pyramidheadshark
23 verified skills92 total stars
.claude/skills/data-validation
# Data Validation ## When to Load This Skill Load when working with: Pandera DataFrame schemas, Great Expectations suites, data quality checks, input validation for ML pipelines, data contracts between pipeline stages. ## Pandera — DataFrame Schema Validation Define schemas declaratively and validate at pipeline boundaries: ```python import pandera as pa from pandera.typing import DataFrame, Series class InputSchema(pa.DataFrameModel): user_id: Series[int] = pa.Field(ge=0, nullable=Fa
development4
.claude/skills/langgraph-patterns
# LangGraph Patterns ## When to Load This Skill Load when working with: LangGraph state machines, agent nodes, tool definitions, checkpointers, human-in-the-loop interrupts, multi-agent coordination. ## Current Version LangGraph `>=0.2.0` (langgraph-checkpoint for persistence). Always pin exact version in `pyproject.toml`. ## Core Concepts LangGraph models agent workflows as directed graphs: - **State**: typed dict passed between nodes (the single source of truth) - **Nodes**: Python async
tools4
.claude/skills/ml-data-handling
# ML Data Handling ## When to Load This Skill Load when working with: pickle, ONNX, Parquet, Feather, HDF5, large datasets, S3/Object Storage, DVC-like versioning, model artifacts, data pipelines. ## Core Principle Binary ML artifacts (weights, embeddings, datasets) are NEVER committed to Git. They live in object storage (Yandex Cloud Object Storage — S3-compatible) or are reproducible via pipeline. Paths and versions are tracked in code; actual data is not. ## Directory Convention ``` pro
development4
.claude/skills/nlp-slm-patterns
# NLP / SLM Patterns ## When to Load This Skill Load when working with: local language models, anonymization/PII detection, text classification, NER, Ollama, vLLM, Presidio, spaCy, Hugging Face Transformers, SLM inference pipelines. ## Philosophy Local models run locally during development, then deploy to YC GPU VM for production. The adapter interface is identical in both environments — only the endpoint URL changes via `.env`. Anonymization projects deal with real personal data. Default p
development4
.claude/skills/prompt-engineering
# Prompt Engineering ## When to Load This Skill Load when working with: system prompts, few-shot examples, chain-of-thought, prompt templates, `system_prompt`, `few_shot`, prompt evaluation, "промпт", CoT, output formatting, structured output. ## System Prompt Structure A production system prompt has four sections: ``` ## Role Who Claude is in this context. ## Context What the system is, what data Claude has access to, key constraints. ## Task What Claude must do. Be specific — avoid "hel
development4
.claude/skills/rag-vector-db
# RAG & Vector DB Patterns ## When to Load This Skill Load when working with: Qdrant, pgvector, embeddings, chunking, retrieval-augmented generation, semantic search, knowledge bases, document ingestion pipelines. ## Vector DB Choice | Option | When to Use | |---|---| | **Qdrant** | Default choice. Standalone service, excellent filtering, production-ready, Docker-friendly | | **pgvector** | Already have PostgreSQL, simple use case, don't want extra service | | **In-memory (numpy)** | Prototy
devops4
.claude/skills/windows-developer
# Windows Developer Guide ## When to Load Automatically loaded on Windows (`platform_trigger: "win32"`). Applies to: `.py`, `.ps1`, `.bat`, `.cmd` files and any Windows-specific workflow. ## Python on Windows ### Encoding (CRITICAL) Windows defaults to `cp1251` / `cp1252` for file I/O. Always specify UTF-8 explicitly: ```python with open("file.txt", "r", encoding="utf-8") as f: content = f.read() Path("file.txt").read_text(encoding="utf-8") Path("file.txt").write_text(content, encodin
development4
.claude/skills/critical-analysis
## Skill: Critical Analysis (v1.1) Multi-role critique system for ML experiments and architecture decisions. Based on: Solo Performance Prompting (NAACL 2024), CrewAI role taxonomy, Bermingham 13-agent DA-as-gate, De Bono Six Thinking Hats (Black Hat gating). --- ### When to Load (Auto-Default Behavior) **Before any of the following — run Quick Mode without waiting for user request:** - Architecture or design decision (new integration, refactor, layer change) - ML experiment launch (clusteri
development4
.claude/skills/database-migration-safety
# Skill: Database Migration Safety ## When to Load Auto-load when: working with Alembic, raw SQL migrations, schema changes, `migrations/` directory, or `*.sql` files. Triggers on `alembic`, `migration`, `upgrade`, `downgrade`, `schema`, `ALTER TABLE` (≥2 keywords). ## Core Rules Every migration must satisfy these requirements before `alembic upgrade head` runs: 1. **Reversible** — `downgrade()` must be implemented and tested. `pass` is not acceptable. 2. **Staged** — run against a staging/
testing4
.claude/skills/design-doc-creator
# Design Doc Creator ## When to Load This Skill Load when: starting a new project, creating or updating `design-doc.md`, translating client requirements into structured documentation, filling in the design document template. ## Purpose This skill guides the creation of a complete design document. It is always the first deliverable of any project — before environment setup, before tests, before code. ## Document Structure (10 Sections) The template lives at `templates/design-doc.md` in `cla
tools4
.claude/skills/fastapi-patterns
# FastAPI Patterns ## When to Load This Skill Load when working with: FastAPI routers, Pydantic models, dependency injection, middleware, ASGI lifecycle, HTTP endpoints, background tasks. ## Architectural Contract All FastAPI projects follow Hexagonal Architecture: ``` api/ → adapters IN (HTTP boundary) core/ → domain (pure Python, zero framework imports) services/ → application (orchestrates core + adapters) adapters/ → adapters OUT (DB, LLM, S3, external A
development4
.claude/skills/github-actions
# GitHub Actions Patterns ## When to Load This Skill Load when working with: `.github/workflows/*.yml`, CI pipelines, lint/test/build/deploy jobs, matrix strategies, GitHub secrets, environment protection rules. Keywords: `github actions`, `ci`, `workflow`, `lint job`, `test job`, `deploy`, `matrix`, `pipeline` ## Canonical Job Templates ### Lint (ruff + mypy) ```yaml lint: runs-on: ubuntu-latest steps: - uses: actions/checkout@v4 - uses: astral-sh/setup-uv@v4 with:
development4
.claude/skills/multimodal-router
# Multimodal Router ## When to Load This Skill Load when working with: PDF files, Word documents, Excel spreadsheets, images, audio, video files, or any document exceeding 400k tokens that cannot fit in Claude's standard context. ## Model - **Model**: `google/gemini-3-flash-preview` - **Provider**: OpenRouter API - **Context window**: 1M tokens - **Capabilities**: text, images, audio, video, PDF — all natively - **Thinking levels**: minimal / low / medium / high (configurable per task) Gemi
development4
.claude/skills/predictive-analytics
# Predictive Analytics ## When to Load This Skill Load when working with: scikit-learn pipelines, feature engineering, tabular ML, time series, model training/evaluation, MLflow experiment tracking, model registry, cross-validation. ## Project Structure for ML Projects ``` src/{project_name}/ ├── core/ │ └── domain.py ├── ml/ │ ├── __init__.py │ ├── features/ │ │ ├── __init__.py │ │ ├── builder.py # FeatureBuilder — assembles feature matrix │ │ └── transformers.py #
development4
.claude/skills/python-project-standards
# Python Project Standards ## When to Load This Skill Load when working with: `pyproject.toml`, Python files, pre-commit config, dependency management, type hints, linting configuration. ## Stack Versions (as of 2026-Q1) - Python: `>=3.11` (minimum), prefer `3.12` - uv: latest stable (replaces pip, venv, pip-tools — all in one) - Ruff: `>=0.9.0` - MyPy: `>=1.10.0` - pre-commit: `>=3.7.0` ## pyproject.toml Standard Config The canonical `pyproject.toml` for all projects. Copy and adjust `[pr
tools4
.claude/skills/skill-developer
# Skill Developer ## When to Load This Skill Load when: creating a new skill, modifying an existing skill, updating `skill-rules.json`, evaluating skill quality, or refactoring the skill library. ## Skill Anatomy Every skill follows this structure: ``` .claude/skills/{skill-name}/ ├── SKILL.md # main file — MUST be under 500 lines └── resources/ ├── topic-1.md # deep-dive subsections — under 500 lines each └── topic-2.md ``` `SKILL.md` is the entry point
development4
.claude/skills/supply-chain-auditor
# Skill: Supply Chain Auditor ## When to Load Auto-load when: adding dependencies, reviewing packages, updating versions, or discussing `requirements.txt`, `pyproject.toml`, `package.json`. Triggers on `dependency`, `install`, `package`, `CVE`, `audit`, `vulnerable` (≥2 keywords). ## Core Rules Every new dependency addition must pass this checklist before merging: 1. **Pinned** — exact version in production (`==1.2.3` for pip, `"1.2.3"` for npm, not `^` or `~`). 2. **Maintained** — last com
testing4
.claude/skills/test-first-patterns
# Test-First Patterns ## When to Load This Skill Load when writing tests, creating `.feature` files, setting up conftest, discussing test strategy, or reviewing coverage. ## Philosophy Tests are written BEFORE code. Always. No exceptions. The order is: Design Doc → BDD Scenarios → Unit Tests → Implementation. BDD scenarios come from the design document's use cases section — they are a direct translation of business requirements into executable specifications. This makes tests the living do
development4
.claude/skills/experiment-tracking
# Experiment Tracking ## When to Load This Skill Load when working with: MLflow experiments, run logging, model registry, artifact management, experiment comparison, cross-validation with tracking. ## Core Concepts | Concept | Purpose | |---------|---------| | **Run** | Single training execution — logs params, metrics, artifacts | | **Experiment** | Named collection of runs — logical grouping by model type or task | | **Model Registry** | Versioned model store — stages: None → Staging → Prod
data-ai4
.claude/skills/htmx-frontend
# HTMX Frontend ## When to Load This Skill Load when working with: Jinja2 templates, HTMX attributes, static files, server-side rendering, admin panels, dashboards integrated into FastAPI. ## Architecture Decision HTMX frontend lives inside the FastAPI application — no separate frontend service, no build step, no npm. This is intentional and correct for our use cases (internal tools, admin panels, dashboards with moderate load). For high-traffic public frontends, consider a separate service
tools4
.claude/skills/infra-yandex-cloud
# Infra: Yandex Cloud ## When to Load This Skill Load when working with: Terraform, Packer, Yandex Cloud resources, Docker deployment to VMs, Kubernetes prep, CI/CD deploy steps, VM configuration. ## Stack - **Packer** — builds VM images (pre-baked with Docker, dependencies) - **Terraform** — provisions YC infrastructure (VMs, networking, Object Storage) - **Docker Compose** — runs application on provisioned VM - **Helm** — Kubernetes-ready charts prepared from day one (even if K8s not yet u
development4
.claude/skills/claude-api-patterns
# Claude API Patterns ## When to Load This Skill Load when working with: Anthropic SDK, `anthropic` package, Claude API, tool use, streaming responses, message batches, `MessageCreate`, `@anthropic-ai/sdk`. ## SDK Setup ```python import anthropic client = anthropic.Anthropic(api_key=settings.ANTHROPIC_API_KEY) ``` Never hardcode the API key. Always use environment variables validated at startup. ## Basic Message ```python message = client.messages.create( model="claude-sonnet-4-6",
tools4
tests/fixtures/project-with-status/.claude/skills/design-doc-creator
# Design Doc Creator ## When to Load This Skill Load when: design documents, requirements, new project start. Short fixture skill for testing (optional/meta skill).
testing4