Roland Tolnay

Triage pull request comments by fetching from GitHub, analyzing each with a systematic framework to decide fix vs. ignore, resolving dismissed threads, and planning fixes. Use when addressing PR feedback, handling code review comments, or following up on reviews.

Check prompt files for quality issues — wasted tokens, poor positioning, vague instructions. Use when reviewing changes to commands, skills, agents, SKILL.md, or any .md containing LLM instructions.

Create and configure Claude Code hooks for event-driven automation. Use when adding PreToolUse, PostToolUse, Stop, Notification, or other lifecycle hooks to settings.json. Covers all hook events, Python/UV scripts, matchers, and decision control.

Create a standalone prompt that another Claude can execute. Use when writing reusable prompts for coding, analysis, or research tasks. Saves to `./prompts/` as numbered `.md` files.

Create new Claude Code skills through collaborative conversation. Use when building a new SKILL.md, creating a skill from scratch, or turning a workflow into a reusable skill.

Create and configure Claude Code slash commands. Use when building custom /commands, editing command YAML frontmatter, adding arguments or dynamic context, or understanding command structure.

Create and configure Claude Code subagents for the Task tool. Use when defining agent types in agents/ config files, setting up agent tools and prompts, or launching specialized agents with the Task tool.

Generate install.js for Claude Code toolkit repos with manifest tracking, symlink/copy modes, and uninstall support. Use when creating a new distributable collection of commands, skills, agents, or references.

Create, update, and comment on Linear tickets. Use when creating issues, adding comments, changing status, assigning work, or referencing issue IDs like ABC-123. Examples: "create a ticket", "add a comment", "mark done", "my issues".

Design iOS/Android app icons with Nano Banana 2. Use when making, replacing, or refining an app icon — runs a discovery brief, writes the JSON prompt, critiques the generated PNG against a rubric, and outputs refinement prompts.

Apply consistent structure, tone, and formatting to README.md files. Use when drafting, rewriting, or reviewing a project README to make it scannable and developer-friendly.

Gather current information from the web using multiple sources in parallel. Use when the user needs facts, tools, best practices, or community opinions that may not be in training data. Triggers on: "search for", "look up", "find out", "what's the latest", "research".

Semantic web search with inline page content via EXA API. Returns actual page text with results — search + fetch in one free call. Use for discovery when you need content, not just URLs. Free tier: 1000 searches/month. Triggers on: "exa search", "search with exa", "free web search", "searchexa".

Send messages, search conversations, read channel history, and manage status in Slack. Use when sending DMs, posting to channels, searching messages, reading threads, editing messages, setting status, or scheduling messages. Examples: "message Roland", "search for deployment", "set my status to deep work".

Research subagent with API usage logging. Use instead of general-purpose when spawning subagents from the research skill. Has full tool access plus PostToolUse hooks that log WebSearch/WebFetch calls.

Recommend the best mental framework for your problem

Create a pull request with a context-rich summary derived from code changes and conversation context

Explain the current issue, options, or behavior so I can make a confident decision

Commit, comment, attach commit, and mark a Linear ticket as done

Generate a high-quality README for a public GitHub project

Analyze the current conversation and create a handoff document for continuing this work in a fresh context

Reflect on recent work, extract principles from past conversations, and capture learnings

After a fix or improvement, explore the codebase for other places where the same learning might apply

Analyze unpushed commits for squash and streamlining opportunities

Second-opinion verification of completed work — analyzes autonomously, then interrogates interactively before declaring issues

Work on an existing Linear ticket — diagnose, design, execute

Refactors Flutter/Dart code to follow quality guidelines. Applies code patterns, widget organization, folder structure, and simplification. Spawned by execute-phase/adhoc.

Analyzes Flutter/Dart code for structural issues during milestone audits. Reports findings only — does NOT fix anything.

Simplifies Flutter/Dart code for clarity, consistency, and maintainability. Spawned by execute-phase/adhoc after code changes.

Capture lessons from code refactorings into reusable docs for future LLM code writers. Use after completing a refactoring.

Extract backend patterns from current project into reusable CDN-ready module

Extract reusable Flutter/Dart patterns from project code into LLM-optimized reference docs. Use when capturing widget patterns, architecture conventions, or coding standards as documentation.

Extract UI patterns from current Flutter project into a reusable implement-ui skill

Learn from recent Flutter/Dart code changes and update the local coding principles file. Use after completing Flutter implementation work.

Generates a comprehensive CLAUDE.md file for Flutter projects through systematic discovery and user consultation

Organize Flutter/Dart code to follow project conventions. Use after implementation to restructure folders, fix widget file organization, align naming patterns, or clean up code to match project standards.

Reduce complexity in Flutter/Dart code. Use when code is too nested, hard to read, or has duplication. Extracts widgets, flattens logic, removes unnecessary abstraction.

Review Flutter/Dart code for architectural and structural design issues. Use when reviewing PRs, auditing widget design, evaluating state management, or identifying problems that make code hard to evolve.