base

Create a timestamped memo with ULID in ~/projects/private-content/memo/

Complete git workflow - create branch, split commits, and create PR

Summarize recent Claude Code release notes in chronological order

General-purpose problem-solving agent for tasks that don't fit specialized agents. Use as fallback for broad inquiries, workflow advice, architecture explanations, and cross-domain tasks.

Create a git branch with an appropriate name derived from current changes. Use when the user asks to create a branch, name a branch, make a branch for current work, or wants to move changes to a new branch. Also use when changes exist on main/master that should be moved to a feature branch before committing.

Generate CodePen 2.0 demos with copy-paste ready code. Use when the user wants to create interactive demos, playground examples, UI component showcases, browser API demonstrations, CSS animations, or any front-end experiment for CodePen. Also trigger when the user mentions "CodePen", "pen", "playground demo", or wants to share a working code snippet that runs in the browser.

Create GitHub pull requests following required project conventions. Always consult this skill when creating a PR, opening a pull request, or submitting changes for review — including "PR作って", "プルリクエスト作成", "create a PR", "make a pull request", "submit for review", "gh pr create", or any request to prepare changes for review on GitHub. Contains mandatory title format [base-branch] type: description, required --draft flag, and --assignee kkhys that differ from defaults and cannot be inferred without this skill.

Creates a personal activity summary across all GitHub repositories for reflection. Use when the user wants to create weekly reports, work reviews, task summaries, activity retrospectives, or review their recent contributions. Also trigger when asked "what did I do this week", "summarize my work", "create a work log", "振り返り", or any variation of reviewing personal GitHub activity over a time period.

Addresses unresolved review comments on the current branch by orchestrating the full fix cycle: reading feedback, implementing changes, verifying, committing, and replying to reviewers. Use this skill whenever the user wants to fix PR review feedback, address reviewer comments, handle review points, or respond to code review. Also trigger on phrases like "レビューコメント直して", "fix the review comments", "address the feedback", "review 対応して", "レビュー指摘を修正", or any request to implement changes requested by PR reviewers. This is the primary skill for closing the PR review feedback loop.

Enforce Conventional Commits format for git commits. Always consult this skill when committing code changes — including "commit", "コミット", "コミットして", "変更を記録", "save changes", "stage and commit", asking whether to use feat or fix, deciding between squash and new commit, or any request to record, amend, or finalize changes in git. Contains mandatory project-specific conventions for type/scope format (plugin name as scope), squash-vs-new-commit strategy selection, and message structure that differ from defaults and cannot be inferred without this skill.

Orchestrate agent teams for parallel implementation tasks. Use when the user requests agent teams, multi-teammate coordination, or parallel feature development across independent modules. Also use when the user mentions 'spawn teammates', 'create a team', 'work in parallel', 'multi-agent', or describes tasks spanning multiple independent modules (frontend/backend/tests) that could benefit from simultaneous development, concurrent code review, or competing-hypothesis debugging.

Post review comments to a GitHub PR as a PENDING review using a dedicated workflow script (post-pr-review.sh) and severity tag system. Use after completing code review (pr-review-toolkit:review-pr, code-reviewer, silent-failure-hunter, or any review agent) to submit structured feedback. This skill provides the post-pr-review.sh script for correct PENDING review creation, 6-level severity tags ([critical]/[warning]/[suggestion]/[nit]/[question]/[praise]), diff line validation, and Japanese reporting. Always use this skill when posting, submitting, or sending review comments to a PR — do not attempt to call the GitHub review API directly without this skill's workflow.

Fetch and analyze unresolved PR review comments using a specialized GraphQL script that extracts thread-level data (thread IDs, resolution status, outdated flags, line ranges) not available through basic gh commands. MUST consult this skill whenever the user wants to check review feedback, see what reviewers said, view unresolved threads, understand pending review comments, or prepare to address reviewer requests on a PR. This is the required first step for the fixing-review-comments workflow. Also trigger on "review comments", "レビューコメント", "unresolved threads", "レビュー指摘", "reviewer feedback", "PR feedback", or any request to read, fetch, list, or summarize review discussions on a pull request.

Resolve all unresolved review threads on the current PR. Use when the user wants to bulk-resolve PR comments after addressing feedback, mark threads as done, or close out review discussions. Also trigger on "resolve threads", "スレッド解決", "レビューコメント解決", "resolve all comments", "コメント全部解決して", "close review threads", or any request to mark review threads as resolved on a pull request. Even if the user simply says "resolve" in the context of a PR review workflow, this skill is likely what they need.

Split large changes into logical commits by semantic meaning. Always consult this skill when the user wants to organize uncommitted changes into multiple commits. Trigger phrases include "split commits", "organize into commits", "break up changes", "separate commits", "logical commits", "コミット分割", "コミットを分けて", "コミットを整理", "変更を整理してコミット", "変更をコミットに分割", "うまくコミット分割", "コミット分割して", "コミットどう分ける". Also trigger when the user describes multiple types of uncommitted changes that need organizing — e.g., mentioning a mix of bug fixes and features, refactoring and new code, renames and logic changes, or any situation where changes span multiple concerns. Use when preparing changes for a PR with mixed concerns, or as part of the publish-pr workflow. Do not use for single-concern changes — those go directly to formatting-commit.

Save session knowledge to a secret GitHub Gist with built-in security scanning and content curation. This skill MUST be consulted before creating any knowledge Gist — it contains critical rules for filtering sensitive data (credentials, internal URLs, connection strings) that could leak via semi-public Gist URLs, and guidelines for structuring content so it's scannable and useful months later. Use whenever the user wants to save, record, capture, or preserve learnings, discoveries, findings, insights, or knowledge from the current session — whether to a Gist explicitly or to "somewhere searchable". Also proactively suggest saving at the end of productive sessions. Trigger phrases include "gist に保存", "ナレッジを保存", "知見を残す", "学んだことを保存", "save to gist", "capture findings", "save session knowledge", "どこかにまとめて", "後で見返せるように", or any request to persist technical learnings beyond the current conversation.

Live browser automation using the cmux browser CLI. MUST use this skill whenever the user asks to open a URL in a real browser, interact with a live web page, fill and submit web forms, click buttons on a page, scrape or read content from a website, inspect live page state (DOM, cookies, localStorage), navigate between pages, take browser screenshots, wait for dynamic/SPA content to render, work inside iframes, or debug a web UI by viewing it. Also trigger for Japanese phrases like "ブラウザで開いて", "ページを確認して", "スクショ撮って", "フォームに入力して", "ボタンを押して", "Cookie を確認したい", or any request to visually verify a deployed/staging/localhost page. This skill is for controlling a real browser instance — do NOT use it for writing HTML/CSS/JS code, writing Playwright/Cypress test scripts, making curl/wget/fetch API requests, reading local HTML files, or building UI components.

4 hooks across 3 events