agent-skill-set

Self-review loop for YOUR OWN PR — request AI reviews (Copilot + Gemini), apply their fixes, push, re-request, and repeat until clean. NOT for reviewing someone else's PR. Use when the user asks to self-review their PR, run the AI review loop, or wants Copilot + Gemini to review their own code. Trigger phrases include "self-review", "self-pr-review", "review my PR", "AI review my PR", "review loop", "copilot + gemini review", "run self-review on my PR".

Drive an already-running, LOGGED-IN Chromium browser on macOS via AppleScript — navigate, click, fill forms, scroll, and extract page content from the user's REAL session, with the browser left open and NO profile copy and NO restart. Use when the task needs the existing logged-in session (cookies, auth, current tabs) reused as-is. Trigger phrases: "chrome-use", "ブラウザから取得", "ログイン済みの Chromeで", "ブラウザを操作して", "このページ読んで", "Xのスレッド読んで", "ログインしたまま取得", "自分のセッションでスクレイプ", "read this page in my browser", "scrape with my session", "drive my chrome", "execute JS in my (logged-in) browser", "automate my logged-in browser". Chromium family (Chrome default; Brave/Edge/Arc/Vivaldi via --app). NOT for isolated/throwaway automation where login reuse is unneeded — use playwright-cli for that. macOS only.

Create, build, and install custom Claude Code skills into Hideki's local marketplace. End-to-end workflow from requirements gathering to a fully installed and usable skill. Use when the user asks to create a new skill, build a skill, make a plugin, add a new capability, or says "make me a skill for X". Also use when updating or reinstalling an existing custom skill. Trigger phrases include "create skill", "make skill", "new skill", "build plugin", "skill for X", "update skill".

Turn a finished grill-me (or any design-finalizing) session into a running implementation kickoff. Synthesizes the session's shared understanding into a self-contained implementation brief, hardens it with a small agent team and an adversarial codex-server review loop (until Codex approves), then launches a backgrounded, remote-controllable, autonomous Claude Code session (claude-bg / ccb-style: --permission-mode auto --allow-dangerously-skip-permissions --bg --remote-control) that runs /prd-council seeded with the brief to produce the execution-ready plan and begin implementation. Confirm-before-launch by default. Document/prompt generation + session spawning; it does not itself write feature code. Use AFTER a grill session when the user wants to hand the agreed design off to a fresh autonomous session. Trigger patterns (match any variation): grill-to-impl / grill to impl / grill→impl / grillの成果を実装へ / grillしたセッションから実装 / grillした内容でbriefを作って / briefを作ってCodexにレビュー / Codexにレビューさせてからprdーcouncil / 別セッションでprd-counc

Run OpenAI Codex via the Codex App Server (JSON-RPC-over-stdio) for streaming multi-turn chat sessions with persistent threads, structured output, image input, and rich event streams. Default entry point for ChatGPT/GPT/Codex-like conversations from the terminal. Uses the user's ChatGPT subscription (Plus / Pro / Team) exclusively via `codex login` — never consumes OPENAI_API_KEY billing, by design. Implementation: TypeScript on deno with npm:@openai/codex-sdk; spawns the existing system codex binary. Decoupled async invocation: every turn returns turn-id in <1 s; the actual SDK work runs in a detached worker that streams to per-turn files under ~/.codex-server/turns/<turn-id>/ — Bash's 2-min timeout never applies. Prefer this over codex-cli for streaming UX, multi-turn dialogues that need live state, structured (JSON schema) output, attaching images, or programmatic event handling. Use codex-cli only for one-shot batch invocations that pre-capture to an -o file. Trigger patterns (match any variation): codex

Turn a feature idea into an execution-ready document set through an adversarial PRD "council": grill the user for requirements, draft a PRD, then debate it with OpenAI Codex round-by-round until BOTH sides approve, and finally emit a Technical PRD (one overall summary + one per UseCase) plus a task list with agent assignments, dependencies, and acceptance criteria — designed so a PdM-role agent could later distribute the work to specialist agents. Document-generation only (no agent execution). Writes local files under docs/prd/<slug>/; never publishes to external trackers. Uses codex-server (the user's ChatGPT subscription) for the debate loop. Use when the user wants a PRD reviewed/approved by Codex, a "PRD council", a Technical PRD for task distribution, or a task breakdown from a PRD. Trigger patterns (match any variation): prd-council / prd council / PRD council / PRD + {Codex, GPT, ChatGPT} + {議論, 相互承認, レビュー, 承認, debate, review, approve} / CodexとPRDを議論 / CodexとPRDを相互承認 / PRDをCodexと詰める / technical prd / t

End-to-end TDD development workflow with multi-agent team review and skill-based quality gates. Plans, discusses, and implements features using strict RED-GREEN-REFACTOR test-driven development. Handles worktree-based branch isolation (from latest remote), PRD and technical requirements creation, plan-mode exploration, embedded team discussion at every deliverable phase, skill-agent quality gates (review-local, orch-qa, scenario-gen, pm-review, e2e-test), mandatory test case design via test-scenario skill, TDD implementation, fresh agent implementation review, conventional commits, optional PR creation with self-pr-review and address-pr-comments. Use when the user asks to implement a feature, fix a bug, refactor code, add tests, or any multi-step development task. Triggers on "implement X", "build Y", "add feature Z", "fix bug", "refactor", "implement with TDD", "add feature with tests", "build X following TDD", or when the user invokes /dev-workflow.

Generate AGENTS.md (cross-tool agent context — Codex CLI, Aider, Cursor, etc.) and wire CLAUDE.md as a symlink so Claude Code reads the same file. Delegates content generation to Claude's built-in /init when needed, then renames + symlinks. Idempotent. Triggers — /agents-init, "generate agents.md", "create agents.md", "share claude and codex context", "set up agents file", "init agents", "agents-init", "cross-tool agent context", "wire claude and agents".

Disciplined diagnosis loop for hard bugs and performance regressions. Reproduce → minimise → hypothesise → instrument → fix → regression-test. Use when user says "diagnose this" / "debug this", reports a bug, says something is broken/throwing/failing, or describes a performance regression.

Compact the current conversation into a handoff document for another agent to pick up.

Find deepening opportunities in a codebase, informed by the domain language in CONTEXT.md and the decisions in docs/adr/. Use when the user wants to improve architecture, find refactoring opportunities, consolidate tightly-coupled modules, or make a codebase more testable and AI-navigable.

Build a throwaway prototype to flesh out a design before committing to it. Routes between two branches — a runnable terminal app for state/business-logic questions, or several radically different UI variations toggleable from one route. Use when the user wants to prototype, sanity-check a data model or state machine, mock up a UI, explore design options, or says "prototype this", "let me play with it", "try a few designs".

Test-driven development with red-green-refactor loop. Use when user wants to build features or fix bugs using TDD, mentions "red-green-refactor", wants integration tests, or asks for test-first development.

Turn the current conversation context into a PRD and publish it to the project issue tracker. Use when user wants to create a PRD from the current context.

Triage issues through a state machine driven by triage roles. Use when user wants to create an issue, triage issues, review incoming bugs or feature requests, prepare issues for an AFK agent, or manage issue workflow.

Tell the agent to zoom out and give broader context or a higher-level perspective. Use when you're unfamiliar with a section of code or need to understand how it fits into the bigger picture.

Operate Notion from the terminal via the official Notion REST API on TypeScript + deno — search workspace, get/create/update/archive pages, query databases with filters and sorts, list/append/delete blocks, list and retrieve users, list databases, retrieve page comments. Two-tool stack: mise + deno (no node_modules, no pnpm, no bash). Token stays outside the agent: existence-only check, scoped `--allow-env=NOTION_TOKEN,NOTION_API_KEY,NOTION_TOKEN_FILE` and `--allow-net=api.notion.com` flags only, no `.env` files, never echo or print the token. Token resolved in order: NOTION_TOKEN → NOTION_API_KEY → contents of the file at NOTION_TOKEN_FILE (suits agenix / sops-nix / 1Password CLI mounts). Walks first-time users through creating a Notion Internal Integration and sharing pages with it. Trigger patterns (match any variation): notion / Notion / notion-cli / notion cli / notion api / notion page / notion pages / notion database / notion db / notion search / search notion / find in notion / create notion page / ne

Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the decision tree. Use when user wants to stress-test a plan, get grilled on their design, or mentions "grill me".

Grilling session that challenges your plan against the existing domain model, sharpens terminology, and updates documentation (CONTEXT.md, ADRs) inline as decisions crystallise. Use when user wants to stress-test a plan against their project's language and documented decisions.

Autonomously fetch and apply AI reviewer comments on a GitHub PR — creating a draft PR and requesting reviewers if needed. Default mode processes Copilot/Gemini bot comments without human intervention. Use when the user asks to address PR comments, fix review comments, handle PR feedback, respond to reviewer comments, apply review suggestions, act on PR feedback, get AI review, run copilot review, auto-review, address review, fix PR feedback.

Run frontend E2E tests with video evidence. Generates a Playwright test from the scenario CSV and runs it via CLI with video recording. Supports Chrome profile reuse (--chrome-profile) for testing with existing cookies, sessions, and extensions via launchPersistentContext. Generates a test report with video, screenshots, and trace. Does NOT modify application code — only creates documents (markdown, CSV, test scripts). Supports --non-interactive for fork / CI use (fails fast on scenario gaps instead of asking). Use when the user asks to run E2E tests, verify frontend behavior, do end-to-end testing, check UI flows, or test a web app. Trigger phrases include "e2e", "E2E test", "end-to-end test", "e2e testing", "frontend test", "UI test", "playwright test", "browser test", "verify the UI", "test this page".

Senior QA/QC engineer that evaluates existing codebases for test quality. Runs existing tests, diagnoses failures, identifies missing test coverage, and writes missing tests. Framework-agnostic — works with any language and test runner. Use when the user asks to run tests, diagnose failures, do test gap analysis, QA, quality assurance, test coverage analysis, find missing tests, test audit, quality check, quality report, fix failing tests, evaluate test quality, or assess test health of a codebase. Trigger phrases include "run tests", "QA", "quality assurance", "test audit", "test coverage", "missing tests", "gap analysis", "quality report", "fix failing tests", "diagnose test failures", "quality check", "evaluate test quality", "test health".

Run Playwright CLI commands via Bash for test execution, code generation, browser management, reporting, and debugging. Covers the full `npx playwright` surface: test, codegen, install, show-report, show-trace, screenshot, pdf, open, clear-cache, merge-reports, and more. Use when the user asks to run playwright tests, generate test code, install browsers, view a test report or trace, take a page screenshot or PDF, open a URL in a specific browser, clear the playwright cache, merge shard reports, debug tests, or use playwright UI mode. Trigger phrases include "playwright test", "run playwright", "npx playwright", "codegen", "install browsers", "show report", "show trace", "playwright screenshot", "playwright pdf", "open in chromium", "open in firefox", "open in webkit", "clear playwright cache", "merge reports", "shard tests", "playwright debug", "playwright ui mode", "record test", "run e2e tests with playwright cli", "playwright headed", "playwright retries", "update snapshots", "playwright grep", "list play

Guided workflow for recording browser interactions with Playwright codegen and transforming the raw output into production-quality test suites. Handles the full lifecycle: planning the recording session, launching codegen, reviewing captured actions, enhancing with assertions and page objects, organizing test files, scaffolding config, and running until green. Use when the user wants to create new Playwright tests from browser recordings, turn codegen output into structured specs, or bootstrap a test suite for a web application. Trigger phrases include "record test", "codegen to test", "generate playwright test", "create test from recording", "record and generate test", "playwright codegen workflow", "turn recording into test", "bootstrap test suite", "record browser test", "generate spec from codegen", "create e2e test from recording".

Review local git changes from a PMBOK-based product management perspective. Analyzes changes against 7 PMBOK knowledge areas: Scope, Risk, Stakeholder, Quality, Integration, Schedule, and Resource management. Produces a structured report identifying project-level impacts and risks. The PMBOK analyst role itself lives in the `pm-review:pm-review` subagent; this skill is a thin orchestrator that gathers the change set and project context before delegating. Use when the user asks for a PM review, product management review, PMBOK analysis, project impact assessment, scope review, risk review, or stakeholder impact analysis of local changes. Trigger phrases include "pm review", "product management review", "PMBOK review", "project impact", "scope review", "risk review", "stakeholder impact", "pm-review", "review from PM perspective", "project management review".

Review a teammate's pull request. The reviewer role (multi-lens analysis, severity calibration, suggestion writing) lives in the `pr-review:pr-review` subagent; this skill is the orchestrator — it resolves the PR, fetches diff and changed files, delegates analysis to the subagent, stages the resulting comments for user approval, then submits a batch review via the GitHub API. NOT for self-review of your own PR (use `self-pr-review` instead). Trigger phrases include "review PR #123", "code review PR", "peer review", "check this PR", "give PR feedback", "review pull request", "review someone's PR".

Review local git changes from 8 expert perspectives by spawning the matching reviewer subagents in parallel. Produces a consolidated report with Critical/Important/Nice-to-have severity. Lightweight pre-commit or pre-push quality gate — no PR or branch push required. Use when the user asks to review local changes, check changes before committing, get a team review of working tree changes, or run a pre-commit review. Trigger phrases include "review local", "review my changes", "review local changes", "pre-commit review", "review before commit", "review before push", "team review my changes", "check my changes", "review working tree", "local code review", "review diff", "review my diff".

Generate test scenarios from git branch changes. Compares current branch to main/master, identifies changed files, classifies them by type (frontend, API, backend, config), and generates test scenarios covering those changes. Produces a Markdown report with embedded screenshots for UI changes and CSV files compatible with the e2e-test skill for automated execution. Captures live browser screenshots via Playwright MCP for frontend changes. Use when the user asks to generate test scenarios, create test cases from changes, analyze branch for testing, generate test plan from diff, create e2e scenarios from code changes, or build test coverage for a branch. Trigger phrases include "generate test scenarios", "test scenarios for this branch", "create test cases from changes", "what should I test", "generate test plan", "test coverage for my changes", "scenario generator", "test scenario generator", "generate scenarios from diff", "create e2e scenarios", "test scenario from branch".

Install a PreCompact hook that auto-generates .agent/local/HANDOVER.md before context compaction, plus a SessionStart hook that restores context from a previous handover. Run once to install, then it works automatically forever. Use when the user says "handover", "setup handover", "install handover hook", or wants seamless context transfer between sessions.

Operate Slack from the terminal via `slack-cli` (@urugus/slack-cli). Send messages, check unreads, search history, upload files, manage reactions/pins, schedule messages, and manage channels — all through Bash commands. Prefer this skill over `mcp__slack__*` MCP tools when you need: scheduling, unreads, search with pagination, reactions/pins, file uploads, multi-workspace profiles, canvas operations, or jq pipelines for filtered output. Trigger phrases include "slack-cli", "slack cli", "send slack message", "check slack unreads", "slack history", "search slack", "schedule slack message", "slack channel", "upload to slack", "slack reactions", "slack pins", "slack canvas", "slack users", "slack unread".

Generate a persistent PROJECT-KNOWLEDGE.md profile for any codebase that gives subagents instant deep domain expertise. Analyzes project structure, architecture, code conventions, tech stack, data flow, and domain concepts. The profile is designed to be loaded into Agent tool prompts so subagents can answer deep codebase questions like an expert teammate. Use when the user asks to profile a project, generate project knowledge, create a codebase profile, build a context pack, analyze a codebase for subagents, or says "profile this project". Trigger phrases include "profile project", "generate project knowledge", "codebase profile", "subagent context", "knowledge profile", "project profile", "profile this codebase", "build project knowledge", "create context pack", "generate codebase knowledge", "index this project", "make subagent", "generate subagent", "deep knowledge", "project expert".

Senior-level .NET development guidance for architecture, code review, scaffolding, EF Core, ASP.NET Core, Blazor, and MAUI targeting .NET 8 and .NET 10. Provides opinionated best practices, modern C# idioms, security hardening, performance optimization, and testing strategies at a staff/senior engineer level. Trigger patterns (match any variation): .NET / dotnet / dot net / .net / C# / csharp / c sharp / c-sharp / ASP.NET / aspnet / asp.net core / Entity Framework / EF Core / ef core / Blazor / MAUI / maui / NuGet / nuget / Clean Architecture + .NET / CQRS + .NET / DDD + .NET / {review, scaffold, create, design, architect, optimize, migrate} + {C#, .NET, dotnet, csharp} / "dotnet new" / "dotnet build" / "dotnet test" / "dotnet publish" / ".NET best practices" / "C# patterns" / "senior .NET" / ".NET architecture"

英語の技術ドキュメント (Hugging Face docs / GitHub README / arXiv 論文 / 技術ブログ等) を、日本人が読んで AI 生成と気づかない自然な日本語の解説サイトに変換します。フェッチ → 構成 → HTML/CSS/JS 生成 → 多角レビュー → 修正 → ローカル配信、まで一気通貫。日本語自然さチェックは `@agent-naturalize-ja:naturalize-ja` subagent に委譲 (辞書本体・grep・質的レビューはそちら側に集約)。技術正確性は codex-cli、UX/レスポンシブは汎用 Agent が担当。Triggers: 「英語ドキュメントを日本語で解説するサイトを作って」「Japanese explainer for [URL]」「英語の技術ドキュメントをやさしく日本語にして」「Convert this English doc to a Japanese site」「AI っぽくない日本語で技術解説」「Hugging Face docs を日本語化」「arXiv を日本語で噛み砕いて」「en-to-ja explainer」「/en-to-ja-explainer」

任意の日本語テキスト (ファイル / 貼り付けテキスト / ディレクトリ) を読み、AI 生成と気づかれる「不自然な日本語表現」を検出・置換します。プレ AI 時代の技術ブログ調へ整えるための禁止フレーズ辞書 (15 カテゴリ。実フィードバックと公開研究を統合) と、grep + 質的レビューの二段監査を内蔵 (subagent をさらに spawn せず、自分のコンテキスト内で完結)。en-to-ja-explainer など他スキルから委譲して使うことも想定。Triggers: 「AI っぽい日本語をレビューして」「この文章を自然な日本語にして」「ChatGPT 臭を抜きたい」「日本語の AI 臭をチェック」「ナチュラルな日本語に直して」「AI 生成バレしないように直して」「naturalize Japanese」「naturalize-ja」「/naturalize-ja」

Retrofit the OIAE self-improvement loop to existing skills and analyze feedback to propose evidence-based amendments. Adds Retrospective, Feedback Check, and version tracking to skills that lack them. Reads feedback/log.md to identify recurring issues and propose targeted skill improvements. Use when the user asks to improve a skill, retrofit feedback to a skill, add the improvement loop, fix a skill based on feedback, or analyze skill performance. Trigger phrases include "improve skill", "retrofit skill", "add feedback loop", "skill keeps failing", "fix skill", "analyze skill feedback", "skill performance", "add retrospective", "skill improvement".

Switch the user's Claude Code session back to classifier-guided auto mode by clearing the bypass-toggle flag. Use after approving a plan from the iPhone Claude app (or anywhere bypass-toggle is on) when you want subsequent tool calls to go through the auto-mode classifier instead of being auto-allowed. Companion to bypass-toggle. Only affects Claude Code (Cursor and other agents are not touched). Trigger phrases include "/auto-mode", "auto on", "auto モード", "auto に戻して", "オートに戻して", "オート", "classifier に戻して", "クラシファイアに戻して", "bypass やめて", "bypass 解除", "switch to auto", "back to auto", "leave bypass mode".

Toggle effective bypassPermissions for the user's Claude Code sessions via a PreToolUse hook + flag file. Use when the user wants to skip further permission prompts after approving a plan — especially when the approval came from the iPhone Claude app (Remote Control) and they don't want to attach a TTY to Shift+Tab. ONLY affects Claude Code (Cursor and other agents read their own configs). Trigger phrases include "/bypass-toggle", "bypass on", "bypass off", "bypass status", "stop asking for permission", "skip permission prompts", "no more permission asks", "バイパス on", "バイパス off", "バイパスにして", "バイパス止めて", "プロンプト止めて", "もう確認しないで", "承認スキップ".

Run OpenAI Codex CLI from the terminal for batch / one-shot `codex exec` invocations with output captured to an `-o` file. Use for fire-and-forget runs, scripted CI, code review batches (`codex review`), and resuming a specific session by id (`codex exec resume`). Always use this skill instead of `mcp__codex__codex` and `mcp__codex__codex-reply` MCP tools — CLI is faster and more capable. For streaming UX, multi-turn dialogues that need live state, structured (JSON schema) output, attaching images, or programmatic event handling → **use codex-server instead**. codex-server uses the user's ChatGPT subscription via Codex App Server and decoupled async invocation that bypasses the Bash 2-minute timeout entirely. Trigger patterns (batch-only — chat/streaming triggers belong to codex-server): codex-cli / codex CLI / codex exec / codex review / codex resume / codex fork / batch + {codex, gpt} / one-shot + {codex, gpt} / non-interactive + codex / codex {to a file, -o, output file, captured output} / CI + {codex, gpt

Sync a session-recap directory (summary.md + note.md + meta.json) to Notion under the explicit "My Agent Notes / Session Recaps" database via the notion-cli skill. The Notion target is fixed by design — the value of this skill is being explicit about its destination, not configurable. Appends to the shared session-recap manifest on success and retries previously-failed uploads on next invocation. Cleanup and retention are owned by session-recap, not this skill. Standalone — pass a recap dir path, or invoke in retry-only mode. Auto-invoked by session-recap via the chain in its ~/.claude/session-recaps/.recap-config.json. Use when the user says "recap to notion" / "Notionに recap を上げて" / "recap upload" / "/recap-to-notion" / "recap retry" / "再アップロード" / "Notion同期".

Read a Claude Code session's transcript and produce two markdown artifacts plus a meta.json: a scannable Summary (TL;DR, key terms, files, decisions, open threads) and a complementary Detailed Note (full glossary, chronological flow, command-by-command annotations, pitfalls). The output language follows the watched session's dominant language (>= 50% Japanese characters → Japanese, otherwise English; overridable via --language). Standalone — pass a session ID, or omit to pick from recent sessions in the current directory. After generation, dispatches a configurable chain of follow-up skills (default ["recap-to-notion"]) read from ~/.claude/session-recaps/.recap-config.json. Owns the manifest schema and runs cleanup of confirmed-uploaded local recap dirs past retention. Read-only on the transcript itself. Use when the user says "session recap" / "セッションのまとめ" / "セッションサマリ" / "学びをまとめて" / "用語集を作って" / "summarize the session" / "recap session" / "session note" / "セッションのノート" / "/session-recap" / "セッション解説まとめ" / "recap掃