Adopt the analysis-workflow plugin; commit notebook outputs; zarr

.claude/settings.json

@@ -0,0 +1,24 @@
+{
+    "extraKnownMarketplaces": {
+        "quadbio": {
+            "source": {
+                "source": "github",
+                "repo": "quadbio/claude-plugins"
+            }
+        }
+    },
+    "enabledPlugins": {
+        "analysis-workflow@quadbio": true
+    },
+    "permissions": {
+        "deny": [
+            "Bash(git add -A:*)",
+            "Bash(git add --all:*)",
+            "Bash(git reset --hard:*)",
+            "Bash(git clean -f:*)",
+            "Bash(git clean -fd:*)",
+            "Bash(git push --force:*)",
+            "Bash(git commit --no-verify:*)"
+        ]
+    }
+}

.github/workflows/lint.yaml

@@ -26,5 +26,3 @@ jobs:
           frozen: true
       - name: Run pre-commit checks
         run: pixi run pre-commit run --all-files
-      - name: Verify notebooks contain no outputs
-        run: git ls-files -z '*.ipynb' | xargs -0 -r pixi run nbstripout --verify

.gitignore

@@ -150,8 +150,23 @@ dmypy.json
 *.gmt
 *.gmx
 
-# Directories to ignore
+*.zarr
+
+# Data never enters git; only its READMEs and placeholders do.
+data/**
+!data/**/
+!data/**/README.md
+!data/**/.gitkeep
+
+# Directories to ignore. Unanchored, so they also match inside every analysis task directory
+# (analysis/<topic>/.../<name>_vN/): the untracked half of a task, anchored to the main checkout.
 figures/
+outputs/
+logs/
+
+# A task's results/ (small evidence tables) is tracked even though *.csv / *.txt are ignored above.
+!analysis/**/results/**/*.csv
+!analysis/**/results/**/*.txt
 
 # OS specifics
 **.DS_Store

AGENTS.md

@@ -1,39 +1,36 @@
 # AGENTS.md — working conventions
 
-This file owns the working conventions and is the canonical guidance for any coding agent.
-`README.md` is the user-facing overview; anything documented there is referenced from here,
-never restated.
+This file is the canonical guidance for any coding agent. `README.md` is the user-facing overview;
+anything documented there is referenced from here, never restated.
 
-## Layout
+## Workflow
 
-- **Notebooks**: `analysis/[INITIALS]-[YYYY]-[MM]-[DD]_description.ipynb`
-- **Data**: `data/<dataset>/{raw,processed,resources,results}/`, gitignored
-- **Package**: `src/<package>/`, installed editable from the checkout
+The human and agent lanes, task lifecycle, output paths, `data/` layout, working objects and their
+write-back are owned by the [analysis-workflow](https://github.com/quadbio/analysis-workflow)
+plugin, enabled in `.claude/settings.json`: load its skill before writing analysis code, outputs or
+data. Pull requests are reviewed against [`REVIEW_GUIDE.md`](REVIEW_GUIDE.md).
 
-## Paths
+## This repo
 
-Never hardcode a path into `data/` or `figures/` — every path hangs off `FilePaths`:
+<!-- Replace with this project's facts: its datasets and where each working object lives,
+environment specifics, companion code packages. Rules the plugin owns are not restated here. -->
+
+- **Package**: `src/<package>/`, installed editable from the main checkout
+- **Paths**: every dataset path hangs off `FilePaths` in `src/<package>/_constants.py`:
 
 ```python
 from myanalysis import FilePaths
 
-FilePaths.DATA           # data/
-FilePaths.FIGURES        # figures/ — curated output: talk and paper figures
-FilePaths.EXAMPLE_DATASET / "processed" / "adata.h5ad"
+FilePaths.EXAMPLE_DATASET / "processed" / "adata.zarr"
 ```
 
-`FilePaths.ROOT` is resolved from git, so it names the *main* checkout even when called from a
-worktree and shared data does not follow your branch. Add a dataset as a constant in
-`_constants.py`; each one keeps the `{raw,processed,resources,results}` layout by convention.
-
 ## Environments
 
-Dependencies live in `pixi.toml`, not `pyproject.toml` — the latter carries package metadata and
-the test config. Run `pixi install` after pulling a change to `pixi.toml`, in the main checkout.
+Dependencies live in `pixi.toml`, not `pyproject.toml`, which carries package metadata and the
+test config.
 
 | Task | Command |
 | --- | --- |
 | Run Python | `pixi run python script.py` |
 | Run tests | `pixi run test` |
-| Add conda package | `pixi add <package>` |
-| Add PyPI package | `pixi add --pypi <package>` |
+| Add a conda / PyPI package | `pixi add <package>` / `pixi add --pypi <package>` |

README.md

@@ -83,14 +83,13 @@ Then finish by hand:
 
 ```bash
 pixi install                   # create environment from pixi.toml
-pixi run install-hooks         # pre-commit hooks + notebook output-stripping filter
+pixi run install-hooks         # pre-commit hooks
 pixi run install-kernel        # register Jupyter kernel
 ```
 
-> `install-hooks` also sets up the [nbstripout](https://github.com/kynan/nbstripout)
-> git filter. Notebook outputs are then stripped from commits automatically while
-> staying in your working copy. **Run it once in every clone** (including remote
-> servers and worktrees), or outputs may slip into git.
+> **Notebook outputs are committed.** They are the record of what a notebook actually
+> produced, and GitHub renders them. Keep them small: clear a notebook by hand before
+> committing if it carries a huge embedded image or an accidental dump.
 
 💡 **Tip**: Use `pixi shell` to enter the environment interactively—then you can run commands directly without the `pixi run` prefix.
 
@@ -126,6 +125,19 @@ git push
 
 ---
 
+## 🤖 Working with coding agents
+
+Agents follow [`AGENTS.md`](AGENTS.md) and the
+[analysis-workflow](https://github.com/quadbio/analysis-workflow) Claude Code plugin (a skill plus
+guard hooks) that this repo enables. Install the plugin once per machine:
+
+```bash
+claude plugin marketplace add quadbio/claude-plugins
+claude plugin install analysis-workflow@quadbio
+```
+
+---
+
 ## ☕ Daily Workflow
 
 ```bash
@@ -187,12 +199,12 @@ Or edit `pixi.toml` directly and run `pixi install`.
 <summary><strong>📓 Data and notebook conventions</strong></summary>
 
 - **Notebook naming**: `[INITIALS]-[YYYY]-[MM]-[DD]_description.ipynb`
-- **Data layout** (one folder per dataset):
-    - `data/<dataset>/raw/` — original data files
-    - `data/<dataset>/processed/` — preprocessed data
-    - `data/<dataset>/resources/` — reference data, annotations
-    - `data/<dataset>/results/` — analysis outputs
-- **Figures**: `figures/` or `data/<dataset>/results/`
+- **Data layout** (one folder per dataset, gitignored):
+    - `data/<dataset>/raw/` — bytes as they arrived; never written by analysis
+    - `data/<dataset>/resources/` — curated inputs that are not data: gene lists, marker tables
+    - `data/<dataset>/processed/` — objects code loads to do new work, as AnnData `.zarr`
+    - `data/<dataset>/results/` — a notebook's own outputs, file names prefixed with the notebook's stem
+- **Figures**: `figures/<topic>/`
 - **Import paths** via the local package:
 
 ```python
@@ -212,13 +224,6 @@ This template uses **pre-commit hooks** to automatically check your code before
 | [Biome](https://biomejs.dev/) | Formats JSON/JSONC files |
 | [pyproject-fmt](https://github.com/tox-dev/pyproject-fmt) | Formats `pyproject.toml` |
 
-**Notebook outputs** are handled separately by an [nbstripout](https://github.com/kynan/nbstripout)
-git *filter* (not a pre-commit hook), set up by `pixi run install-hooks`. The filter
-strips outputs from the committed copy while leaving them in your working tree, so
-your notebooks stay rendered locally but git history stays clean. CI fails the build
-if a notebook with outputs ever lands in the repo (a clone that skipped
-`install-hooks`), via `nbstripout --verify`.
-
 Hooks run automatically on `git commit`. To run manually:
 
 ```bash

REVIEW_GUIDE.md

@@ -0,0 +1,90 @@
+# Review Guide
+
+PR review playbook for **review agents running on GitHub**. Imperative voice.
+
+**Scope: review only.** Comment and suggest. Do not push commits or apply fixes.
+
+> Everything above "This repo" comes from the analysis template, which takes it from the analysis-workflow
+> plugin; change it there and pull it with `cruft update`. Reviewers cannot load the plugin, which is why this
+> copy exists.
+
+The code has usually already run. That is not a reason to lower the bar: requiring a rerun in better shape is a
+normal outcome. Say what to redo, not just what was wrong.
+
+## Data governance
+
+Every write to a shared object (a working object under `data/*/processed/`, a SpatialData store, anything other
+code loads):
+
+- **Accumulate by addition.** Adding keys to freshly re-read state is commutative, so concurrent sessions can't
+  lose each other's work. Removal isn't: that means a new dated copy.
+- **Writing back to a shared object needs sign-off on that specific diff.**
+
+## Priorities
+
+### 1. Writes that lose data, or silently do nothing
+
+A write to a working object goes through `commit_adata`, gated behind a dry run. Flag:
+
+- an in-memory AnnData written back over a working object (`write_zarr`/`write_h5ad` onto its path): it erases
+  whatever other sessions added since it was read;
+- a recomputed key committed under an existing name: `commit_adata` writes only keys not yet on disk, so it is a
+  no-op and the PR reports a result the object doesn't hold. Use a new key or a new dated copy;
+- a replaced SpatialData image, labels, points or shapes element, where a new element name would do.
+
+### 2. Promotions, which leave no diff
+
+`data/` is gitignored, so an object written into `data/<dataset>/processed/` appears nowhere in the diff; the task
+README's `Write-back` section is the only trace. The right subdirectory is not guessable. Check that the PR *says*
+where it went and that it was agreed, while moving it is still cheap.
+
+### 3. The task contract
+
+Agent work belongs in a task directory `analysis/<topic>/.../<name>_vN/` with a `README.md` naming real inputs,
+outputs and write-back keys. Outputs go through `task_paths(__file__)` to the task's own `results/`, `reports/`,
+`figures/`, `outputs/`. Writes to `data/<dataset>/results/` or the central `figures/` are the human lane and are a
+finding.
+
+Every key committed to a shared object carries the task's version suffix **and** appears in that README: an
+unrecorded key can't be traced back, an unsuffixed one collides with the next version.
+
+### 4. Scientific correctness
+
+The point of the repo. Check that the claim in the PR body follows from what the code computes: the right cells,
+the right grouping, a control where one is needed, and a comparison not confounded by condition, time point,
+sample or batch. A number that reaches a figure is worth more scrutiny than anything below.
+
+### 5. Reinvention
+
+New code needs a reason to exist. Look outward before accepting it: scverse and the Python ecosystem, then the
+repo's sibling code packages, then earlier task directories; a near-match found by grep counts. Name the
+candidate and what's wrong with it rather than concluding nothing fits.
+
+The acute case is a helper defined twice in one task or copied between tasks: copies drift, and when they compute
+a quantity compared *across* tasks, the result is a wrong conclusion. Copying figure code between versions of one
+task is fine.
+
+### 6. Conciseness
+
+Prose is reviewed like code. READMEs, docstrings, comments and the PR body: no restating the diff, no filler
+preamble, no exhaustive caveats. Padding rots the same way dead code does, and these docs are load-bearing for
+the next agent.
+
+## Do not report
+
+- Anything the CI linters already gate: formatting, imports, line length.
+- Lock files, and anything under `data/`.
+- Committed notebook outputs: review changed cells' code, never the output blobs.
+- Absolute paths built from `main_checkout()`: task outputs are anchored there deliberately.
+- Worktree, temp-dir and cwd-relative `data/`/`figures/` path literals in `analysis/` code: a hook blocks them
+  before the file is written.
+
+## Drift older than the diff
+
+Parts of a repo may predate these rules. When a PR touches such a file, report the drift once at the lowest
+severity and don't block; it gets fixed when a task next touches it. Search the open issues before filing one.
+
+## This repo
+
+<!-- Repo-specific rules go here: its shared stores and their write helpers, its companion packages, its
+confounders. -->

analysis/ML-2026-01-27_demo_scRNA_workflow.ipynb

@@ -497,11 +497,7 @@
    "id": "37",
    "metadata": {},
    "outputs": [],
-   "source": [
-    "output_path = FilePaths.EXAMPLE_DATASET / \"processed\" / \"pbmc3k_processed.h5ad\"\n",
-    "adata.write(output_path)\n",
-    "print(f\"Saved to: {output_path}\")"
-   ]
+   "source": "output_path = FilePaths.EXAMPLE_DATASET / \"processed\" / \"pbmc3k_processed.zarr\"\nadata.write_zarr(output_path)\nprint(f\"Saved to: {output_path}\")"
   },
   {
    "cell_type": "markdown",

analysis/XX-2026-01-27_sample_notebook.ipynb

@@ -127,9 +127,7 @@
   {
    "cell_type": "markdown",
    "metadata": {},
-   "source": [
-    "Store data in `data/<dataset>/` with subfolders `raw/`, `processed/`, `resources/`, `results/`. Access paths via `FilePaths`:"
-   ]
+   "source": "Store data in `data/<dataset>/` (layout in the README) and access paths via `FilePaths`:"
   },
   {
    "cell_type": "code",

data/example_dataset/README.md

@@ -1,6 +1,4 @@
 # Dataset structure
 
--   `raw`: Raw state of the data we received.
--   `processed`: Processed / intermediate data.
--   `resources`: Reference data, gene sets, annotations.
--   `results`: Any results we compute for this dataset.
+`raw/`, `resources/`, `processed/` and `results/`, as described under *Data and notebook
+conventions* in the top-level `README.md`.