BioRouter Skills

A collection of skills for BioRouter — UCSF's local-first AI workspace for biomedical research.

Skills are reusable instruction bundles that BioRouter loads on demand to shape how the model writes code, drafts documents, reasons about a problem, and runs research. Some run automatically when relevant files are touched (style enforcement, language conventions, bioinformatics tasks). Others are user-invocable and respond to a slash command like /ralph, /auto-research, or /anti-ai-writing in the chat panel.

This repository hosts the skills curated by the Baranzini Lab at UCSF. Each top-level folder is a single skill; a few folders (superpowers/, bio-skills/) are bundled collections of many sub-skills.

---

Installing a skill

1. Download the skill's .zip from the Releases page.
2. Unzip it.
3. Move the resulting folder into ~/.config/biorouter/skills/.
4. Open BioRouter. The skill appears under Skills automatically. User-invocable skills respond to /skill-name.

You can also drag a .zip (or a single SKILL.md) onto Skills → Add Skill inside BioRouter, or clone individual folders directly into ~/.config/biorouter/skills/ to track updates with git.

---

Core skills

General-purpose skills — research, writing, planning, code style, and design.

Skill	Trigger	What it does
scientific-research	/scientific-research	End-to-end research pipeline: scope → literature review → hypotheses → analysis plan → verification → cited write-up. Orchestrates bio-skills and the BioRouter extensions; never fabricates citations or results.
empirical-research-router	/empirical-research-router	Stage-gated empirical research workflow: design register, data audit, identification, estimation, robustness, replication, and reporting.
causal-identification-gates	/causal-identification-gates	Audits whether RCT, observational, IV, RDD, DiD, synthetic-control, matching, DML, or survival designs support causal language.
replication-package-audit	/replication-package-audit	Checks research packages for clean-run reproducibility, relative paths, environment capture, seeds, logs, and table/figure mapping.
claim-evidence-integrity	/claim-evidence-integrity	Verifies that manuscript claims, numbers, citations, figures, and tables are grounded in traceable evidence.
citation-temporal-integrity	/citation-temporal-integrity	Confirms citations are real, relevant, current enough for the claim, and temporally valid.
empirical-benchmark-harness	/empirical-benchmark-harness	Provides known-answer benchmark patterns for testing whether empirical-analysis agents avoid common causal and reproducibility failures.
scientific-artifacts	/scientific-artifacts	Plans and verifies DOCX/PDF/PPTX/XLSX/LaTeX/scientific figure artifacts with source-to-exhibit mapping.
open-science-review	/open-science-review	Reviews data/code availability, reporting checklists, protocols, ethics/privacy notes, and open-science readiness.
scientific-data-engineering	/scientific-data-engineering	Plans scalable scientific data workflows with Polars, Dask, Zarr, TileDB, Arrow/Parquet, HDF5, and dataset versioning.
statistical-modeling-bayes	/statistical-modeling-bayes	Designs and critiques statistical, Bayesian, survival, hierarchical, power, and uncertainty-reporting workflows.
scientific-machine-learning	/scientific-machine-learning	Plans scientific ML with leakage checks, transformers, graph neural networks, active learning, SHAP, and reproducibility gates.
geospatial-science	/geospatial-science	Handles geospatial analysis with CRS checks, spatial joins, rasters/vectors, maps, and spatial statistics.
neurophysiology-analysis	/neurophysiology-analysis	Plans biosignal and neural-recording analysis: ECG/EDA/PPG, spike sorting, Neuropixels metadata, and peri-event QC.
quantum-scientific-computing	/quantum-scientific-computing	Routes quantum and scientific simulation work across Qiskit, Cirq, PennyLane, QuTiP, and validation checks.
materials-informatics	/materials-informatics	Guides computational materials workflows with structures, crystals, compositions, phase diagrams, and dataset provenance.
biomedical-imaging-pathology	/biomedical-imaging-pathology	Plans DICOM/pathology/PACS workflows with tiling, metadata, QC, cohort provenance, and privacy-safe reporting.
citation-library-management	/citation-library-management	Builds, cleans, verifies, and deduplicates Zotero/BibTeX/RIS/CSL citation libraries.
laboratory-automation-planning	/laboratory-automation-planning	Plans Opentrons/cloud-lab workflows with plate maps, protocol simulation, sample metadata, and execution gates.
econometrics-toolkit	/econometrics-toolkit	Plans applied econometrics in Python, R, or Stata: panel data, IV, DiD, RDD, clustering, and replication tables.
systematic-review-prisma	/systematic-review-prisma	Runs systematic-review workflows: PICO, search strings, screening logs, PRISMA flow, extraction, risk of bias, and synthesis.
simulation-optimization	/simulation-optimization	Plans discrete-event simulation, multi-objective optimization, sensitivity analysis, scheduling, and what-if modeling.
physical-science-computing	/physical-science-computing	Routes astronomy, physics, units, coordinate systems, spectra, numerical experiments, and physical-science workflows.
metabolic-modeling-cobrapy	/metabolic-modeling-cobrapy	Guides COBRApy/FBA metabolic modeling with GPR rules, media constraints, knockouts, objectives, and provenance.
drug-discovery-benchmarks	/drug-discovery-benchmarks	Plans ADMET, virtual-screening, molecular-property, scaffold-split, leakage, and assay-provenance benchmark checks.
clinical-decision-support-review	/clinical-decision-support-review	Reviews clinical decision support, treatment plans, and reports for evidence grounding, safety scope, and no-PHI handling.
scientific-visual-communication	/scientific-visual-communication	Plans schematics, posters, slides, figure panels, infographics, visual abstracts, and source-to-visual traceability.
grant-funding-strategy	/grant-funding-strategy	Structures research grants, specific aims, milestones, reviewer-risk responses, and funder-fit checks.
regulatory-quality-systems	/regulatory-quality-systems	Reviews scientific software, device, and lab processes against quality, validation, traceability, and design-control gates.
gpu-compute-optimization	/gpu-compute-optimization	Plans GPU-bound training/inference, batching, profiling, memory pressure, mixed precision, and reproducible compute.
research-evaluation-venues	/research-evaluation-venues	Evaluates papers, scholars, venues, impact claims, reviewer fit, and publication positioning.
qualitative-thematic-analysis	/qualitative-thematic-analysis	Plans qualitative coding, thematic analysis, codebooks, memoing, inter-rater checks, and audit trails.
financial-accounting-econometrics	/financial-accounting-econometrics	Plans empirical finance/accounting workflows with event studies, abnormal returns, panel data, and table replication.
stata-research-workbench	/stata-research-workbench	Plans Stata .do/.dta research workflows: audits, logs, tables, graph export, replication, and R&R updates.
survey-weights-complex-samples	/survey-weights-complex-samples	Handles complex survey designs with strata, PSUs, replicate weights, domain analysis, calibration, and weighted inference.
psychometrics-sem-measurement	/psychometrics-sem-measurement	Plans reliability, CFA/SEM, mediation/moderation, invariance, item analysis, and APA-style measurement reporting.
social-science-text-as-data	/social-science-text-as-data	Plans dictionaries, topic models, embeddings, sentiment, supervised coding, LLM-label validation, and measurement-error checks.
econ-finance-data-sources	/econ-finance-data-sources	Routes empirical econ/finance sources including WRDS, CRSP, Compustat, SEC/EDGAR, NBER, SSRN, and public datasets.
empirical-paper-submission-rr	/empirical-paper-submission-rr	Plans journal targeting, referee reports, response letters, revise-and-resubmit execution, and appendix organization.
structural-economics-theory	/structural-economics-theory	Routes game theory, equilibrium, DSGE/HANK, structural IO, calibration, simulation, and identification-proof workflows.
causal-heterogeneity-policy-learning	/causal-heterogeneity-policy-learning	Plans CATE, causal forests, meta-learners, policy learning, conformal causal inference, fairness, and subgroup reporting.
arbor-optimization	/arbor-optimization	Runs evaluator-driven branch-and-prune exploration for improving research artifacts, prompts, analyses, and designs.
hugging-science-resource-discovery	/hugging-science-resource-discovery	Finds and evaluates Hugging Face scientific models, datasets, Spaces, and papers with license, provenance, and benchmark checks.
molecular-dynamics	/molecular-dynamics	Plans MD setup and trajectory analysis with OpenMM, GROMACS, AmberTools, MDAnalysis, MDTraj, and reproducibility checks.
general-time-series-forecasting	/general-time-series-forecasting	Plans non-omics forecasting with classical models, ML, foundation models, backtesting, leakage checks, and uncertainty.
bids-neuroimaging	/bids-neuroimaging	Audits BIDS neuroimaging datasets, derivatives, validation, metadata, and reproducible preprocessing readiness.
clinical-ai-modeling	/clinical-ai-modeling	Plans clinical AI modeling workflows with EHR cohorts, validation, bias checks, calibration, safety, and no-PHI handling.
computational-fluid-dynamics	/computational-fluid-dynamics	Plans CFD workflows with meshes, boundary conditions, turbulence models, solver validation, and convergence checks.
anti-ai-writing	/anti-ai-writing	Pass/fail checklist that strips AI tells out of prose, articles, and essays.
taste-skill	Auto (frontend design)	Anti-slop frontend design — landing pages, portfolios, dashboards, and paper companion sites that don't look templated.
develop-biorouter-extension	/develop-biorouter-extension	Step-by-step guide for building a .brxt extension — manifest, MCP server, bundled skills.
ralph	/ralph	End-to-end planning for the Ralph autonomous agent loop — generates a markdown PRD from a feature idea, then converts it to prd.json sized for one-story-per-iteration execution. Adapted from snarktank/ralph (MIT), itself based on Geoffrey Huntley's Ralph pattern.
ucsf-hpc	/ucsf-hpc	SSH setup, SLURM templates (CPU/GPU/H200), file transfer, and module management for the UCSF CHPC cluster.
python-scripting	Auto (Python)	Enforces Python naming, typing, error handling, and project-structure conventions.
r-scripting	Auto (R)	Tidyverse conventions and documentation standards for R.
ggplot-visualization	Auto (R plotting)	Applies the lab's ggplot2 style whenever R plotting code is written.
superpowers	Mixed	A bundled collection of 13 engineering-discipline skills (brainstorming, TDD, systematic debugging, parallel agents, plan writing, code review, git worktrees, more).

---

Developer & authoring skills

Engineering, review, and skill-authoring tools — adapted from Anthropic's official Claude Code plugins (Apache 2.0). Each upstream plugin ships as slash-commands and subagents (none of them carry lifecycle hooks); BioRouter has no commands/ or agents/ loader, so each plugin's commands and subagents are folded into one self-contained SKILL.md the model loads on demand. hookify is the exception — BioRouter has a hook engine, so it writes genuine, enforced hooks in <project>/.biorouter/hooks.yaml instead of honor-system notes.

Skill	Trigger	What it does
code-review	/code-review	Multi-stage, confidence-scored review of a pull request (or local diff); drops low-confidence findings and posts a formatted PR comment via gh.
code-simplifier	/code-simplifier	Simplifies recently-changed code for clarity while preserving exact behavior; defers language style to python-scripting/r-scripting.
commit-commands	/commit-commands	Three git workflows — clean single commit; commit + push + open PR; and clean up [gone] branches and their worktrees.
skill-creator	/skill-creator	The canonical guide to authoring BioRouter skills — frontmatter, progressive disclosure, packaging, and testing.
hookify	/hookify	Captures project guardrails (warn/block rules) and writes them as real Biorouter hooks in <project>/.biorouter/hooks.yaml — a block rule stops the action, a warn rule surfaces a notice. Requires hooks.allow_project_hooks: true.
code-modernization	/code-modernization	Phased legacy-modernization workflow: preflight → assess → map → extract rules → brief → transform → harden → status.
playground	/playground	Generates self-contained single-file interactive HTML playgrounds for configuring something visually and copying out a prompt.
frontend-design	Auto (frontend work)	Anthropic's anti-slop frontend guidance; complements the taste-skill.
autoskill-workflow-discovery	/autoskill-workflow-discovery	Detects repeated workflows and drafts native BioRouter skills, hooks, checklist updates, or extension plans.

---

BioSkills — bioinformatics bundle

bio-skills/ is a large, context-triggered bundle covering the working biologist's toolkit: sequence I/O, alignment, variant calling, single-cell, ATAC/ChIP-seq, proteomics, metabolomics, pathway analysis, machine learning, workflow management, and more — 466 individual skills across 63 categories in this snapshot. These load automatically when BioRouter sees a relevant file or task (a .vcf, a scanpy import, a Seurat object). See bio-skills/INDEX.md for the full catalog.

Integrated from GPTomics/bioSkills (MIT-origin upstream) and distributed here as part of the Apache-2.0 BioRouter skills package with provenance preserved in the attribution table. BioRouter's house style prefers uv over pip; skills with pip install lines carry a one-line uv tip banner while leaving the original commands intact for reproducibility.

scientific-research is designed to delegate to this bundle — and to the SPOKE, CDW, OMOP, and MedCP extensions — for any biomedical data access, so the orchestrator never has to recall facts from memory.

---

Repository layout

biorouter-skills/
├── LICENSE                        ← Apache-2.0
├── NOTICE                         ← upstream attribution notes
├── auto-research/SKILL.md
├── anti-ai-writing/SKILL.md
├── taste-skill/SKILL.md
├── develop-biorouter-extension/SKILL.md
├── ggplot-visualization/SKILL.md
├── python-scripting/SKILL.md
├── r-scripting/SKILL.md
├── ucsf-hpc/SKILL.md
├── ralph/
│   ├── SKILL.md                   ← the skill itself
│   ├── agent-instructions.md      ← prompt for the autonomous loop
│   └── prd.json.example
├── superpowers/                   ← 13 sub-skill collection
│   ├── brainstorming/
│   ├── test-driven-development/
│   └── …
└── bio-skills/                    ← bioinformatics bundle (63 categories)
    ├── INDEX.md
    ├── single-cell/
    ├── variant-calling/
    └── …

Every skill folder contains a SKILL.md with YAML frontmatter (name, description, optional user-invocable). Anything else in the folder is supporting material the skill can reference.

---

Contributing a new skill

1. Fork this repo.

2. Add a new top-level folder named after your skill (kebab-case).

3. Drop a SKILL.md into it with frontmatter:

   ---
   name: my-skill
   description: One sentence — what it does and when to use it. This is the line the model reads to decide whether to load the skill.
   license: Apache-2.0
   user-invocable: true   # optional
   ---
   
   # My Skill
   
   The skill body. Be concrete, give examples, and keep it focused on one job.

4. Open a pull request. A maintainer will review and cut a release with a matching .zip so BAAM picks it up automatically.

A good description: is the single most important line — it is what the model sees when deciding whether to invoke the skill, so be specific about when to use it, not just what it does.

---

Sources & attribution

This repository is licensed under the Apache License 2.0. Every distributed SKILL.md declares license: Apache-2.0 so BioRouter and BAAM can surface the license consistently. Several skills are adapted from permissively licensed open-source projects; we preserve those upstream credits and original license families in the table below. Lab-original skills are written by the Baranzini Lab.

Skill	Source	License	What changed for BioRouter
bio-skills	GPTomics/bioSkills, plus BioRouter-original Apache-2.0 additions inspired by Scientific Agent Skills	Distributed as Apache-2.0 in BioRouter; GPTomics origin was MIT	Dropped the bio- name prefix, set user-invocable: false, and added a one-line uv tip banner to skills that shipped pip install commands (originals left intact). 466 skills / 63 categories in this snapshot, including new PyDESeq2, compound hygiene, DeepChem, protein language model, DiffDock, Nextflow/nf-core, causal ML, SHAP, Python visualization, and literature-discovery skills.
taste-skill	Leonxlnx/taste-skill	Distributed as Apache-2.0 in BioRouter; upstream origin was MIT	Wired to BioRouter's auto-trigger on frontend work; content kept. Attribution header retained in the SKILL.md.
superpowers	obra/superpowers by Jesse Vincent	Distributed as Apache-2.0 in BioRouter; upstream origin was MIT	Folded the 13 engineering-discipline skills in; dropped the upstream using-superpowers dispatcher and its SessionStart hook (BioRouter discovers skills natively). verification-before-completion and finishing-a-development-branch gained an optional Stop-hook recipe to hard-enforce "tests pass before finishing."
ralph	snarktank/ralph by Ryan Carson, after Geoffrey Huntley's Ralph pattern	Distributed as Apache-2.0 in BioRouter; upstream origin was MIT	Reworked the PRD → prd.json → one-story-per-iteration flow around BioRouter's agent loop and /ralph invocation.
code-review, code-simplifier, commit-commands, skill-creator, code-modernization, playground, frontend-design	anthropics/claude-plugins-official	Apache-2.0	Each upstream plugin's slash-commands and subagents folded into one SKILL.md (these plugins ship no hooks). Shell routed through the Developer extension; lab-specific defaults adjusted. Each SKILL.md keeps a source header.
hookify	anthropics/claude-plugins-official → hookify	Apache-2.0	Rewritten to generate real, enforced BioRouter hooks (<project>/.biorouter/hooks.yaml) — the upstream's enforced-hook behavior, expressed in BioRouter's native hook schema.
scientific-research	Lab-original	Apache-2.0	— Orchestrates bio-skills and the SPOKE / CDW / OMOP / MedCP extensions.
empirical-research-router, causal-identification-gates, replication-package-audit, claim-evidence-integrity, citation-temporal-integrity, empirical-benchmark-harness, scientific-artifacts, open-science-review	Lab-original synthesis inspired by k-dense-ai/scientific-agent-skills and brycewang-stanford/Auto-Empirical-Research-Skills	Apache-2.0	Rewritten as native BioRouter skills with stage gates, citation/claim audits, and reproducibility checks. No upstream text copied.
scientific-data-engineering, statistical-modeling-bayes, scientific-machine-learning, geospatial-science, neurophysiology-analysis, quantum-scientific-computing, materials-informatics, biomedical-imaging-pathology, citation-library-management, laboratory-automation-planning, econometrics-toolkit, systematic-review-prisma	Lab-original capability-cluster synthesis inspired by k-dense-ai/scientific-agent-skills and brycewang-stanford/Auto-Empirical-Research-Skills	Apache-2.0	Rewritten as native BioRouter routing and quality-gate skills for remaining tool/library families. No upstream text copied.
simulation-optimization, physical-science-computing, metabolic-modeling-cobrapy, drug-discovery-benchmarks, clinical-decision-support-review, scientific-visual-communication, grant-funding-strategy, regulatory-quality-systems, gpu-compute-optimization, research-evaluation-venues, qualitative-thematic-analysis, financial-accounting-econometrics	Lab-original capability-cluster synthesis inspired by k-dense-ai/scientific-agent-skills and brycewang-stanford/Auto-Empirical-Research-Skills	Apache-2.0	Rewritten as native BioRouter routing and verification skills for simulation, physical science, regulatory, visual, grant, and empirical discipline clusters. No upstream text copied.
stata-research-workbench, survey-weights-complex-samples, psychometrics-sem-measurement, social-science-text-as-data, econ-finance-data-sources, empirical-paper-submission-rr, structural-economics-theory, causal-heterogeneity-policy-learning	Lab-original synthesis inspired by brycewang-stanford/Auto-Empirical-Research-Skills	Apache-2.0	Rewritten as native BioRouter empirical/social-science workbench skills. No upstream text copied.
autoskill-workflow-discovery, arbor-optimization, hugging-science-resource-discovery, molecular-dynamics, general-time-series-forecasting, bids-neuroimaging, clinical-ai-modeling, computational-fluid-dynamics	Lab-original capability-cluster synthesis inspired by k-dense-ai/scientific-agent-skills	Apache-2.0	Rewritten as native BioRouter skills for workflow discovery, evaluator optimization, scientific resource discovery, molecular dynamics, forecasting, BIDS neuroimaging, clinical AI modeling, and computational fluid dynamics. No upstream text copied.
anti-ai-writing	Lab-original	Apache-2.0	— Prose checklist for stripping AI tells.
ucsf-hpc	Lab-original	Apache-2.0	— UCSF CHPC cluster: SSH, SLURM templates, modules.
python-scripting, r-scripting, ggplot-visualization	Lab-original	Apache-2.0	— House style for Python, R, and ggplot2; auto-triggered by file context.
develop-biorouter-extension	Lab-original	Apache-2.0	— Guide for the .brxt extension format.

Maintained by the Baranzini Lab at UCSF.