empirical-prompt-tuning
agent 向けテキスト指示（skill / slash command / task プロンプト / AGENTS.md 節 / コード生成プロンプト）を、バイアスを排した実行者に動かしてもらい、両面（実行者の自己申告 + 指示側メトリクス）で評価して反復改善する手法。改善が頭打ちになるまで回す。プロンプトや skill を新規作成・大幅改訂した直後、またはエージェントの挙動が期待通りにならない原因を指示側の曖昧さに求めたいときに使う。
design-team
規模の大きいタスクの設計段階で、Architect (構築役) と Devil's advocate (反対役) を独立 subagent として並行起動し、指示から抜けた視点/選択肢/死角を炙り出してから設計を確定する。互いの出力を見せずアンカリングを防ぎ、本体が突合して統合設計を作り 2nd round で再レビューさせる。「設計チーム」「死角を洗い出して設計」「devil's advocate」で起動。方針自体の go/no-go は premise-questioning、単一視点の設計手順は arch を使う。
deep-review
High-risk or multi-file code review using security, performance, and maintainability perspectives. Use for 'deep review', large diffs, auth/secrets/DB/API, or release gates. Front-load this description for Codex implicit matching; explicit invocation via $deep-review always works.
consult
実装中に判断が必要になった時、技術選定・設計相談が必要な時に使用。相談テンプレートで構造化された問題提示を強制。「どっちがいい」「どう設計するか」「迷ってる」「技術選定」「比較して」もトリガー。
commit-msg
Generate Conventional Commits message proposals from staged diff. Never run git commit. Front-load this description for Codex implicit matching; explicit invocation via $commit-msg always works.
bug-hunt
Gitのstagedまたはunstaged差分には必ず1件以上の実害バグがある前提で、差分・周辺コード・契約・テスト・実行結果を多角的に調査し、再現可能なバグを最低1件以上特定する。"バグの間違い探し"、"差分からバグを見つけて"、"staged/unstagedを疑って"、"必ず1件バグがある"、"変更コードを壊しに行って"で発動。単なるスタイル指摘や一般的コードレビューには使わない。
test
Add or improve tests for existing behavior or new specification. Use for unit/integration test additions. Do not use for CI-only changes. Front-load this description for Codex implicit matching; explicit invocation via $test always works.
test-coverage-guard
既存テストの信頼性を検証し、偽陽性を検出・排除するガードレール。テストがGREENになった後に発動する。「このテスト信頼できる?」「偽陽性」「カバレッジ稼ぎ検出」で発動。
tdd
機能実装やバグ修正でテストを書く・変更する作業に使用。RED-GREEN-REFACTORサイクルを適用する。「TDD」「テストから書く」「テスト駆動」「テスト先に」で発動。
semantic-generation
|
security-review
Security-only review for auth, authorization, secrets, input validation, injection, XSS, SSRF, path traversal, unsafe deserialization, CORS/CSRF. Front-load this description for Codex implicit matching; explicit invocation via $security-review always works.
rules-required
Select and apply only the markdown rules relevant to the current task. Use when task or project instructions require rules, applicable rules are unclear, or a rules guard requests reactivation. Front-load this description for Codex implicit matching; explicit invocation via $rules-required always works.
rules-compliance-review
Use after code changes, before final answers, or whenever rules/*.md compliance is uncertain. Runs deterministic rules-enforce first, then performs a semantic audit of codex/rules/*.md against the current diff. Do not use as a replacement for hooks; use it as the human/model-level second pass.
review
Code review for a diff, file, PR, or implementation result. Use for 'review', 'check', 'inspect'. Escalate to deep-review for high-risk diffs. Front-load this description for Codex implicit matching; explicit invocation via $review always works.
refactoring
振る舞いを変えずにコード構造を改善する際に使用。「リファクタ」「整理して」「きれいにして」「重複を消す」「関数を分割」で発動。
current-location-weather
Get current weather conditions for a location. Use this skill when users ask about weather, temperature, current conditions, forecasts, or any weather-related queries like "What's the weather?", "Is it raining?", "How hot is it outside?", "What's the temperature?", or "Should I bring an umbrella?". Supports automatic location detection via IP or specific location queries.
ollama-web-search
Call Ollama's cloud web_search/web_fetch API, and handle the API key without it ever touching chat or session transcripts.
copy
Professional copywriter for SaaS and startups. Expert in landing page copy, positioning, messaging, conversion optimization, and voice-of-customer research. Use when writing compelling copy for SaaS products, landing pages, marketing materials, or when you need help with product positioning and messaging strategy.
kinde
>-
landing-page-breakdown
Comprehensive landing page design analysis for extracting typography, color palette, spacing systems, visual elements, and conversion optimization insights. Use when a user provides a landing page URL for design analysis, wants to understand what makes a page effective, needs to extract design specifications, or wants to learn from high-converting landing pages.
use-browser
>-
things
>-
sys-arch
Design production-grade software systems with expert knowledge of architecture patterns, distributed systems, cloud platforms, and operational excellence. Use this skill when architecting complex systems, evaluating technology choices, designing scalable infrastructure, or making critical architectural decisions requiring trade-off analysis.
sqlite
>-
go-errors
>-
go-dev
>-
go-concurrency
>-
go-sql
>-
go-logging
>-
front-dev
>-
go-lint
>-
go-http
>-
apple-dev
Comprehensive macOS and iOS development expertise covering Swift best practices, SwiftUI design patterns, Human Interface Guidelines, Apple frameworks, and performance optimization. Use when developing native Apple applications, implementing SwiftUI interfaces, working with Apple frameworks (Foundation, UIKit, AppKit, Core Data, CloudKit, etc.), optimizing app performance, following HIG principles, or writing production-quality Swift code for iOS, macOS, watchOS, or tvOS platforms.
multi-agent-config
Manage portable AI coding-agent configuration for OpenAI Codex, Claude Code, and Cursor, with explicit fallbacks for unsupported mappings. Use when initializing multi-agent projects, syncing Agent Skills and project rules, translating supported MCP formats, migrating between agent platforms, or maintaining shared configuration for teams using different coding agents.
go-cli
>-
go-style
>-
go-testing
>-
moonshot-classify-task
Classifies a user request into task types (feature, modification, bugfix, refactor) and extracts intent keywords. Use at the start of PM analysis.
explain-diff-html
Use when the user asks for a rich explanation of a code change, diff, branch, or PR. Produces HTML output.
doc-auto-sync
Detects code changes, auto-updates related docs (PROJECT.md, README, CHANGELOG, generated docs), and bootstraps project documentation structure.
moon-relay-kernel
Explicit-only entrypoint for Moon Relay Kernel task routing and adaptive workflow execution. Use only when the user explicitly names `moon-relay-kernel`, invokes `$moon-relay-kernel`, or explicitly asks to use the Kernel skill or Kernel mode for the current task. Do not infer activation from installed availability, AGENTS.md, track markers, repository context, or a task that merely concerns Kernel.
adr-c4-writer
Internal Moonshot Architecture stage for writing C4 model artifacts and ADRs from accepted architecture decisions.
architecture-artifacts
Produce architecture artifacts and a provenance-bound Task Contract Seed before Kernel execution.
architecture-gate-reviewer
Internal Moonshot Architecture stage for reviewing whether an architecture package is ready for implementation handoff.
architecture-option-generator
Internal Moonshot Architecture stage for generating architecture options from ASRs, constraints, and current evidence.
architecture-tradeoff-reviewer
Internal Moonshot Architecture stage for comparing architecture options and selecting a justified recommendation.
asr-extractor
Internal Moonshot Architecture stage for extracting architecturally significant requirements and quality attribute scenarios from normalized requirements.
assumption-ledger
Keep product-definition work moving by recording ambiguity as assumptions or blockers instead of stopping the workflow.
audit
Run a technical UI quality audit across accessibility, performance, responsive behavior, theming, and design anti-patterns.
browser-verifier
Runs runtime/browser verification for web projects using URL health checks and optional E2E commands.
Page 1546 of 1700 · 84990 results
