modeling.

Once euclid mode is in place, I will then start modeling a small fraction of lenses and we can iterate
on what features and functionality need adding given the science paper.

[Jammy2211]

Shipped (PRs open, awaiting merge) — 2026-07-16

PRs

• chore: remove profiling/ and skills/ folders; fix stale scripts docs euclid_strong_lens_modeling_pipeline#27 — Phase 0: remove profiling/ + skills/, fix stale scripts docs
• feat: euclid mode — pipeline-paired skills + dedicated euclid wiki (#73) #74 — Phases 1+2: euclid_* skills family + wiki/euclid/ sub-wiki

Summary

Euclid mode is implemented end-to-end: five euclid_* skills pair the pipeline's surfaces (setup, data prep, staged modeling, workflow products, HPC) to the assistant, and a dedicated wiki/euclid/ sub-wiki carries the mission/instrument/EXT/photo-z context plus the Q1 strong-lensing Discovery Engine series, with the bibliography copied verbatim from the collaboration's shared euclid.bib and the 9 pre-Euclid classics verified against public arXiv/Crossref metadata.

Session Notes

• One assistant PR instead of two: Phases 1 and 2 share the feature/assistant-euclid-mode branch and reference each other (skills point at wiki pages), so they ship as separate commits (a4f04f8, 9eebceb) in one PR rather than one PR per phase.
• DR1 draft handling: Euclid_DR1_impact_image_processing.pdf (Fogliardi et al. 2026, in prep) was ingested as content only — wiki/euclid/sources/euclid-dr1-image-processing.md marks all values preliminary; the PDF is NOT committed to this public repo, and no bib entries were written for in-prep DR1 papers (Aussel/Lines/Holloway/Nightingale/Nersesian/Vincken/Kümmel/McCracken/Polenta 2026) until public metadata exists.
• "Mode" architecture: implemented as a skills family + sub-wiki, not a modes/ interaction preset (those stay teacher/assistant), per plan approval.
• Heart gate: shipped through an unrelated RED ("PyAutoLens: 1 uncommitted source change(s)") on explicit user acknowledgement; no scripts changed in either PR (markdown/deletions/docstrings), pipeline python verified with compileall, citations validated.
• Next: model a small fraction of lenses through euclid mode and iterate on features the science paper needs (user-driven follow-up).

[Jammy2211]

Both PRs merged (euclid_strong_lens_modeling_pipeline#27, autolens_assistant#74) — euclid mode is live. Closing; feature iteration continues as new tasks once real lenses are modeled through it.