Add CLAUDE.md with development workflow and quality guidelines

.claude-plugin/marketplace.json

@@ -6,7 +6,7 @@
   },
   "metadata": {
     "description": "Claude scientific skills from K-Dense Inc",
-    "version": "2.20.0"
+    "version": "2.24.0"
   },
   "plugins": [
     {
@@ -25,6 +25,7 @@
         "./scientific-skills/cellxgene-census",
         "./scientific-skills/cirq",
         "./scientific-skills/cobrapy",
+        "./scientific-skills/copilot-skills-integration",
         "./scientific-skills/dask",
         "./scientific-skills/datacommons-client",
         "./scientific-skills/datamol",
@@ -42,6 +43,7 @@
         "./scientific-skills/gtars",
         "./scientific-skills/histolab",
         "./scientific-skills/imaging-data-commons",
+        "./scientific-skills/infographics",
         "./scientific-skills/hypogenic",
         "./scientific-skills/lamindb",
         "./scientific-skills/markitdown",
@@ -53,6 +55,7 @@
         "./scientific-skills/molfeat",
         "./scientific-skills/neurokit2",
         "./scientific-skills/networkx",
+        "./scientific-skills/neuropixels-analysis",
         "./scientific-skills/paper-2-web",
         "./scientific-skills/pathml",
         "./scientific-skills/pennylane",
@@ -157,7 +160,9 @@
         "./scientific-skills/offer-k-dense-web",
         "./scientific-skills/protocolsio-integration",
         "./scientific-skills/get-available-resources",
-        "./scientific-skills/iso-13485-certification"
+        "./scientific-skills/iso-13485-certification",
+        "./scientific-skills/browser-automation",
+        "./scientific-skills/multi-agent-research"
       ]
     }
   ]

CLAUDE.md

@@ -0,0 +1,216 @@
+# CLAUDE.md — Claude Scientific Skills
+
+## Project Overview
+
+Open-source collection of 142 scientific skills for AI agents (Cursor, Claude Code, Codex). Each skill is a self-contained folder under `scientific-skills/` with a `SKILL.md` file following the open Agent Skills standard. Maintained by K-Dense Inc.
+
+## Development Workflow
+
+### 1. Adding or Updating a Skill
+
+Every skill lives in `scientific-skills/<skill-name>/` and MUST contain a `SKILL.md` with YAML frontmatter:
+
+```yaml
+---
+name: skill-name
+description: One-line description. Be specific about when to use vs alternatives.
+license: MIT
+metadata:
+    skill-author: K-Dense Inc.
+---
+```
+
+**Rules:**
+- Keep `description` actionable — explain WHEN to use this skill and how it differs from similar ones
+- Include `references/` directory for API docs, guides
+- Include `scripts/` directory for example code when applicable
+- Never add dependencies that require compilation without documenting prerequisites
+
+### 2. Registering Skills
+
+After adding a skill, register it in `.claude-plugin/marketplace.json` under `plugins[0].skills`:
+```
+"./scientific-skills/<skill-name>"
+```
+
+Then bump the version in `marketplace.json` → `metadata.version`. This triggers the release workflow.
+
+### 3. Version Bumping & Releases
+
+- Version is in `.claude-plugin/marketplace.json` → `metadata.version` (semver)
+- Bumping the version on `main` triggers `.github/workflows/release.yml`
+- The workflow creates a GitHub release with auto-generated changelog
+- **Always bump version when adding/updating skills**
+
+### 4. Commit Style
+
+Follow existing patterns:
+```
+Add <skill-name> skill for <one-line purpose>
+Update <skill-name> skill to v<X.Y.Z>
+Bump version number to <X.Y.Z> in marketplace.json
+Enhance <skill-name> documentation
+```
+
+Do NOT use conventional commits (no `feat:`, `fix:` prefixes). Keep it simple and descriptive.
+
+### 5. Before Creating a PR
+
+- Verify the new SKILL.md has valid YAML frontmatter (name, description, license, metadata)
+- Verify the skill is registered in `marketplace.json`
+- Verify no broken links in documentation
+- Keep PRs focused: one skill per PR when possible
+
+## Architecture
+
+```
+.
+├── .claude-plugin/marketplace.json   # Skill registry + version (triggers releases)
+├── .github/workflows/release.yml     # Auto-release on version bump
+├── scientific-skills/                 # 142 skills, each with SKILL.md
+│   ├── <skill-name>/
+│   │   ├── SKILL.md                  # Required: frontmatter + documentation
+│   │   ├── references/               # Optional: API docs, guides
+│   │   └── scripts/                  # Optional: example code
+├── docs/                             # Project documentation
+├── README.md                         # Main documentation
+└── LICENSE.md                        # MIT
+```
+
+## Common Mistakes — Do NOT Repeat These
+
+- **Do not add skills without registering them in marketplace.json** — they won't be discoverable
+- **Do not forget to bump the version** — no release will be created
+- **Do not use `interface` in TypeScript** — use `type` instead (if any TS code is written)
+- **Do not create overly long SKILL.md files** — keep them focused and scannable. Prefer code examples over long explanations
+- **Do not duplicate skills** — check existing skills first. If rdkit already covers molecular fingerprints, don't create a separate fingerprint skill
+- **Do not add Python package skills without specifying the install command** — always include `uv pip install <package>` or equivalent
+
+## Skill Categories Quick Reference
+
+| Category | Example Skills | Count |
+|----------|---------------|-------|
+| Python Packages | rdkit, scanpy, biopython, pytorch-lightning | 55+ |
+| Scientific Databases | pubmed, chembl, uniprot, clinvar, cosmic | 28+ |
+| Integrations | benchling, dnanexus, latchbio, omero | 15+ |
+| Analysis & Communication | literature-review, scientific-writing, visualization | 30+ |
+| Research & Clinical | hypothesis-generation, clinical-decision-support | 10+ |
+
+## Key Files
+
+- **Version/registry**: `.claude-plugin/marketplace.json`
+- **Release automation**: `.github/workflows/release.yml`
+- **All skills**: `scientific-skills/*/SKILL.md`
+
+## Python Environment
+
+- Require Python 3.9+, recommend 3.12+
+- Use `uv` for package management (not pip directly)
+- Install command: `uv pip install <package>`
+- Virtual environments: `uv venv && source .venv/bin/activate`
+
+## Planning & Implementation Strategy
+
+Use plan mode for any task that touches multiple files or requires architectural decisions. The pattern:
+
+1. **Plan first** — Start with `/plan` or ask Claude to plan before coding. A good plan should identify all files to modify, the order of operations, and edge cases.
+2. **Validate the plan** — Before implementing, check: does the skill already exist? Is there overlap with existing skills? Will `marketplace.json` need updating?
+3. **Implement in order** — SKILL.md first, then `references/` and `scripts/`, then register in `marketplace.json`, then bump version.
+4. **If implementation drifts from plan** — Stop and re-plan instead of patching. Return to plan mode immediately.
+
+## Quality Checklist for SKILL.md Files
+
+Good SKILL.md files in this repo follow a consistent pattern. Use this as a template:
+
+```markdown
+---
+name: <tool-name>
+description: <What it does>. <When to use it vs alternatives>. <Key differentiator>.
+license: <SPDX identifier>
+metadata:
+    skill-author: K-Dense Inc.
+---
+
+# <Tool Name>: <Subtitle>
+
+## Overview
+One paragraph: what it is, when to apply this skill.
+
+## When to Use This Skill
+Bullet list of specific scenarios.
+
+## Quick Start
+Minimal working code example (under 20 lines).
+
+## Core Capabilities
+Organized by use case, each with code examples.
+
+## Common Patterns
+Real-world patterns users will actually need.
+
+## Troubleshooting
+Known issues and their solutions.
+```
+
+**Size guidelines:**
+- Target: 200-500 lines (sweet spot for usefulness without bloat)
+- Under 150 lines: probably too thin — add more code examples
+- Over 800 lines: too long — split into sections, move reference material to `references/`
+- Skills like `latex-posters` (1600 lines) are the anti-pattern — be more concise
+
+## Skill Description Writing Guide
+
+The `description` field in frontmatter is the most important line. It determines when Claude auto-invokes the skill. Write it as:
+
+```
+<What it does>. <When to use — specific trigger>. <How it differs from similar skills>.
+```
+
+**Good examples from this repo:**
+- rdkit: "Cheminformatics toolkit for fine-grained molecular control. SMILES/SDF parsing, descriptors... For standard workflows with simpler interface, use datamol."
+- scanpy: "Standard single-cell RNA-seq analysis pipeline... For deep learning models use scvi-tools; for data format questions use anndata."
+
+**Bad pattern:** Generic descriptions like "A tool for data analysis" — never say what makes it unique.
+
+## Cross-Skill References
+
+Many skills in this repo overlap. When writing a new skill, always reference alternatives:
+- **rdkit** ↔ **datamol** (datamol wraps rdkit with simpler API)
+- **scanpy** ↔ **scvi-tools** ↔ **anndata** (different layers of single-cell analysis)
+- **biopython** ↔ **gget** ↔ **bioservices** (different levels of database access)
+- **pubmed** ↔ **biorxiv** ↔ **openalex** (different literature databases)
+- **chembl** ↔ **pubchem** ↔ **drugbank** (different chemical databases)
+
+Add a "When NOT to use this skill" or "See also" section when there's overlap.
+
+## MCP Servers Available in This Environment
+
+This project has access to the following MCP servers. Reference them in skills when relevant:
+
+| Server | Purpose |
+|--------|---------|
+| `biothings-mcp` | Gene/variant/drug annotation (MyGene, MyVariant, MyChem) |
+| `gget-mcp` | Quick gene/protein lookups via gget |
+| `synergy-age-mcp` | Synergistic drug combinations database |
+| `opengenes-mcp` | Aging/longevity gene database |
+| `pharmacology-mcp` | Drug pharmacology data |
+| `biocontext-kb` | Biological context knowledge base |
+
+## Parallel Workflow with Git Worktrees
+
+This project uses git worktrees extensively (see `~/.claude-worktrees/`). Each worktree can run an independent Claude session. Use this for:
+
+- Working on multiple skills in parallel (one skill per worktree)
+- Testing changes without affecting the main branch
+- Running independent Claude sessions that don't interfere
+
+Create a new worktree: `git worktree add ~/.claude-worktrees/claude-scientific-skills/<name> -b <branch-name>`
+
+## Lessons Learned — Update This Section After Corrections
+
+<!-- Add entries here after every correction using the format below -->
+<!-- Each entry captures a mistake so it's never repeated -->
+
+- **Path encoding**: Claude Code encodes CWD paths non-trivially for `~/.claude/projects/` directories. Don't assume simple `/` → `-` replacement. Hidden directories (starting with `.`) get the dot stripped, creating double-dashes. Use fuzzy matching on the last two path components instead.
+- **SKILL.md size**: Keep under 500 lines. Long skills (800+) cause context bloat when auto-invoked. Move API reference docs to `references/` subdirectory.
+- **License field**: Use SPDX identifiers (`MIT`, `BSD-3-Clause`, `Apache-2.0`), not `Unknown`. Check upstream package license before writing the skill.