dotfiles

Unified pull request workflow: prepare a safe branch, run read-only review, fix Required findings, create/update a GitHub draft PR, and verify CI plus automated review comments. Subcommands: `create` (simplify + PR, skip code review), `review` (read-only findings only), `fix` (Required fixes + commit, no push), `simplify` (simplification only), `verify` (existing PR checks/reviews only). Use when creating or reviewing PRs, self-reviewing changes, simplifying PR changes, or requesting "PR作成", "レビュー", "簡素化". Do NOT use for responding to review comments or a lightweight review of the current working diff without the PR workflow (use the built-in /code-review instead).

Orchestrate medium-to-large software development by turning a broad goal into an implementation and verification plan made of PR-sized subtasks and operational nodes, defining verification harnesses and gates for each node, and stopping after epic creation on the first invocation until the user explicitly asks to implement or execute approved nodes. Use when the user asks for large-scale development planning, PR decomposition, multi-PR execution, technical selection, or autonomous delivery of a broad epic. Do NOT use for a single small bug fix, one obvious PR, pure code review, ordinary SDD work that already fits in one spec/PR, or implementing an epic during the same first call that created it.

Create a git worktree using the project's `{repo}-worktree/{branch}` convention, fetching origin first and basing new branches on the highest-priority remote ref (`origin/staging`, then `origin/main`, then `origin/master`). Use when the user asks to create a new worktree or start work on a new branch in a worktree (e.g. "ワークツリー作って", "worktree 作成", "新しい作業ブランチ", "新規ブランチで作業"). Do NOT use for removing/listing worktrees (use `gwc` / `git worktree list`), or for the PR flow (use my-pr — it creates its own worktree on protected branches).

Explain an existing GitHub pull request in the CCV chat as a code-oriented walkthrough for human review. Covers the PR intent, every changed file, behavioral and contract changes, faithful code excerpts, flows, and review attention points while omitting test-detail explanations. Use when the user asks to understand, inspect, walk through, or visually review a PR created in CCV. Do NOT use for creating or updating PRs, posting review comments, fixing findings, running tests, or performing a formal findings-only code review.

Scan a codebase for refactoring opportunities, write a SARIF-compatible report, and register each finding as a GitHub issue labelled by severity, effort, and Fowler category. After issue creation, report the created issues and suggested feature names without choosing an implementation workflow. Use when the user asks to "リファクタリング候補洗い出し", "技術的負債リスト化", "refactor scan", "refactoring issues", or requests surveying a repo for refactor work. Do NOT use for applying refactors, code review of a specific PR (use `my-pr review`), or for fixing a specific bug.

Generate release notes from merged PR history and commits. Categorizes changes by type and creates GitHub releases with semantic versioning. Use when the user asks to create a release or generate release notes (e.g., "リリースノート", "release note", "リリース作成"). Do NOT use for changelogs, commit summaries, or PR descriptions.

Interactive guide for creating and improving Claude Code skills (SKILL.md files). Use when the user explicitly asks to create, update, or fix a **skill definition**. Triggers: "スキルを作って", "スキル作成", "create a skill", "new skill", "build a skill", "improve this skill", "update skill", "fix my skill", "skill doesn't trigger". Do NOT use for general coding tasks, PR creation, code review, or any task that merely contains the word "create" without referring to skill authoring.

Manage Claude Code Viewer (CCV): preview in-progress CCV changes on ccv-preview, register HTML artifacts, download files, or schedule jobs via CCV local API. Use when user asks for CCV preview, ccv-preview, browser visual QA, UI/UX validation, CCV artifacts, CCV file download, or CCV scheduler jobs. Treat preview/プレビュー as the CCV app preview service first; use artifacts only when the user explicitly asks for artifact/dashboard/report HTML. Do NOT use for generic web app previews outside CCV, production deployment, Cloudflare config, or Anthropic remote triggers.

Delegate tasks to an external CLI agent (OpenAI Codex, Google Gemini, or Claude Code) for second opinions or parallel execution. Use when the user explicitly mentions "codex", "gemini", or "claude" or asks to delegate work to one of them (e.g., "codexで実行", "codexに聞いて", "geminiで実行", "Geminiに聞いて", "claudeで実行", "Claudeに聞いて", "run with codex", "ask gemini", "ask claude"). Also use when the user mentions Claude's "fable" model alias for delegation or consultation (e.g., "fableで実行", "fableに相談", "Fableに聞いて", "ask fable"). Do NOT use for general coding tasks that don't mention codex, gemini, claude, or fable.

Disciplined bug and performance-regression diagnosis loop: build a feedback loop, reproduce, hypothesize, instrument, fix, and regression-test. Use when the user reports a bug, says something is broken/failing/throwing, asks to debug or diagnose, or describes a performance regression. Do NOT use for planned feature work or broad architecture review.

Clean up disk space by removing caches and unnecessary files on macOS and Ubuntu. Targets: Docker (images, build cache, volumes), package managers (Homebrew, apt, pip, npm, yarn), and temporary files (__pycache__, .mypy_cache, Xcode DerivedData). Reports disk usage before/after with freed space summary. Use when the user asks to free disk space, clean caches, or mentions "ディスク整理", "キャッシュ削除", "disk cleanup", "free space", "clean docker", "clean caches". Do NOT use for file organization, deduplication, or finding large files.

Remove "AI slop" from Japanese prose so AI-drafted text reads like a human wrote it. Fixes not just em-dashes and pet phrases, but absent authorship, proposition-style H2s, grandiosity, both-sides hedging, and uniform rhythm. Use when drafting, editing, or reviewing Japanese text, or when the user asks to strip AI臭 / AIっぽさ — e.g. "AI臭を消して", "この文章をレビューして", "公開前チェック", "下書きを直して". Do NOT use for English prose or for non-prose code review.

Delegate subagent-suitable work to subagents and keep the main agent focused on orchestration, decisions, integration, and final verification. Use automatically before non-trivial work when any part can be isolated as independent research, implementation, testing, review, or multi-step worker work; also use when the user says "subagent", "委譲", "並列実行", or asks to use lighter models for worker tasks. Do NOT use for simple one-shot answers, tiny one-file edits, irreversible actions, or decisions that require user approval.

Check the latest Claude (Anthropic), OpenAI Codex, and Google Gemini model releases from official primary sources, then update model selections in this dotfiles repo (home/dot_claude/settings.json, home/dot_codex/private_config.toml.tmpl, home/dot_gemini/settings.json). Also scans the invoking repository for hardcoded model IDs in GitHub Actions workflows, Python code, and shell scripts, and offers to bump them. Use when the user asks to "モデル更新", "モデルを最新に", "最新モデル確認", "model bump", "update models", or "claude/codex/gemini のモデル更新". Do NOT use for one-off model selection in a single conversation, general model questions, or non-config code changes.