environment

Deep codebase exploration to fully understand the current project before starting work. Use this skill whenever the user says "onboarding", "onboard", "get familiar", "learn the codebase", "understand this repo", "context load", or at the start of a session when the user wants Claude to be fully prepared to work on the project. Also use when switching to an unfamiliar part of the codebase.

Core agent-browser usage guide. Read this before running any agent-browser commands. Covers the snapshot-and-ref workflow, navigating pages, interacting with elements (click, fill, type, select), extracting text and data, taking screenshots, managing tabs, handling forms and auth, waiting for content, running multiple browser sessions in parallel, and troubleshooting common failures. Use when the user asks to interact with a website, fill a form, click something, extract data, take a screenshot, log into a site, test a web app, or automate any browser task.

Audit a ralphex/farm implementation plan (docs/plans/*.md) for self-containment and task concreteness before running it through ralphex or queuing it on ralphex-farm. ralphex runs every task and review in a FRESH session that re-reads only the plan (plus the committed repo and a progress log), so anything that tells the agent to "go read elsewhere", "match 1:1 with an external codebase", or just names a big outcome without saying how to build it silently wastes huge effort and may never converge. Fans out parallel auditor agents that check the plan against rules from real farm failures and returns a farm-ready / fix-these verdict. Use whenever a ralphex plan was just written or is about to be executed or queued - "audit this plan", "check my plan before the farm", "is this plan farm-ready", "review the plan before ralphex", or right after /ralphex-plan or /planning:make and before creating a farm ticket. Do NOT use to create plans (that is ralphex-plan) or operate the farm (that is ralphex-farm).

Cross-repo GitHub radar - a personal snapshot of the open pull requests and issues that involve you, run from any directory (account-wide, not tied to the current repo). Covers your own repos (every open PR/issue, any author) plus anywhere you are involved on other people's repos (authored, assigned, mentioned, or review-requested). Use this whenever the user wants a GitHub overview or status, asks what PRs or issues need their attention, what is waiting on them, what they have open on GitHub, what they might be forgetting, before standup, or when catching up after time away - including phrasings like "github radar", "my open PRs", "anything waiting on me on github", "what did I leave open", even if they don't name a specific repo. Also supports an optional single-repo scope - when the user names one repository ("github radar for turtle-hub", "what's open in pkarpovich/tuclaw", "PRs and issues in <repo>"), pass that repo and the report covers only it.

Work with ralphex-farm: an autonomous executor that polls Linear, picks up Todo issues whose description carries a `<!-- ralphex-farm -->` YAML block (repo + plan + branch, optional mode), runs ralphex against the named plan inside Docker, and opens a PR. Invoke whenever the user wants to: create a Linear ticket for a plan ("create a ralphex farm task", "add task for plan", "queue plan to the farm"), re-run or recover a failed task ("re-run", "recover", "review only", "move back to Todo"), add a new repository to the farm, trigger an immediate sync, debug why a ticket was not picked up, or answer questions about how the farm works. Also trigger on any mention of the `<!-- ralphex-farm -->` metadata block, repos.yaml in the farm context, `/var/ralphex/...` host paths, the farm's `/api/sync` or `/api/repos` endpoints, or Claude plugin provisioning in the farm.

Go 1.25+ development standards and coding conventions. Use when writing, reviewing, or refactoring Go code. Triggers on Go file creation/editing, code reviews, architecture decisions, testing strategies, error handling patterns, concurrency, or when user asks about Go best practices.

End-to-end preparation of a new Lego Cubes podcast episode (Pavel's Russian-language podcast about content + dev projects). Walks through 7 stages with check-ins — collect Content bullets from Tuclaw/Podcast files, fill the episode note, suggest an episode title in the established poetic style, find IMDB/Steam/Goodreads links, produce a LEGO-style cover image prompt for Nano Banana Pro, after the recording publish the .mp4 + .jpg + .nfo bundle to the NAS for Plex/Kodi, and finally archive the per-content impression files into References/. Trigger on any mention of preparing, assembling, filling, publishing, or archiving a Lego Cubes episode — including phrases like "prepare episode 30", "Lego Cubes 30", "the next episode", "fill out the episode note", "I recorded the podcast", "put it on the NAS", "make the nfo", "faststart the recording", "move impressions to References", "archive the episode", or a bare episode number with podcast context. Pavel typically writes these requests in Russian — same triggers app

Modern Python 3.13+ development standards and best practices. Use when writing, reviewing, or refactoring Python code. Triggers on Python file creation/editing, code reviews, architecture decisions, async patterns, type hints, testing strategies, or when user asks about Python best practices.

Modern CSS (2026) style guide for plain CSS and CSS Modules. Apply automatically when writing, editing, or reviewing .css, .module.css, or .scss files. Targets Chrome (last 4) and Safari 26+. No fallbacks, no @supports hedges, no legacy support. Prescribes Grid/Flexbox over position:absolute for layout; intrinsic sizing (min-content, fit-content, clamp()) over fixed pixels; container queries for component-level responsive; CSS custom properties for every repeated value; OKLCH color space; light-dark() for theming; :has() and @scope over deep selector chains; @layer for cascade order over !important; native CSS Nesting; View Transitions API for navigation; scroll-driven animations via animation-timeline; anchor positioning for floating UI; text-wrap balance/pretty for typography; field-sizing for auto-resizing form controls; popover API for overlays. When editing existing CSS, apply these rules even if surrounding code is older - do not cargo-cult legacy patterns.

Diagnose and fix excessive Postgres egress (network data transfer) in a codebase. Use when a user mentions high database bills, unexpected data transfer costs, network transfer charges, egress spikes, "why is my Neon bill so high", "database costs jumped", SELECT * optimization, query overfetching, reduce Neon costs, optimize database usage, or wants to reduce data sent from their database to their application. Also use when reviewing query patterns for cost efficiency, even if the user doesn't explicitly mention egress or data transfer.

Find high-quality Claude Code skills for a topic by spawning parallel subagents that audit each candidate's AUTHOR (not the SKILL.md text, which is gameable in five minutes by a slop generator). Use whenever the user asks to find, search for, recommend, or pick a skill for a topic - phrases like "find me a skill for X", "find skill for X", "recommend a skill for X", "I need a skill that does X", "look for skills about X". Each candidate goes through prose-driven investigation across four signals (OSS contributions in the topic, repo context, external reputation, the care put into the author's own other repos), and you return the top 3 with categorical verdicts (real_expert / promising_unknown / skip), never numeric scores. Do NOT use for auditing the user's own existing skills (that's skill-audit) or creating a new skill from scratch (that's skill-creator).

Expert prompt engineering for Claude 4.x models (Sonnet 4.5, Opus 4.5, Haiku 4.5). Use when creating system prompts, optimizing existing prompts, designing agentic workflows, or improving prompt effectiveness. Triggers on requests like "optimize this prompt", "write a system prompt", "improve these instructions", "create an agent prompt", or any task involving prompt design for Claude.

Grilling session that challenges your plan against the existing domain model, sharpens terminology, and updates documentation (CONTEXT.md, ADRs) inline as decisions crystallise. Use when user wants to stress-test a plan against their project's language and documented decisions.

Use rust-analyzer's Structural Search and Replace (SSR) to change lots of Rust code. SSR matches by AST structure and semantic meaning, understanding type resolution and path equivalence.

Rust coding style guide. Apply automatically when writing or modifying Rust code. Enforces for-loops over iterators, let-else for early returns, variable shadowing, newtypes, explicit matching, and minimal comments.

Rust documentation conventions (RFC 1574). Apply when writing doc comments on public Rust items. Covers summary sentences, section headings, type references, and examples.

Create or extend notes in Pavel's personal Obsidian vault at `/Users/pavel.karpovich/Obsidian/PK Workspace/`, following his adapted kepano-style vault grammar (plural categories, YYYY-MM-DD dates, heavy internal wikilinking, root-first organization, 1-10 rating). Use this skill WHENEVER Pavel wants to capture something into his vault - even when he does not say "Obsidian" or "vault" explicitly. Trigger phrases include "save a note about <X>", "log this", "add to vault", "I watched <movie>", "I read <book>", "I finished <game>", "I bought <jacket/jeans/sneakers>", "I met <person>", "new idea", "evergreen about", "meeting note for", "project note on", "weekly note", "add <X> to references", and any bare mention of a movie/book/game/show/person/place/product that implies cataloguing. Knows the 19 active note types - Journal, Meeting, Project Note, Task, Idea, Evergreen, Quote, Job Interview, Rent Payment, Weekly, Movie, TV Show, Game, Book, Person, Product, Company, Place, Recipe, Podcast - and where each one li

Query Obsidian vault files by frontmatter properties using Dataview-style syntax, AND discover what property names and values actually exist in the vault. Use when the user asks to find notes, search the vault, or filter notes by metadata (tags, status, dates, categories, projects). ALSO use proactively before writing or editing a vault note when the canonical schema is not 100% certain: run `ovq --values <field> --count` to discover real category and property values, and run a `categories contains` query plus read 1-2 sample files to confirm property names before assuming them. Triggers include "find notes where", "search vault for", "which notes have", "find meetings", "find projects with status", and any agent task that touches Obsidian frontmatter on a vault you do not own.

Capture the full state of the current session into a handoff document so a fresh agent can pick up cold with zero loss of context. Not a summary, not a brief; an exhaustive record of every task tackled (what we did, why we did it, how we did it), every user preference learned, every unresolved decision, and every loose end. Use when the user says "handoff", "save session", "checkpoint", "session snapshot", "next agent will continue", "wrap up the session", "hand off the context", or when wrapping up a multi-topic session and the next conversation needs the full picture. Saves to `<repo-root>/.local/handoffs/` so the doc lives with the project and survives across sessions. SKIP for short single-task sessions where the artifacts (commits, PRs, file edits) already carry the whole story - handoff only earns its keep when the session contains material that lives only in the conversation (decisions, dead ends ruled out, user corrections, preferences learned).

Audit one or more Claude Code skills against best practices and report violations. A Python script runs mechanical checks (1024-char description limit, name format and dir match, frontmatter required fields, body 500-line target, ASCII hyphens, bilingual trigger padding, reference-file ToC, allowed-tools syntax). The model adds semantic checks on top (what + when, negative cases, concrete triggers, progressive disclosure, sibling-skill coordination). Use when the user asks to audit, check, review or validate a skill, before adding a new skill to the dotfiles, or before committing changes to an existing skill. Also activate when asked HOW to write a good skill, what fields to use, when to split into references/, or about the 1024-char description limit, the bilingual-padding pitfall, or the progressive-disclosure pattern.

Convert any web article, blog post, documentation page, or release-notes URL into clean markdown with a YAML metadata block (title, author, site, published date, source URL, word count) via the public defuddle.md HTTP API — just `curl https://defuddle.md/<url-without-scheme>`. Use this INSTEAD of WebFetch whenever the user shares a URL that points at readable prose, even if they did not explicitly say "extract", "clean up", or "summarize" — e.g. "open <url>", "what does this article say", "read this blog post", a bare URL pasted into chat, or "what does <author> say about X" with a link. WebFetch summarizes the page through an LLM and drops the metadata, so quotes drift and the publication date is lost; defuddle returns the article text verbatim, which matters when the user wants to discuss, cite, or archive what was written. SKIP this skill (use WebFetch or another tool instead) when the URL ends in `.md` / `.txt` (already plain), `.pdf` / `.png` / other binary, is a JS-rendered SPA or dashboard (defuddle ca

Draft a Slack / email / chat message, propose it in chat first, and save to `<repo-root>/.local/messages/` only after the user approves. Invoke when the user says "draft a message", "write to <person> about", "slack <person>", "send a message to <person>", or any similar request to compose a chat / email message for review before sending. Pavel often asks in Russian — same triggers apply regardless of language.

Fetch up-to-date, version-specific documentation and code examples for libraries via the Context7 platform. Use when you need authoritative API references, current syntax, or to verify a library behavior beyond your training cutoff. Especially valuable for fast-moving libraries (React, Next.js, Tailwind, FastAPI, Pydantic, LangChain, LangGraph, Anthropic SDK, OpenAI SDK, Bun, Deno, Prisma, tRPC, OpenTelemetry SDKs, etc.).

Search real-world code examples across millions of GitHub repositories using grep.app. This skill should be used when looking for implementation patterns, API usage examples, library integration patterns, or production code references. Supports literal code search, regex patterns, and filtering by language/repo/path.

Multi-pass code review of branch changes against the default branch. Broad parallel sweep across quality, implementation, tests, simplification, and docs, with optional external (codex) second-opinion consultation. Reports findings only — does not modify code. Use when user asks for "deep review", "thorough review", "multi-pass review", "review my branch".

Audit, clean up, and structure SSH config and keys. Use when user asks to "review ssh config", "clean up ssh", "check ssh keys", "add ssh host", "remove dead ssh hosts", "organize ssh keys", or any SSH configuration management task. Also triggers on mentions of ~/.ssh/config, SSH key organization, or 1Password SSH agent setup.

Audit CLAUDE.md files against best practices. Analyzes length, structure, anti-patterns, and generates improvement recommendations with concrete suggestions. Invoke when user says "audit claude.md", "check claude.md", "review claude.md", "improve claude.md".

Build multi-step LLM agents for the Continuum system. Use when creating agent definitions with nodes, edges, tools, and execution flows. Triggers on requests to "create an agent", "build a workflow", "design an agent graph", or when discussing Continuum agent architecture. Supports parallel execution, conditional routing, tool integration, and structured outputs.

Design patterns and architectural principles for maintainable Python code. Use when designing new components, refactoring existing code, choosing between patterns (factory vs simple dict, inheritance vs composition), planning module structure, or implementing dependency injection. Triggers on architecture discussions, "how should I structure", refactoring requests, or when complexity needs to be managed.

Execute Bruno .bru API requests via curl, or create new .bru files. Use when user asks to run, execute, or test an API request from a Bruno collection, references a .bru file, or wants to create a new Bruno request file.

Analyze Langfuse trace JSON exports using jq queries. Use when user provides a Langfuse trace file (.json) for analysis, debugging agent behavior, checking token usage, or investigating tool calls. Triggers on phrases like "analyze trace", "check this langfuse", "debug agent run", or when user shares a trace JSON file path.