feiskyer/claude-code-settings

Brainstorming explores user intent, requirements, and design options through collaborative dialogue before implementation. It is used before building new features, components, or systems when design decisions are involved, not for bug fixes or tasks with an obvious implementation path.

Codex runs OpenAI Codex/GPT models through the Codex CLI to autonomously implement code, review changes, and review plans. Use it when delegating coding tasks, getting a second-opinion review, or running tasks in sandboxed modes.

Deep Research orchestrates multiple headless claude -p subprocesses to split a research goal into parallel sub-goals, gather sources, and aggregate findings into a polished report file. Use it for systematic web research, competitive analysis, batch link processing, or evidence synthesis.

Fix GitHub Issue guides analysis, branch creation, implementation, testing, and PR submission for a reported issue. Use it when working from an issue number, issue URL, or a request to fix an issue end to end.

GPT Image generates and edits images using OpenAI GPT Image API models (gpt-image-2, gpt-image-1, etc). Use it when the user explicitly asks for OpenAI or GPT image generation or editing.

Grill Me runs high-intensity adversarial design reviews and grill sessions, challenging assumptions and missing constraints before implementation. It also maintains a domain model with a terminology glossary and ADRs as decisions crystallize.

Handoff compresses the current conversation into a handoff document, saved to $TMPDIR, so a new agent can continue the work seamlessly. Use /handoff when handing off a session to another agent or after a long conversation.

Nanobanana Image Generation creates and edits images using the Google Gemini API, and acts as the default image skill when a user requests image generation or editing without specifying another provider.

Pull Request Review runs a structured multi-agent workflow to review GitHub pull requests, using parallel specialized reviewers, adversarial verification, and false-positive filtering to produce actionable feedback. Use it when someone asks to review a PR, check a pull request, or references a PR number or GitHub PR URL.

Skill Creator guides the iterative process of drafting, testing, evaluating, and improving Claude Code skills. Use it when creating a new skill, updating an existing one, running evals, or optimizing skill descriptions and triggers.

Tech Article Translator translates English or Japanese technical articles and text into natural, fluent Chinese while preserving Markdown formatting and keeping technical terms untranslated. Use it when the user asks for Chinese translation of pasted text or a file.

YouTube Transcript Extraction extracts subtitles or transcripts from a YouTube video URL and saves them as a local text file. Use it when you need a video's captions for summarization, analysis, or repurposing.