Agent Claude Code
Use this agent when the user requests to update, annotate, or write notes for a Python file that has an associated markdown documentation file. Triggering scenarios include:\n\n1. Explicit requests like 'update the note for torchcell/datasets/scerevisiae/sameith2015.py' or 'annotate the md file for this Python…
Command Claude Code
Relevant files: experiments/005-kuzmin2018-tmi/scripts/GOgraphdatefiltercomparison.py experiments/005-kuzmin2018-tmi/scripts/dcellgoperturbationvisualization.py torchcell/graph/graphanalysis.py /Users/michaelvolk/Documents/projects/torchcell/notes/assets/images/gographunfilteredvisualization2025-05-13-17-54-25.png…
Command Claude Code
We are writing a pipeline for training the Dango Model. The way they set lambda is based on the difference.
Command Claude Code
The torchcell project is migrating from a label-based data format to a COO (Coordinate) format for phenotype data. In the old format, phenotype values were stored as data["gene"]["fitness"] or data["gene"]["geneinteraction"]. In the new COO format, all phenotype values are stored in a single tensor with accompanying…
Command Claude Code
We are writing the experimental pipeline for Dcell. It is very similar to the pipeline for the Dango model. We want to keep it similar to the Dango pipeline as possible. We want to record all of the same metrics, use the same code structure, etc.
Command Claude Code
We are writing the experimental pipeline for Dcell. It is very similar to the pipeline for the Dango model. We want to keep it similar to the Dango pipeline as possible. We want to record all of the same metrics, use the same code structure, etc.
Command Claude Code
We have written the experimental pipeline for Dcell. It is very similar to the pipeline for the Dango model. have kept it similar to the Dango pipeline as possible. We kept a record or all of the same metrics, use the same code structure, etc.
Command Claude Code
We're working on improving the Gene Ontology (GO) term handling in TorchCell, specifically focusing on properly integrating GO annotations from multiple sources within the graph representation module. The goal is to ensure that the Ggo graph in TorchCell correctly incorporates GO terms from both raw SGD (Saccharomyces…
Command Claude Code
Command "hetero_cell_bipartite_dango_gi_refactor" from Mjvolk3/torchcell, covering heterocellbipartitedangogi model refactoring plan, current status summary, context, key files and completed tasks.
Command Claude Code
The GeneInteractionDango model experiences NaN values during training when using ICLoss in the Lightning training pipeline, but works fine with LogCosh loss and also works fine with ICLoss when run standalone (outside the Lightning pipeline).
Command Claude Code
A general-purpose dataset class for inference that follows the Neo4jCellDataset pattern but removes the complexity of database queries, conversion, deduplication, and aggregation. This dataset uses LMDB for efficient storage and retrieval of experiments, following the same format as Neo4jCellDataset. Put the new class…
Command Claude Code
Command "load_coo_context" from Mjvolk3/torchcell, covering new data, old transforms with named phenotypes and old test results.
Command Claude Code
We are riding a pipeline for training the Dango Model.
Command Claude Code
@/Users/michaelvolk/Documents/projects/torchcell/torchcell/data/experimentdataset.py @/Users/michaelvolk/Documents/projects/torchcell/torchcell/datasets/scerevisiae/kuzmin2018.py @/Users/michaelvolk/Documents/projects/torchcell/torchcell/datasets/scerevisiae/costanzo2016.py…
Command Claude Code
We are refactoring pipeline for training the Dcell Model. We want to base it off the structure of the recent Dango Model Pipeline. The issue is that we have had a decent amount of updates since working with this code.
Command Claude Code
We have refactored the pipeline for training the Dcell Model. We have based it off the recent Dango Model Pipeline.
Settings file Claude Code
Agent settings declaring 17 allowed tools.
Skill Claude CodeCodex
Commit staged changes with an auto-generated bulleted commit message. Run /stage first to stage files.
Skill Claude CodeCodex
Move a path (dataset, LMDB build, schema, code, note) into the deprecation graveyard instead of deleting it. Use when retiring old data structures or stale builds -- especially before a rebuild -- since rm -rf of directories is blocked. Wraps scripts/deprecate.sh (move + provenance manifest); nothing is auto-deleted…
Skill Claude CodeCodex
Run the deterministic merge-queue drainer once (land queued worktree branches to main). Normally triggered automatically by /enqueue-merge and a cron; use this only to force a drain by hand.
Skill Claude CodeCodex
Author a NEW draw.io diagram from scratch by writing the .drawio XML directly -- at true Nature print size, in the repo palette, verified by headless render. Use when asked to draw/diagram a mechanism, workflow, schematic, or concept figure that is not a matplotlib plot. For refreshing plot panels already placed in a…
Skill Claude CodeCodex
Enqueue a finished worktree branch and watch it land through the single-writer merge queue from this session -- resolving any rebase conflict here, and printing a distinct dove signal when it is merged + cleaned up + safe to close. Takes a branch name (defaults to the current worktree's branch).
Skill Claude CodeCodex
Audit and refresh the matplotlib SVG panels embedded in a draw.io figure (notes/assets/drawio/.drawio.svg) without opening the GUI -- swap regenerated panels in at true size, rebuild the preview/PDF, and rebuild editing.pdf if it is a paper figure. Use after re-running plot scripts, or to check which panels of a…
Skill Claude CodeCodex
Pull OCR literature artifacts from the keyed torchcell literature endpoint and verify sha256. Use on the M1 Mac (or any client) to list citation keys, fetch a paper's manifest/markdown/SI, or download an artifact with integrity checking.