skills

Use whenever building a UI table, list view, or dashboard panel showing records with amounts, dates, or statuses (invoices, tickets, orders, tasks, leads, log entries, etc.) — in React/HTML/artifacts, TanStack Table column defs, or when writing/reviewing component code for one. Consult before rendering a "flat" table (raw columns, no aggregation, no visual encoding), even if the user just says "make a table" or "show this data" — a bare table is rarely the best answer once rows carry a magnitude, a due date, or a status. Also use when asked to "make this table nicer", "more like a dashboard", or to add KPIs/summary above a table.

Use this skill whenever the user wants to bootstrap, scaffold, or initialize a new full-stack app with a Vite + React + TanStack + shadcn/ui frontend and a FastAPI + Postgres backend. Triggers on requests like "create a new app", "set up a project", "scaffold a full-stack app", "init a new project", or anything involving starting a fresh React/FastAPI application from scratch.

Discover repos across the whole Azure DevOps org — not just the one you're sitting in — and figure out where a piece of functionality actually lives before assuming it doesn't exist locally. Generates ALL_REPOS.md, a map of every project/repo in the org, each repo's description, and whether it's already cloned locally (and where). Use this whenever the user asks about "other repos", "which repo has X", cross-repo/cross-service work, wants to find or clone a repo they don't have locally yet, or when you're stuck looking for code that isn't in the current repo and it might live elsewhere in the org. Trigger on things like "is there already a repo for this", "check across our other repos", "what repos do we have", "where's the shared X library", or "I don't think this exists yet, can you check devops".

Conventions for versioning a PostgreSQL schema as plain `.sql` files inside a `database/` folder in the repo, instead of an ORM migration framework. Use when creating/organising a `database/` folder, adding a new table/view/function/type, or deciding where a migration script goes. Triggers on things like "add a new table to database/", "where do table definitions live", "database folder structure", "schema layer", "migration script", ".prod file", ".init.sql".

PostgreSQL coding standards for Python projects using psycopg (no ORM). Use this skill whenever the user is writing or reviewing code that queries PostgreSQL with psycopg, creates SQL files, maps query results to Python objects, or asks how to structure database access code. Triggers on things like "add a database query", "write a SQL query", "create a repository", "fetch from postgres", "insert into the database", "update the DB", "parameterized query", "pydantic model from DB row", "sql file", "dynamic SQL", or any mention of psycopg, psycopg2, asyncpg, or SQLAlchemy where the user seems open to a different approach. Also use when the user asks about connection pooling, transactions, cursor factories, or how to organise database code.

Set up and work with a local PostgreSQL test database in Docker for integration/e2e tests. Use this skill whenever the user wants to add, configure, troubleshoot, or run queries against a local test Postgres instance — even if they don't say "postgres" explicitly. Triggers on things like "set up a test database", "add postgres fixtures", "init test DB", "local postgres for tests", "add a column", "create a table", "change the schema", "run a SQL query on my test DB", "reset the database", "database is out of sync", "test data", or anything involving the local test Postgres database. Also use this skill when the user asks how to inspect the test DB, seed rows, or apply a migration to the local environment.

PostgreSQL-based async task queue for Python using Procrastinate. Use this skill whenever the user is working with Procrastinate — setting it up, defining tasks, deferring jobs, running workers, configuring retries, periodic tasks, or deploying to Azure (Web App or Container App). Also triggers on: "background jobs", "task queue postgres", "deferred task", "async worker", "procrastinate worker", "procrastinate app", "run worker in azure", "start script worker", "background task fastapi", "background task django procrastinate".

Guide for working with DuckDB — CLI usage, SQL execution, reading files (CSV, Parquet, JSON, Excel), extensions (httpfs, spatial, json, excel, postgres, etc.), and the Python duckdb package. Use this skill whenever the user is working with DuckDB, whether via CLI, SQL scripts, or Python. Trigger on: duckdb queries, reading local/remote files with duckdb, installing duckdb extensions, duckdb in Python/pandas, exporting data, attaching or querying PostgreSQL from DuckDB, or any question about DuckDB behavior or syntax.

Correct patterns for channel-based asyncio concurrency in Python using the aiochannel library. Use this skill whenever the user is working with aiochannel, writing producer/consumer pipelines in asyncio, or asking how to pass data between async tasks without a shared queue. Trigger on any mention of aiochannel, Channel[T], async producer/consumer, pipeline stages in asyncio, or "how do I pass items between coroutines". Also trigger when the user is about to write a producer that feeds a consumer task — even if they don't name aiochannel explicitly.

Browser automation and visual verification with Playwright in Python. Use this whenever the user wants to drive a browser from Python — taking screenshots of a running dev server or deployed page, scraping rendered content, automating clicks/forms, mocking network requests, or writing pytest browser tests. Also use this when the user wants Claude to "see" what a UI looks like (the bundled screenshot helper drops a PNG in $TMPDIR which Claude can read directly), when they mention pytest-playwright, when they say things like "verify my dashboard renders" / "screenshot localhost" / "test the signup flow in a browser", or any time a task needs a real rendered page rather than raw HTML. Async API is the default; sync and pytest patterns are documented inline.

Set up code formatting and pre-commit hooks using prek (fast Rust-based alternative to pre-commit) with prek.toml config. Use whenever the user wants to configure prek, add formatters, set up pre-commit hooks, or enforce code style. Triggers on "set up prek", "add formatting", "configure formatters", "pre-commit setup", "add ruff", "set up prettier", "configure sqlfmt". Works on new and existing projects. Always use when init-app-stack has just run.

Autonomous iterative optimization loop for Python, SQL, ML, and Spark/Databricks. Define a measurable goal; the skill runs code changes, measures, and keeps/reverts until you stop it or the budget runs out. Use for: Python runtime (cProfile, hyperfine), SQL latency, pytest pass rate/coverage, ty/pyright errors, ML metrics (loss/accuracy/F1), memory, PySpark/Spark Connect. Trigger on: "optimize this", "autoopt", "autoresearch", "keep trying until it's faster", "hill-climb this", "run experiments", "iterate autonomously", "optimize SQL performance", "tune this model", "optimize this Spark job", "tune my PySpark pipeline", "improve test coverage automatically". DO NOT USE FOR: one-shot fixes, code review without a metric, tasks with no measurable outcome.

SQL and data warehouse coding guidelines for the BME data platform. Use whenever writing, reviewing, or refactoring SQL — including SparkSQL, T-SQL, and Databricks SQL. Apply when naming tables, columns, CTEs, views, functions, or procedures; building fact or dimension tables; working with business keys, surrogate keys, or metadata columns; or setting up the semantic layer. Also apply for Spark SQL: QUALIFY, GROUP BY ALL, collations, NULL-safe patterns, conditional aggregation, timestamp intervals. Trigger on: "write sql", "review sql", "fact table", "dimension table", "surrogate key", "business key", "gold layer", "spark sql", "databricks sql", "t-sql", "naming convention", "sql style", "semantic layer", "combi field", "qualify", "group by all", "deduplication", "conditional aggregation".

Dimensional data modeling guide for the Fabricks platform — covers the full pipeline from staging through raw, transf, core, and semantic layers; fact and dimension table design; key strategy; SCD types; bridge tables; degenerate dimensions; and config. Use whenever designing or reviewing fact/dimension tables, deciding where logic belongs in the pipeline, choosing SCD1/SCD2, constructing business/surrogate keys, modeling many-to-many relationships, or understanding layer architecture. Trigger on: "design fact table", "design dimension", "SCD type", "slowly changing dimension", "surrogate key", "business key", "bridge table", "layer architecture", "transf vs core", "udf_identity", "udf_key", "__valid_from", "__is_current", "new dim table", "new fact table", "data model", "bronze", "silver", "gold", "semantic layer", "staging".

Databricks SQL query optimizer and error fixer: analyzes a slow or broken SQL query, rewrites it for speed using SQL-level optimizations only, validates byte-for-byte result equivalence, and benchmarks both versions with statistical significance testing. Also diagnoses and fixes SQL errors (syntax errors, cast failures, column resolution, type mismatches) using --goals fix. Trigger on: "/databricks-sql-autotuner", "optimize this SQL", "make this query faster", "tune my Databricks query", "benchmark SQL on Databricks", "speed up this spark SQL", "why is my query slow on Databricks", "fix my SQL query", "SQL syntax error", "casting error in SQL", "AnalysisException", "ParseException", or whenever a user pastes a SQL query and mentions errors, slowness, or failures.

Python code optimizer and error fixer: analyzes a Python file or function, rewrites it for speed and quality one change at a time, validates with pytest, benchmarks with timeit, checks with ruff/ty, and measures complexity reduction. Also diagnoses broken Python code (SyntaxError, ImportError, TypeError, AttributeError, runtime crashes) using --goals fix. Trigger on: "/python-autotuner", "optimize this Python", "make this function faster", "benchmark my Python function", "reduce complexity", "ruff keeps failing", "my Python function is too slow", "simplify this Python code", "fix this Python error", "SyntaxError in Python", "my Python code is crashing", "TypeError in Python", or whenever a user shares Python code and mentions errors, crashes, performance, quality, slowness, or complexity.

Use this skill whenever working with FastAPI — building APIs, adding routes, structuring projects, streaming responses, handling authentication, database access, or running the server. Triggers on: "fastapi", "build an API", "add an endpoint", "stream response", "SSE", "server-sent events", "async endpoint", "dependency injection", "pydantic model", "granian", or any backend API work in Python. Always use this skill when the user is writing or modifying a FastAPI application, even if they don't say "FastAPI" explicitly.

Azure Entra ID SSO for FastAPI using cookie-based sessions (MSAL, /login → /callback → session). Trigger on: Azure AD login, Entra SSO, protect routes, RBAC, DEV_MODE bypass, download links with auth. Covers App Registration, AuthMiddleware, role checks via Postgres, and cookie auth enabling native <a download> links.

Apply and enforce Python-specific coding standards. Use alongside coding-guidelines for any Python file — covers typing, Pyright, dataclasses, enums, abstract base classes, mutable defaults, and module-level state. Trigger on any Python work session, PR review, or when the user asks to "follow Python guidelines", "check Python style", or "enforce Python standards".

Apply and enforce TypeScript-specific coding standards. Use alongside coding-guidelines for any TypeScript file — covers typing, no-any policy, interfaces, abstract classes, async typing, discriminated unions, and pre-commit checks. Trigger on any TypeScript work session, PR review, code generation task, or refactoring. Also trigger when writing TypeScript even if the user doesn't explicitly ask for guidelines — if the user asks you to write a function, add a feature, fix a bug, or review code in a .ts or .tsx file, apply these standards automatically without being asked.

KendoReact component library patterns and best practices for React developers. Trigger whenever the user works with KendoReact components, imports from @progress/kendo-react-*, asks how to set up Grid, Form, DateInputs, Charts, or any KendoReact component, asks about theming or licensing, or asks to "use Kendo", "add a Kendo grid", "set up KendoReact", or similar.

Guide for running PySpark code locally against a remote Databricks cluster via Spark Connect (databricks-connect). Use this skill when asked how to set up Databricks Connect, connect to a cluster from a local machine, run notebooks or scripts locally with Spark semantics, understand the get_spark_session() pattern, configure .env for Spark Connect, manage the local-vs-cluster execution model, or use the autoresearch loop with Databricks. Trigger on: "connect to Databricks locally", "databricks-connect setup", "spark connect", "get_spark_session", "run notebook locally against cluster", "local Spark", "databricks connect config", "DATABRICKS_CLUSTER_ID", "DATABRICKS_PROFILE env".

Comprehensive best practices for TanStack libraries in React applications — covering TanStack Query (React Query) data fetching, caching, and mutations; TanStack Router type-safe routing and search params; TanStack Query + Router integration patterns; TypeScript usage; and common pitfalls to avoid. Use this skill whenever someone is: building data fetching logic or server state management with TanStack Query / React Query; setting up or refactoring routing with TanStack Router; integrating Query with Router in loaders; asking about staleTime, cache invalidation, query key factories, optimistic updates, search params, suspense patterns, SSR hydration, or prefetching. Also trigger when the user writes useQuery, useMutation, useNavigate, createFileRoute, createRootRouteWithContext, queryOptions, or ensureQueryData.

Use this skill whenever the user is working with Kendo UI for Vue — including the Data Grid, DropDownList, ComboBox, AutoComplete, or any @progress/kendo-vue-* package. Covers server-side paging/filtering/sorting, inline CRUD editing, row selection, custom cell templates, master-detail rows, grouping, Excel export, and virtualization for large datasets. Trigger on any mention of kendo grid, telerik vue, @progress/kendo-vue-grid, or when the user is building a sortable/filterable/pageable data table in Vue. Also trigger when a user asks about adding editing, export, or selection to an existing Kendo grid.

Build Python web UIs with NiceGUI — covering layout, widgets, data binding, routing, and the AgGrid data grid. Use this skill whenever the user mentions NiceGUI, wants to build a Python web app, dashboard, or internal tool, asks about ui.aggrid, ui.page, or any NiceGUI component. Triggers on: "nicegui", "ui.aggrid", "ui.page", "NiceGUI app", "python web ui", "python dashboard", "aggrid python", "build a web interface in python".

Use this skill whenever the user is working with Kendo UI for Angular — including the Data Grid, TreeList, TreeView, DropDownList, AutoComplete, or any @progress/kendo-angular-* package. Covers server-side paging/filtering/sorting, inline CRUD editing, row selection, custom cell templates, master-detail rows, grouping, Excel export, and virtualization for large datasets. Trigger on any mention of kendo grid, kendo treelist, kendo treeview, telerik angular, @progress/kendo-angular-grid, or when the user is building a sortable/filterable/pageable data table in Angular. Also trigger when a user asks about adding editing, export, tree navigation, or dropdown filtering to an existing Kendo Angular component.

Master skill for the bmsuisse platform — routes to the relevant skills based on sub-command. Always enables caveman communication style. Use /bms for core guidelines (SQL + Python + TypeScript + glossary), /bms sql for SQL + Databricks optimization, /bms python for Python + autotuner, /bms data for dimensional modeling + Fabricks. Trigger on: "/bms", "/bms sql", "/bms python", "/bms data", "bmsuisse mode", "activate bms".

Databricks CLI operations: auth, profiles, data exploration, bundles, and notebook execution. Use this skill for ANY Databricks task — running or debugging notebooks on the cluster, executing SQL queries, exploring tables, deploying jobs, or fixing notebook errors. Trigger whenever the user mentions "run the notebook", "execute on Databricks", "run on the cluster", "fix the notebook error", "query the table", or anything involving the Databricks workspace.

Remove AI slop from code and pull requests. Use this skill whenever the user wants to clean up AI-generated code, review a diff for unnecessary noise, audit a file or PR for style inconsistencies, over-abstraction, defensive checks that don't belong, redundant comments, or anything that "doesn't feel like it belongs here". Also use it proactively after any AI-assisted coding session to apply the Research → Plan → Execute → Review → Revise loop and ensure the output is clean, reviewable, and consistent with the existing codebase. Trigger on: "remove slop", "clean up AI code", "deslop", "review diff", "does this look AI-written", "polish this PR", "make this feel more human", "clean up before commit".

Detect and surgically fix AI writing patterns in English, German, French, or Italian text. Use this skill whenever the user asks to 'remove AI writing', 'fix AI tells', 'make this sound less like ChatGPT/Claude', 'avoid AI patterns', 'make this more human', 'de-slop my text', 'nettoyer le texte IA', 'KI-Schreibstil entfernen', 'rimuovere la scrittura IA', or pastes text that sounds stiff, puffed-up, or formulaic. Also trigger proactively after writing or editing long prose sections — even if the user doesn't explicitly ask for AI pattern removal. The skill applies targeted word-level and sentence-level fixes without rewriting content or changing meaning.

Comprehensive C-AL performance analyzer for Classic Microsoft Dynamics NAV (Navision). Targets Classic NAV anti-patterns: N+1 queries, LOCKTABLE overuse, FlowField overuse, key/index misalignment, heavy transactions, and page trigger abuse. Zero-dependency Python — no packages required. Use when the user asks about analyzing codeunits, tables, or pages, finding performance bottlenecks, locking/blocking issues, or optimizing Classic NAV / C-AL code. Trigger: "/codeunit-analyzer", "analyze codeunit", "scan bottlenecks", "performance issues".

Use this skill whenever company-specific jargon, acronyms, or domain terminology is needed to answer correctly.

Analyzes all SQL files in the Fabricks.Runtime repository, builds a dependency DAG, runs performance heuristics, and produces a Markdown report with ranked findings and improvement suggestions. Optionally fetches row counts and EXPLAIN COST plans from Databricks. Pass --fix to automatically create a branch and apply SQL rewrites.

Use this skill whenever the user wants to verify, revise, or improve a scientific essay, academic paper, or any written document. It checks citation consistency, verifies bibliography entries via CrossRef and Google Scholar/Semantic Scholar, and detects AI-generated writing patterns (slop). Trigger this skill whenever the user mentions 'check citations', 'verify references', 'scientific revision', 'revise essay', 'clean up writing', 'check for AI slop', 'improve writing quality', 'check google scholar', or 'review my paper', especially in academic or professional writing contexts.

Universal SQL performance optimization assistant for comprehensive query tuning, indexing strategies, and database performance analysis across all SQL databases (MySQL, PostgreSQL, SQL Server, Oracle, Databricks). SQL style: sqlfmt lowercase, CTEs over subqueries, QUALIFY over row_number subqueries, GROUP BY ALL, meaningful aliases, alias-prefixed columns, no select *.