Cursor

Enforce SOLID design principles during code generation, refactors, and code review (SRP/OCP/LSP/ISP/DIP). Use when changing architecture, introducing new abstractions, reducing coupling, eliminating type-switches, or reviewing PRs for maintainability.

KeePassXC cursor.kdbx — natywny keyring per OS, keepass-db.path bez fallbacków, agent instaluje brakujące narzędzia.

On-demand over-engineering review/audit using the "laziness ladder" (DietrichGebert/ponytail) — the always-on rung is already in USER_RULES.txt STATIC PRINCIPLES; this skill is for deep diff/repo audits and debt tracking. Use when the user asks to review a diff or repo for over-engineering, unnecessary abstractions, unused dependencies, or dead code, or invokes /ponytail_review.

Keep the workspace tidy while the agent works: no orphan scratch docs/scripts, confine temp to .agent-scratch/ or tmp/agent-*, clean before declaring done. Use during implementation, multi-file edits, shell/write batches, or when marking work done — continuous hygiene, not post-hoc /cleanup.

Next.js App Router frontend stack workflow: Server Components-first, Prisma server-only (Node runtime), Tailwind + shadcn/ui, TanStack Query, React Hook Form + Zod, and conventions for work under frontend/. Use when implementing or refactoring UI, routes, data fetching, mutations, or forms in this stack.

The user's dedicated browser for agents — a real browser signed into their accounts, with live logins and a persistent profile. Use it for any task that touches a website or browser (open, read, act, fill, sign in, download, verify). The user installed it precisely so agents default here unprompted — over in-app browser tools, devtools/playwright automation, or headless fetching.

Build or modify plain HTML/CSS/JavaScript (no framework) using modular ES modules (ESM), SRP, minimal global state, and shippable runnable files with lightweight verification. Use for static pages, small UI widgets, vanilla JS refactors, and quick prototypes without React/Vue/Angular.

Enforce consistent Python typing (type hints + return types), concise Google-style docstrings, PEP 8/Black formatting, and unit tests when creating or editing Python code. Use when working on .py files, Python APIs, refactors, or bug fixes where maintainability and correctness matter.

Conduct deep, source-grounded research with freshness checks and citations, then synthesize an actionable answer. Use for time-sensitive, niche, or high-stakes topics; when the user asks to “zweryfikuj”, “zrób research”, or when uncertainty is high.

Use BrowserOS MCP to automate the user's visible Chromium browser (sessions, cookies, manual CAPTCHA/2FA). Use when anti-bot protection, user intervention, OAuth in existing session, or Klavis Strata app integrations are needed.

FastAPI + Pydantic v2 backend workflow: implement/modify API endpoints with thin handlers + service layer, add pytest tests, manage idempotent Alembic migrations, and implement streaming ETL/parsers with invalid-row CSV logging. Use this skill when working on Python backend APIs, schema changes, or ingestion/parsing pipelines.

Ground answers in the repository by reading relevant files, configs, and code paths. Use when the user asks to modify/debug code, configuration, CI/CD, or wants repo-specific reasoning.

Drive BrowserClaw (BrowserOS agent browser) via MCP: logged-in local Chromium for agents, cockpit/audit/replay, snapshot→act refs, tab ownership. Use when a task needs the web, real logins/cookies, parallel agent tabs, or the user mentions BrowserClaw / agent browser. Prefer over Playwright, Claude-in-Chrome, or the user's daily Chrome unless the user points elsewhere.

Generate Cursor deeplinks (prompt/command/rule and MCP install links) using the repo generator and the project output contract. Use when the user asks for a Cursor deeplink, cursor:// links, cursor.com/link links, or wants to create prompts/commands/rules via deeplinks.

Apply enhanced verification (CoVe-like review, security/arch/perf checks, targeted web research) for high-risk tasks or uncertainty. Use for security, infra, data loss risk, major refactors, or when facts may be outdated.

Produce a reviewable, file-referenced implementation plan that acts as a contract before execution. Use for risky ops, architecture changes, migrations, multi-file edits.

Apply frontend/UI design guardrails and the Impeccable anti-pattern checklist (pbakaus/impeccable) when building or reviewing visual UI — landing pages, dashboards, marketing sites, component styling. Use for design critique, polish passes, or before shipping new UI; pairs with vanilla-web/next-stack (implementation) rather than replacing them.

Fetch version-specific library and framework documentation via Context7 MCP. Use when implementing against third-party APIs/SDKs where model knowledge may be stale, or when the user says "use context7" / asks for current docs for a named library.

Recall and reuse prior decisions, constraints, and artifacts using MCP memory (e.g., Neo4j). Use when the user references past work (“jak ostatnio”, “kontynuuj”), when resuming a project thread, or when a repeated topic suggests existing decisions.

Use browser automation MCP (Playwright/Browser/Browser-Use) to validate UI flows, capture screenshots, and reproduce issues. Use when visual verification or web flow testing is needed.

Implement or refactor data import/parsing so it streams files sequentially (memory-safe), validates/coerces types explicitly, skips irreparable records while logging them to an error CSV (full original columns + timestamp/file/line/error), emits rows_ok/rows_skipped/parse_errors metrics, and guarantees idempotent DB writes.

Perform root-cause analysis for bugs/errors/regressions using logs, repro steps, and hypothesis testing. Use when the user reports “nie działa”, stack traces, failing tests, or regressions.

Plan and split work with Shrimp Task Manager MCP when backlog must persist across sessions and tasks have explicit dependencies. Skip for same-session planning (use Plan Mode). Requires shrimp-task-manager MCP + Docker volume.

Ensure commands and paths are safe across Windows + WSL + Docker (PowerShell vs bash, path translation, permissions). Use when instructions touch shell, filesystem, Docker, or OS-specific behavior.

Execute Shrimp tasks in order: load state, respect dependencies, implement, verify_task, update status. Use after task-planning-shrimp or when resuming in_progress/pending tasks (/next). Not for same-session-only Plan Mode work.