pbi-task-split
PBI（Product Backlog Item）を agent 実装向けの粗めの実装タスクまたは phase gate に分割する。親 issue の ready 判定、implementation contract、blackbox/runtime acceptance、依存関係、review checkpoint を整理し、方針未決・仕様矛盾・単なる明確化を sub-issue 化しない。UI PBI では approved screenshot/golden が source of truth の場合だけ visual-ui-contract を使い、共通部品・tokens・UI vocabulary が source of truth の場合は vocabulary contract で並列実装可能にする。
plan-on-md
A skill to save plans to local Markdown after creating them in plan mode
plan-steps-split
Split plan steps into individual markdown files in Obsidian vault. Use after creating a plan with plan-on-md when you want each step as a separate trackable file.
prototype
Generate 3 HTML prototype variants using a team of parallel agents. Each agent creates a distinct design pattern (e.g., hover effects, animations, layouts) based on user prompts or docs/goal.md. Use when the user wants to explore multiple design directions for a UI component.
existing-code-first
Use for any coding implementation, bug fix, refactor, test addition, code review, or agent handoff where new code may be added or existing behavior may be changed. Before creating new functions, classes, files, utilities, services, hooks, components, schemas, constants, validators, test helpers, or dependencies, search existing code for equivalent or similar responsibilities and prefer reuse, extension, relocation, or consolidation over parallel reimplementation.
refactoring
Refactor code to improve maintainability by consolidating redundant logic and applying DRY principles
review-comment-triage
Triage GitHub PR review comments after a PR is open and decide whether each comment is worth fixing, should be rejected, is optional cleanup, or needs user/product judgment. Use when the user asks if a PR review comment is valid, worth fixing, should be addressed, or asks to inspect GitHub review feedback without immediately changing code.
review-diffs
Review git diffs with a manager-led workflow that preserves issue requirements. First derive a shared review contract from the user request and any linked issue or PR, then select review perspectives from a catalog (a mandatory floor plus diff-specific optional lenses, including a rendered UI/visual lens with web and Flutter-golden backends), run those perspectives as parallel tmux review agents, and finally perform a manager pass that checks total requirement coverage before summarizing.
tax
Review transaction exports for Japanese tax filing, classify deductible expense candidates, and import selected rows into MoneyForward with safe browser automation, verification, and reconciliation.
tdd-plan-review
Review and reorder existing implementation plans to follow TDD methodology. Use when user has a plan that puts implementation before tests, or when validating plan structure for TDD compliance. Reorganizes steps to ensure test-first approach.
tdd-plan
Create implementation plans following TDD methodology with test-first approach. Each test file is immediately followed by its implementation (fine-grained RED→GREEN cycles).
tmux-sender
tmux の別ペインにコマンドを送信する。「ペインで実行して」「tmuxで送信」などのリクエストで使用。
ui-critique
Critique a rendered UI for visual hierarchy, clutter, "AI-generated" tells, and drift from an established design world — when there is no golden image to match against, only a sense that "it looks off / too busy / too AI". Renders the real pixels (headless browser screenshot), judges against a named failure-mode taxonomy, and returns ranked, subtraction-biased fixes that preserve the world. Triggered by "review this UI", "this screen looks cluttered/AI-ish", "critique this design", "why does this feel off", "keep the vibe but clean it up".
visual-ui-contract
Create and enforce visual contracts for screenshot-driven UI work only when an approved screenshot, mockup, ideal image, golden baseline, or explicit visual fidelity target is the source of truth. Do not use for UI work governed by shared tokens, primitives, or a UI vocabulary unless an approved reference image is also binding.
wiki-knowledge-card
Generate a polished, self-contained dark-theme HTML "visual study sheet" for a term/word/concept in the Obsidian Wiki at ~/uuta/Wiki, then link it from a sibling Markdown note. Use when the user wants a visual vocabulary/knowledge card, an HTML version of a term explanation, or says things like "make a visual card for X", "knowledge card", "Wikiに<term>のhtmlを作って", or invokes /knowledge and asks for an .html. Produces <Term>.html (house style — hero, SVG diagrams, bilingual EN/JP) and ensures <Term>.md contains an open link.
wordplay-manzai
日本語の掛け合い漫才台本を作るスキル。ユーザーが「ナイツ風」「語感縛りの漫才」「特定の漢字や響きをねじ込む漫才台本」などを求める時に使用。特定の存命コンビを直接模写せず、淡々と間違えるボケ、冷静から激昂へ移るツッコミ、縛りワードへの異常な執着、法則の看破、強引なオチを持つオリジナル台本を作る。
worktree-env
Copy or link ignored env-like files into a git worktree from a source checkout or secrets directory using a tracked manifest such as `docs/env-paths.txt`. Use for worktree bootstrap when a repo declares local runtime/build config paths, especially after `agent-workspace` creates or reuses an issue worktree.
acceptance-criteria-clarifier
Clarify ambiguous implementation requests into explicit, testable acceptance criteria before coding, issue splitting, agent delegation, or review. Use when success is vague, scope or out-of-scope is missing, multiple valid interpretations exist, blackbox/runtime verification is unclear, risky assumptions are needed, or the user asks to make requirements or acceptance criteria explicit. Avoid using for trivial mechanical edits where the requested outcome is already objectively verifiable.
activity-digest
その日の作業ログを横断して activity digest を作る。ユーザーが「activity digest」「今日何をしたかログから見たい」「作業ログをまとめたい」「証跡ベースで振り返りたい」などを求めたときに使う。特定プロジェクト名に依存せず、その日に触った作業対象を git、GitHub、Codex、Claude、tmux、shell snapshot から復元する。
agent-workspace
Prepare and clean up a project workspace and git worktrees so agents can work safely per issue. Use `main/` as the base checkout and `.worktrees/<issue_number>/` as the dedicated worktree directory. After creating or reusing a worktree, copy ignored env/local config files declared by `docs/env-paths.txt` from `main/` into the worktree. Do not create worktrees directly under home or inside an existing checkout. Use branch names like `feat/<issue_number>`. Use this for requests such as "create a worktree", "set up an agent workspace", or "prepare an issue-specific working directory".
blackbox-test-design
Turn an input specification into a minimal, high-coverage black-box test set using Equivalence Partitioning + 2-value Boundary Value Analysis. Use when designing or reviewing test cases for functions, forms, validators, or APIs from their input domains (ranges, formats, enums) — not from source code. Triggered by phrases like "design test cases", "what should I test for this input", "boundary tests", or "EP/BVA for this field".
claude-tmux-pm
tmux 上の Claude Code セッションに GitHub の sub-issue を順番に割り当てるスキル。親 Issue 配下の sub-issue を番号順に見て、明示的な implementation contract / Done when / Not done if / blackbox verification を持つ ready issue を `feat/{issue_number}` ブランチの専用 worktree（basename は issue number）で Claude Code に実装させる。workspace / worktree の配置規約は `agent-workspace` に従う。Claude は実装と検証まで行い、Codex が phase gate や diff review を挟んでから commit / push / PR を行う。L または contract 不足の issue は通常割り当てず、先にユーザーと詰める。「Claude Code に割り当てて」「tmux の Claude に投げて」「agent に issue を振って」などで使用。
claude-tmux-review-loop
tmux 上の Claude Code に issue または明示タスクを割り当て、進捗を監視し、実装完了ごとに diff review を行い、findings があれば同じ pane に修正依頼を返して review clean になるまで繰り返すスキル。仕様不足や判断待ちが出たらユーザーに確認する。「Claude を監督して」「tmux の Claude を見ながらレビューを回して」「review clean になるまで Claude と往復して」などで使用。
codex-review
tmux で動作する Codex CLI にプロンプトを送信し、ポーリングでレスポンスが完了するまで待機して取得する。「codex に聞いて」「codex で調べて」などのリクエストで使用。
component-extraction
Use when a UI change, bug fix, or review touches repeated inline UI blocks and Codex should decide whether to extract a small presentational domain component. Helps identify repeated UI with shared visual or interaction responsibility, keep domain state management in callers, define explicit component state such as enabled/disabled, and avoid over-extracting generic atoms.
github:create-pr
Create a GitHub PR using the current branch's issue title and project PR template. Designed to be called after all dev steps complete.
daily-review
日次の振り返りを定型で進める。ユーザーが「今日の振り返り」「日次レビュー」「daily review」などを求めたときに使う。まず activity digest とログから実際の活動を調査して要約し、その要約を踏まえて短い質問で不足分を集め、自然な振り返り文、明日の最重要1個、必要なら template/checklist/skill/automation 化候補まで整理する。
github-issue-images
Fetch and inspect screenshots embedded in GitHub issues or pull requests, especially `github.com/user-attachments/assets/...` URLs that often return `404` without GitHub authentication. Use when an issue or PR includes images and an agent needs the actual pixels from CLI or tmux.
sub-issue-maker
Provide coarse, agent-ready task granularity for cutting GitHub sub-issues. Use implementation contracts and blackbox acceptance, avoid creating sub-issues for policy clarification, unresolved decisions, or specification contradictions.
simple-review-format
Output in the specified format when a review of specific markdown is requested.
review-format
Output in the specified format when a review of specific markdown is requested.
learn
AI-Driven Learning Assistant. Structured learning based on Plan → Learn → Practice → Track methodology. Stores everything for a topic in a single ~/uuta/Learning/{topic}/curriculum.md page.
frontend-design
Create distinctive, production-grade frontend interfaces with high design quality. Use this skill when the user asks to build web components, pages, or applications. Generates creative, polished code that avoids generic AI aesthetics.
rag-systems
Build RAG systems - embeddings, vector stores, chunking, and retrieval optimization
ai-agent-basics
Master AI agent fundamentals - architectures, ReAct patterns, cognitive loops, and autonomous system design
multi-agent
Build multi-agent systems - orchestration, coordination, workflows, and distributed architectures
llm-integration
Integrate LLMs into applications - APIs, prompting, fine-tuning, and context management
agent-memory
Implement agent memory - short-term, long-term, semantic storage, and retrieval
agent-safety
Ensure agent safety - guardrails, content filtering, monitoring, and compliance
tool-calling
Implement tool calling - function schemas, API integration, validation, and error handling
security-patterns
Security architecture, authentication, authorization, and compliance patterns
devops-patterns
Infrastructure, deployment, and operations patterns for Docker, Kubernetes, and CI/CD
documentation
API documentation with OpenAPI and developer portals
frontend-patterns
Frontend development and API integration patterns for React, TypeScript, and state management
versioning
API versioning strategies and backward compatibility
backend-patterns
Production-grade backend patterns for Node.js, Python, Go, and Java/Spring frameworks
database-patterns
Database design, optimization, and caching strategies for SQL, NoSQL, and Redis
testing
API testing strategies and contract testing
api-architecture
REST, GraphQL, and hybrid API architecture patterns for building scalable and maintainable APIs
scaling-patterns
Enterprise scaling patterns for microservices, event-driven architecture, and distributed systems
Page 1447 of 1708 · 85399 results
