Alt

Drive Alt changes test-first in outside-in order — E2E (Playwright/Hurl) → CDC (Pact) → Unit (RED-GREEN-REFACTOR) — and close with a local CI parity sweep (format/lint/type/security) for every touched microservice. Use when implementing a feature, fixing a bug, or refactoring, when deciding what test to write first, or when the user says "TDDで". Also use before declaring work complete, since "tests pass" is not "CI will pass".

Alt の Architecture Decision Record を日本語で `docs/ADR/NNNNNN.md` に書き起こす。番号採番、frontmatter（title/date/status/tags/affected_services/aliases/supersedes）、Context・Decision・Consequences の書き分け、wikilink 形式、OSS 公開向けの情報衛生を扱う。実装が一段落して決定を記録するときに使う。ユーザが「ADR書いて」「ADRにまとめて」「ADRに記録して」「実装が終わったのでドキュメントに」と言ったとき、または設計上の判断を伴う変更を終えたときに使う。

Interrogate a plan against the project's documented decisions and domain language — challenging fuzzy terms, cross-referencing the code, and surfacing where the plan contradicts an existing ADR or canonical contract. Use when the user wants a plan stress-tested against what the project has already decided, rather than in the abstract. Prefer plain grill-me when there is no documented model to test against, and plan-context-loader when the goal is to gather vault context rather than be questioned.

Hunt for deepening opportunities — shallow modules whose interface costs as much to learn as their implementation saves — and present them as a visual before/after report, informed by the decisions already recorded in docs/ADR/. Use when the user wants to improve architecture, find refactoring opportunities, consolidate tightly-coupled modules, reduce the cost of understanding a subsystem, or make code more testable. Use clean-architecture instead when the question is whether a specific change violates Alt's layer rules.

設計・計画タスク時に、Obsidian vault から関連 ADR・contract・review・runbook を 必要最小限だけ集めて計画の精度を上げる。 次のときに使う: - 「計画を立てて」「設計して」「プランを作って」 - ADR や過去の決定を確認したい - Knowledge Trail / Knowledge Home や reproject-safe / immutable 設計が絡む

Alt の Go 規約を適用する。エラーラップ、context 第一引数、slog、テーブル駆動テスト、http.Server の 4 タイムアウト、Redis Streams の XACK/XAUTOCLAIM 対、fail-fast 設定を扱う。Go のコードを書く・直す・レビューするときに使う。ユーザが「Go」や規約名に触れなくても、Go サービス（alt-backend, auth-hub, pre-processor, search-indexer, mq-hub, rag-orchestrator, altctl）の実装・修正に入るなら使う。

Alt の Python 3.14+ 規約を適用する。型ヒント必須と Pyrefly、具体例外と原因チェーン、Pydantic/frozen dataclass の境界保護、asyncio と同期推論の分離、spawn プロセスプール、起動時 fail-closed を扱う。Python のコードを書く・直す・レビューするときに使う。ユーザが「Python」や規約名に触れなくても、Python サービス（news-creator, tag-generator, metrics, recap-subworker, recap-evaluator, acolyte-orchestrator）の実装・修正に入るなら使う。

Alt の Rust Edition 2024 規約を適用する。thiserror でのドメインエラー、借用優先、tokio + tracing、配線失敗を握り潰さない、flush タスクの JoinHandle 保持、char boundary 安全な切り詰め、リトライ境界でのエラー型保持、libtorch 常駐時の jemalloc を扱う。Rust のコードを書く・直す・レビューするときに使う。ユーザが「Rust」や規約名に触れなくても、Rust サービス（rask-log-aggregator, rask-log-forwarder, recap-worker）の実装・修正に入るなら使う。

Alt の Svelte 5 / SvelteKit 規約を適用する。Runes（$state/$derived/$effect）、$props()、$effect をコールスタック越えで自己再発火させない、snippet でのコンポーネント合成、load 関数でのデータ取得、{@html} の DOMPurify 必須、keyed each の重複キー、非同期コールバックの stale-response guard を扱う。Svelte コンポーネントや SvelteKit のルート・load 関数を書く・直すときに使う。ユーザが「Svelte」に触れなくても alt-frontend-sv の UI 実装に入るなら使う。

Alt の TypeScript 規約を適用する。strict + noUncheckedIndexedAccess、境界では unknown、型ガード優先、satisfies、判別共用体の網羅性、Zod での境界バリデーション、起動時 env 検証の fail-fast、redirect パラメータの URL パース検証、connect-es エラーの numeric enum、wire スキーマの一本化を扱う。TypeScript のコードを書く・直す・レビューするときに使う。ユーザが「TypeScript」に触れなくても alt-frontend-sv / auth-token-manager / alt-perf の実装に入るなら使う。

Alt の Clean Architecture レイヤ規約（Handler → Usecase → Port → Gateway → Driver）を適用し、層の逆依存・層越境・誤った層への実装を検出する。サービスのコードを新規実装・リファクタ・レビューするとき、どの層に置くべきか判断するとき、handler/usecase/gateway/driver/port ディレクトリ配下を編集するときに使う。ユーザが「Clean Architecture」や「層」に触れなくても、ハンドラに fetch や SQL を書き足すような層をまたぐ変更に入るなら使う。

Generate multiple radically different interface designs for a module using parallel sub-agents. Use when user wants to design an API, explore interface options, compare module shapes, or mentions "design it twice".

Alt スタックを Docker Compose で操作する。サービスの起動・停止・再ビルド・ログ確認・health check、18 個の stack ファイルに分割された構成のどこにどのサービスが居るかの引き当て、起動しない/変更が反映されない/ポート衝突の切り分けを扱う。ユーザが「起動して」「ログ見せて」「再ビルドして」「立ち上がらない」のようにコンテナ操作を求めたとき、または compose 配下の yaml を編集するときに使う。

Work a GitHub issue end to end in Alt — read the issue, locate the affected code, drive the fix test-first, and commit it. Use when the user points at an issue by number or URL and wants it fixed, or says "fix issue 42" / "この issue 直して".

Interview the user relentlessly about a plan or design, one question at a time, walking every branch of the decision tree until the open choices are actually resolved rather than deferred. Use when the user wants to stress-test a plan, be grilled on a design, think something through out loud, or says "grill me". Use grill-with-docs instead when the plan should also be checked against the project's recorded decisions and terminology.

Audits code, schema, and migrations for violations of immutable data-model invariants: append-first event log, reproject-safe projectors, disposable read models, versioned artifacts, merge-safe upserts, and no business-fact `time.Now()`. Use when changing migrations, projectors, event handlers, append-only stores, or any read model derived from events; or when the user mentions "イミュータブル", "event sourcing", "projector", "reproject", "append-only", "projection", "read model", or kawasima's resource/event modeling. Applies to any append-only subsystem, not one named service. Not for stateless pure functions, TTL caches, or plain CRUD tables with no event log behind them.

Run a conversational QA session — the user describes bugs one at a time, and each becomes a durable, user-focused GitHub issue written in the project's own domain language, broken into parallelizable sub-issues when it spans separable concerns. Use when the user wants to report bugs, walk through problems they hit while using the product, file issues conversationally, or mentions a QA session. Use log-seeker instead when the goal is to diagnose a failure rather than record it.

Re-read Alt's authoritative project context — CLAUDE.md, the path-scoped rules, and the wiki entry point — and treat it as binding for the rest of the session. Invoke manually after a long session, after context compaction, or when the working assumptions have drifted from what the repo actually says.

Review a GitHub pull request against the failure modes specific to Alt — Clean Architecture layer direction, missing CDC coverage for a crossed service boundary, unwired components that unit tests still pass, and immutable-data-model violations. Use when the user asks to review a PR by number or URL. The bundled /review skill covers the generic code-quality pass; this one adds what Alt has been burned by.

Audits code for security vulnerabilities using OWASP Top 10:2025, OWASP ASVS 5.0, the OWASP Secure Code Review Cheat Sheet, and (for AI/agentic code paths) the OWASP Top 10 for Agentic Applications 2026 as authoritative sources. Produces a structured findings report with severity, OWASP/CWE mapping, evidence, and concrete remediation. Use when the user asks for "security review", "脆弱性チェック", "セキュリティ監査", "OWASP レビュー", threat-modeling a diff/module, or when a change touches authentication, authorization, crypto, input validation, deserialization, file upload, SQL/NoSQL/command building, secrets, logging, error handling, dependency updates, or LLM/agent/RAG code paths.

Run one Alt service's test suite, detecting the language and the right runner from the files in its directory. Use when the user asks to test, re-test, or check a specific service by name. For the full pre-handoff sweep across every touched service — formatters, linters, type checkers, security scanners — use tdd-workflow Phase 5 instead, since a passing test suite is not a passing CI run.

Web research skill that prioritizes official documentation while covering broad literature. Searches, fetches, and synthesizes findings into a structured actionable report. Use when: - 「調べて」「リサーチして」「公式ドキュメント確認して」 - "research X", "look up", "find docs for" - Evaluating libraries, frameworks, or approaches - Investigating errors, migration guides, or breaking changes - Needing up-to-date information beyond training data

Investigates Alt's container logs and databases read-only to diagnose incidents. Reads `docker compose logs`, the ClickHouse `rask_logs` store (otel_logs / http_logs / otel_traces / otel_http_requests / otel_error_logs / sli_metrics), and the PostgreSQL databases (alt-db, pre-processor-db, recap-db, rag-db, acolyte-db, kratos-db, knowledge-sovereign-db), correlates log errors with trace_id and DB state, and returns an evidence-backed root-cause report. Use when the user says 「ログ調べて」「DB見て」 「コンテナログ精査して」「なんで〇〇がエラー/落ちてる/遅い」, "check the logs", "why is X failing/crashing/slow", "investigate this incident", or right after an outage or alert.