@jstn-sdk/rcs 0.1.0

package/docs/openclaw-integration.ru.md

@@ -0,0 +1,44 @@
+# Руководство по интеграции OpenClaw (локализованный тюнинг промптов)
+
+> [← Back to Docs Home](./index.html) · [Integrations Landing](./integrations.html)
+
+**Language Switcher:** [English](./openclaw-integration.md) | [한국어](./openclaw-integration.ko.md) | [日本語](./openclaw-integration.ja.md) | [简体中文](./openclaw-integration.zh.md) | [繁體中文](./openclaw-integration.zh-TW.md) | [Tiếng Việt](./openclaw-integration.vi.md) | [Español](./openclaw-integration.es.md) | [Português](./openclaw-integration.pt.md) | [Русский](./openclaw-integration.ru.md) | [Türkçe](./openclaw-integration.tr.md) | [Deutsch](./openclaw-integration.de.md) | [Français](./openclaw-integration.fr.md) | [Italiano](./openclaw-integration.it.md) | [Українська](./openclaw-integration.uk.md)
+
+
+Эта страница локализует раздел **“Prompt tuning guide (concise + context-aware)”** из основной англоязычной документации.
+
+Полное руководство по интеграции (gateway, hooks, проверка) см. в [English guide](./openclaw-integration.md).
+
+## Тюнинг промптов (кратко + с учетом контекста)
+
+## Где редактировать шаблоны промптов
+
+- `notifications.openclaw.hooks["session-start"].instruction`
+- `notifications.openclaw.hooks["session-idle"].instruction`
+- `notifications.openclaw.hooks["ask-user-question"].instruction`
+- `notifications.openclaw.hooks["stop"].instruction`
+- `notifications.openclaw.hooks["session-end"].instruction`
+
+## Рекомендуемые контекстные токены
+
+- Всегда включать: `{{sessionId}}`, `{{tmuxSession}}`
+- По событию: `{{projectName}}`, `{{question}}`, `{{reason}}`
+
+## Стратегия детализации (verbosity)
+
+- `minimal`: очень короткие уведомления
+- `session`: сжатый операционный контекст (рекомендуется)
+- `verbose`: расширенный контекст статуса/действий/рисков
+
+## Быстрая команда обновления (jq)
+
+```bash
+CONFIG_FILE="$HOME/.codex/.rcs-config.json"
+
+jq '.notifications.verbosity = "verbose" |
+    .notifications.openclaw.hooks["session-start"].instruction = "[session-start|exec]\nproject={{projectName}} session={{sessionId}} tmux={{tmuxSession}}\n요약: 시작 맥락 1문장\n우선순위: 지금 할 일 1~2개\n주의사항: 리스크/의존성(없으면 없음)" |
+    .notifications.openclaw.hooks["session-idle"].instruction = "[session-idle|exec]\nsession={{sessionId}} tmux={{tmuxSession}}\n요약: idle 원인 1문장\n복구계획: 즉시 조치 1~2개\n의사결정: 사용자 입력 필요 여부" |
+    .notifications.openclaw.hooks["ask-user-question"].instruction = "[ask-user-question|exec]\nsession={{sessionId}} tmux={{tmuxSession}} question={{question}}\n핵심질문: 필요한 답변 1문장\n영향: 미응답 시 영향 1문장\n권장응답: 가장 빠른 답변 형태" |
+    .notifications.openclaw.hooks["stop"].instruction = "[session-stop|exec]\nsession={{sessionId}} tmux={{tmuxSession}}\n요약: 중단 사유\n현재상태: 저장/미완료 항목\n재개: 첫 액션 1개" |
+    .notifications.openclaw.hooks["session-end"].instruction = "[session-end|exec]\nproject={{projectName}} session={{sessionId}} tmux={{tmuxSession}} reason={{reason}}\n성과: 완료 결과 1~2문장\n검증: 확인/테스트 결과\n다음: 후속 액션 1~2개"'   "$CONFIG_FILE" > "$CONFIG_FILE.tmp" && mv "$CONFIG_FILE.tmp" "$CONFIG_FILE"
+```

package/docs/openclaw-integration.tr.md

@@ -0,0 +1,44 @@
+# OpenClaw Entegrasyon Kılavuzu (yerelleştirilmiş prompt ayarı)
+
+> [← Back to Docs Home](./index.html) · [Integrations Landing](./integrations.html)
+
+**Language Switcher:** [English](./openclaw-integration.md) | [한국어](./openclaw-integration.ko.md) | [日本語](./openclaw-integration.ja.md) | [简体中文](./openclaw-integration.zh.md) | [繁體中文](./openclaw-integration.zh-TW.md) | [Tiếng Việt](./openclaw-integration.vi.md) | [Español](./openclaw-integration.es.md) | [Português](./openclaw-integration.pt.md) | [Русский](./openclaw-integration.ru.md) | [Türkçe](./openclaw-integration.tr.md) | [Deutsch](./openclaw-integration.de.md) | [Français](./openclaw-integration.fr.md) | [Italiano](./openclaw-integration.it.md) | [Українська](./openclaw-integration.uk.md)
+
+
+Bu sayfa, İngilizce ana belgede yer alan **“Prompt tuning guide (concise + context-aware)”** bölümünün yerelleştirilmiş sürümüdür.
+
+Tam entegrasyon dokümantasyonu (gateway, hook, doğrulama) için [English guide](./openclaw-integration.md) sayfasına bakın.
+
+## Prompt ayarı (kısa + bağlam farkındalıklı)
+
+## Prompt şablonları nerede düzenlenir
+
+- `notifications.openclaw.hooks["session-start"].instruction`
+- `notifications.openclaw.hooks["session-idle"].instruction`
+- `notifications.openclaw.hooks["ask-user-question"].instruction`
+- `notifications.openclaw.hooks["stop"].instruction`
+- `notifications.openclaw.hooks["session-end"].instruction`
+
+## Önerilen bağlam tokenları
+
+- Her zaman: `{{sessionId}}`, `{{tmuxSession}}`
+- Olaya göre: `{{projectName}}`, `{{question}}`, `{{reason}}`
+
+## Ayrıntı (verbosity) stratejisi
+
+- `minimal`: çok kısa bildirimler
+- `session`: kısa operasyonel bağlam (önerilir)
+- `verbose`: daha fazla durum/aksiyon/risk bağlamı
+
+## Hızlı güncelleme komutu (jq)
+
+```bash
+CONFIG_FILE="$HOME/.codex/.rcs-config.json"
+
+jq '.notifications.verbosity = "verbose" |
+    .notifications.openclaw.hooks["session-start"].instruction = "[session-start|exec]\nproject={{projectName}} session={{sessionId}} tmux={{tmuxSession}}\n요약: 시작 맥락 1문장\n우선순위: 지금 할 일 1~2개\n주의사항: 리스크/의존성(없으면 없음)" |
+    .notifications.openclaw.hooks["session-idle"].instruction = "[session-idle|exec]\nsession={{sessionId}} tmux={{tmuxSession}}\n요약: idle 원인 1문장\n복구계획: 즉시 조치 1~2개\n의사결정: 사용자 입력 필요 여부" |
+    .notifications.openclaw.hooks["ask-user-question"].instruction = "[ask-user-question|exec]\nsession={{sessionId}} tmux={{tmuxSession}} question={{question}}\n핵심질문: 필요한 답변 1문장\n영향: 미응답 시 영향 1문장\n권장응답: 가장 빠른 답변 형태" |
+    .notifications.openclaw.hooks["stop"].instruction = "[session-stop|exec]\nsession={{sessionId}} tmux={{tmuxSession}}\n요약: 중단 사유\n현재상태: 저장/미완료 항목\n재개: 첫 액션 1개" |
+    .notifications.openclaw.hooks["session-end"].instruction = "[session-end|exec]\nproject={{projectName}} session={{sessionId}} tmux={{tmuxSession}} reason={{reason}}\n성과: 완료 결과 1~2문장\n검증: 확인/테스트 결과\n다음: 후속 액션 1~2개"'   "$CONFIG_FILE" > "$CONFIG_FILE.tmp" && mv "$CONFIG_FILE.tmp" "$CONFIG_FILE"
+```

package/docs/openclaw-integration.uk.md

@@ -0,0 +1,44 @@
+# Посібник з інтеграції OpenClaw (локалізоване налаштування промптів)
+
+> [← Повернутися на головну Docs](./index.html) · [Сторінка інтеграцій](./integrations.html)
+
+**Language Switcher:** [English](./openclaw-integration.md) | [한국어](./openclaw-integration.ko.md) | [日本語](./openclaw-integration.ja.md) | [简体中文](./openclaw-integration.zh.md) | [繁體中文](./openclaw-integration.zh-TW.md) | [Tiếng Việt](./openclaw-integration.vi.md) | [Español](./openclaw-integration.es.md) | [Português](./openclaw-integration.pt.md) | [Русский](./openclaw-integration.ru.md) | [Türkçe](./openclaw-integration.tr.md) | [Deutsch](./openclaw-integration.de.md) | [Français](./openclaw-integration.fr.md) | [Italiano](./openclaw-integration.it.md) | [Українська](./openclaw-integration.uk.md)
+
+
+Ця сторінка локалізує розділ **«Prompt tuning guide (concise + context-aware)»** з основної англомовної документації.
+
+Повний посібник з інтеграції (gateway, hooks, перевірка) див. у [English guide](./openclaw-integration.md).
+
+## Налаштування промптів (стисло + з урахуванням контексту)
+
+## Де редагувати шаблони промптів
+
+- `notifications.openclaw.hooks["session-start"].instruction`
+- `notifications.openclaw.hooks["session-idle"].instruction`
+- `notifications.openclaw.hooks["ask-user-question"].instruction`
+- `notifications.openclaw.hooks["stop"].instruction`
+- `notifications.openclaw.hooks["session-end"].instruction`
+
+## Рекомендовані контекстні токени
+
+- Завжди включати: `{{sessionId}}`, `{{tmuxSession}}`
+- За подією: `{{projectName}}`, `{{question}}`, `{{reason}}`
+
+## Стратегія деталізації (verbosity)
+
+- `minimal`: дуже короткі сповіщення
+- `session`: стислий операційний контекст (рекомендовано)
+- `verbose`: розширений контекст статусу/дій/ризиків
+
+## Швидка команда оновлення (jq)
+
+```bash
+CONFIG_FILE="$HOME/.codex/.rcs-config.json"
+
+jq '.notifications.verbosity = "verbose" |
+    .notifications.openclaw.hooks["session-start"].instruction = "[session-start|exec]\nproject={{projectName}} session={{sessionId}} tmux={{tmuxSession}}\nпідсумок: контекст старту 1 реченням\nпріоритет: що робити зараз 1–2 пункти\nувага: ризики/залежності (якщо немає — немає)" |
+    .notifications.openclaw.hooks["session-idle"].instruction = "[session-idle|exec]\nsession={{sessionId}} tmux={{tmuxSession}}\nпідсумок: причина idle 1 реченням\nплан відновлення: негайні дії 1–2 пункти\nрішення: чи потрібен ввід користувача" |
+    .notifications.openclaw.hooks["ask-user-question"].instruction = "[ask-user-question|exec]\nsession={{sessionId}} tmux={{tmuxSession}} question={{question}}\nключове питання: потрібна відповідь 1 реченням\nвплив: наслідки відсутності відповіді 1 реченням\nрекомендована відповідь: найшвидший формат відповіді" |
+    .notifications.openclaw.hooks["stop"].instruction = "[session-stop|exec]\nsession={{sessionId}} tmux={{tmuxSession}}\nпідсумок: причина зупинки\nпоточний стан: збережені/незавершені елементи\nвідновлення: перша дія 1 пункт" |
+    .notifications.openclaw.hooks["session-end"].instruction = "[session-end|exec]\nproject={{projectName}} session={{sessionId}} tmux={{tmuxSession}} reason={{reason}}\nрезультат: завершені результати 1–2 реченнями\nперевірка: підтвердження/результати тестів\nдалі: наступні дії 1–2 пункти"'   "$CONFIG_FILE" > "$CONFIG_FILE.tmp" && mv "$CONFIG_FILE.tmp" "$CONFIG_FILE"
+```

package/docs/openclaw-integration.vi.md

@@ -0,0 +1,44 @@
+# Hướng dẫn tích hợp OpenClaw (bản địa hóa tinh chỉnh prompt)
+
+> [← Back to Docs Home](./index.html) · [Integrations Landing](./integrations.html)
+
+**Language Switcher:** [English](./openclaw-integration.md) | [한국어](./openclaw-integration.ko.md) | [日本語](./openclaw-integration.ja.md) | [简体中文](./openclaw-integration.zh.md) | [繁體中文](./openclaw-integration.zh-TW.md) | [Tiếng Việt](./openclaw-integration.vi.md) | [Español](./openclaw-integration.es.md) | [Português](./openclaw-integration.pt.md) | [Русский](./openclaw-integration.ru.md) | [Türkçe](./openclaw-integration.tr.md) | [Deutsch](./openclaw-integration.de.md) | [Français](./openclaw-integration.fr.md) | [Italiano](./openclaw-integration.it.md) | [Українська](./openclaw-integration.uk.md)
+
+
+Trang này bản địa hóa mục **“Prompt tuning guide (concise + context-aware)”** từ tài liệu tiếng Anh chính.
+
+Để xem tài liệu tích hợp đầy đủ (gateway, hook, xác minh), hãy xem [English guide](./openclaw-integration.md).
+
+## Tinh chỉnh prompt (ngắn gọn + nhận biết ngữ cảnh)
+
+## Vị trí chỉnh sửa mẫu prompt
+
+- `notifications.openclaw.hooks["session-start"].instruction`
+- `notifications.openclaw.hooks["session-idle"].instruction`
+- `notifications.openclaw.hooks["ask-user-question"].instruction`
+- `notifications.openclaw.hooks["stop"].instruction`
+- `notifications.openclaw.hooks["session-end"].instruction`
+
+## Token ngữ cảnh khuyến nghị
+
+- Luôn có: `{{sessionId}}`, `{{tmuxSession}}`
+- Theo sự kiện: `{{projectName}}`, `{{question}}`, `{{reason}}`
+
+## Chiến lược mức độ chi tiết (verbosity)
+
+- `minimal`: thông báo rất ngắn
+- `session`: ngữ cảnh vận hành ngắn gọn (khuyến nghị)
+- `verbose`: nhiều ngữ cảnh hơn về trạng thái/hành động/rủi ro
+
+## Lệnh cập nhật nhanh (jq)
+
+```bash
+CONFIG_FILE="$HOME/.codex/.rcs-config.json"
+
+jq '.notifications.verbosity = "verbose" |
+    .notifications.openclaw.hooks["session-start"].instruction = "[session-start|exec]\nproject={{projectName}} session={{sessionId}} tmux={{tmuxSession}}\n요약: 시작 맥락 1문장\n우선순위: 지금 할 일 1~2개\n주의사항: 리스크/의존성(없으면 없음)" |
+    .notifications.openclaw.hooks["session-idle"].instruction = "[session-idle|exec]\nsession={{sessionId}} tmux={{tmuxSession}}\n요약: idle 원인 1문장\n복구계획: 즉시 조치 1~2개\n의사결정: 사용자 입력 필요 여부" |
+    .notifications.openclaw.hooks["ask-user-question"].instruction = "[ask-user-question|exec]\nsession={{sessionId}} tmux={{tmuxSession}} question={{question}}\n핵심질문: 필요한 답변 1문장\n영향: 미응답 시 영향 1문장\n권장응답: 가장 빠른 답변 형태" |
+    .notifications.openclaw.hooks["stop"].instruction = "[session-stop|exec]\nsession={{sessionId}} tmux={{tmuxSession}}\n요약: 중단 사유\n현재상태: 저장/미완료 항목\n재개: 첫 액션 1개" |
+    .notifications.openclaw.hooks["session-end"].instruction = "[session-end|exec]\nproject={{projectName}} session={{sessionId}} tmux={{tmuxSession}} reason={{reason}}\n성과: 완료 결과 1~2문장\n검증: 확인/테스트 결과\n다음: 후속 액션 1~2개"'   "$CONFIG_FILE" > "$CONFIG_FILE.tmp" && mv "$CONFIG_FILE.tmp" "$CONFIG_FILE"
+```

package/docs/openclaw-integration.zh-TW.md

@@ -0,0 +1,44 @@
+# OpenClaw 整合指南（提示詞調校在地化）
+
+> [← Back to Docs Home](./index.html) · [Integrations Landing](./integrations.html)
+
+**Language Switcher:** [English](./openclaw-integration.md) | [한국어](./openclaw-integration.ko.md) | [日本語](./openclaw-integration.ja.md) | [简体中文](./openclaw-integration.zh.md) | [繁體中文](./openclaw-integration.zh-TW.md) | [Tiếng Việt](./openclaw-integration.vi.md) | [Español](./openclaw-integration.es.md) | [Português](./openclaw-integration.pt.md) | [Русский](./openclaw-integration.ru.md) | [Türkçe](./openclaw-integration.tr.md) | [Deutsch](./openclaw-integration.de.md) | [Français](./openclaw-integration.fr.md) | [Italiano](./openclaw-integration.it.md) | [Українська](./openclaw-integration.uk.md)
+
+
+此頁為英文主文件中 **「Prompt tuning guide (concise + context-aware)」** 章節的在地化版本。
+
+完整整合文件（gateway、hooks、驗證）請參考 [English guide](./openclaw-integration.md)。
+
+## 提示詞調校（精簡 + 情境感知）
+
+## 提示詞模板編輯位置
+
+- `notifications.openclaw.hooks["session-start"].instruction`
+- `notifications.openclaw.hooks["session-idle"].instruction`
+- `notifications.openclaw.hooks["ask-user-question"].instruction`
+- `notifications.openclaw.hooks["stop"].instruction`
+- `notifications.openclaw.hooks["session-end"].instruction`
+
+## 建議的情境 Token
+
+- 建議固定包含：`{{sessionId}}`、`{{tmuxSession}}`
+- 依事件補充：`{{projectName}}`、`{{question}}`、`{{reason}}`
+
+## 詳細度（verbosity）策略
+
+- `minimal`：極短通知
+- `session`：精簡的作業情境（建議）
+- `verbose`：提供更完整的狀態/動作/風險資訊
+
+## 快速更新指令（jq）
+
+```bash
+CONFIG_FILE="$HOME/.codex/.rcs-config.json"
+
+jq '.notifications.verbosity = "verbose" |
+    .notifications.openclaw.hooks["session-start"].instruction = "[session-start|exec]\nproject={{projectName}} session={{sessionId}} tmux={{tmuxSession}}\n요약: 시작 맥락 1문장\n우선순위: 지금 할 일 1~2개\n주의사항: 리스크/의존성(없으면 없음)" |
+    .notifications.openclaw.hooks["session-idle"].instruction = "[session-idle|exec]\nsession={{sessionId}} tmux={{tmuxSession}}\n요약: idle 원인 1문장\n복구계획: 즉시 조치 1~2개\n의사결정: 사용자 입력 필요 여부" |
+    .notifications.openclaw.hooks["ask-user-question"].instruction = "[ask-user-question|exec]\nsession={{sessionId}} tmux={{tmuxSession}} question={{question}}\n핵심질문: 필요한 답변 1문장\n영향: 미응답 시 영향 1문장\n권장응답: 가장 빠른 답변 형태" |
+    .notifications.openclaw.hooks["stop"].instruction = "[session-stop|exec]\nsession={{sessionId}} tmux={{tmuxSession}}\n요약: 중단 사유\n현재상태: 저장/미완료 항목\n재개: 첫 액션 1개" |
+    .notifications.openclaw.hooks["session-end"].instruction = "[session-end|exec]\nproject={{projectName}} session={{sessionId}} tmux={{tmuxSession}} reason={{reason}}\n성과: 완료 결과 1~2문장\n검증: 확인/테스트 결과\n다음: 후속 액션 1~2개"'   "$CONFIG_FILE" > "$CONFIG_FILE.tmp" && mv "$CONFIG_FILE.tmp" "$CONFIG_FILE"
+```

package/docs/openclaw-integration.zh.md

@@ -0,0 +1,44 @@
+# OpenClaw 集成指南（提示词调优本地化）
+
+> [← Back to Docs Home](./index.html) · [Integrations Landing](./integrations.html)
+
+**Language Switcher:** [English](./openclaw-integration.md) | [한국어](./openclaw-integration.ko.md) | [日本語](./openclaw-integration.ja.md) | [简体中文](./openclaw-integration.zh.md) | [繁體中文](./openclaw-integration.zh-TW.md) | [Tiếng Việt](./openclaw-integration.vi.md) | [Español](./openclaw-integration.es.md) | [Português](./openclaw-integration.pt.md) | [Русский](./openclaw-integration.ru.md) | [Türkçe](./openclaw-integration.tr.md) | [Deutsch](./openclaw-integration.de.md) | [Français](./openclaw-integration.fr.md) | [Italiano](./openclaw-integration.it.md) | [Українська](./openclaw-integration.uk.md)
+
+
+此页面是英文主文档中 **“Prompt tuning guide (concise + context-aware)”** 章节的本地化版本。
+
+完整集成文档（gateway、hooks、验证）请参阅 [English guide](./openclaw-integration.md)。
+
+## 提示词调优（简洁 + 上下文感知）
+
+## 提示词模板编辑位置
+
+- `notifications.openclaw.hooks["session-start"].instruction`
+- `notifications.openclaw.hooks["session-idle"].instruction`
+- `notifications.openclaw.hooks["ask-user-question"].instruction`
+- `notifications.openclaw.hooks["stop"].instruction`
+- `notifications.openclaw.hooks["session-end"].instruction`
+
+## 推荐上下文令牌
+
+- 始终包含：`{{sessionId}}`、`{{tmuxSession}}`
+- 按事件选择：`{{projectName}}`、`{{question}}`、`{{reason}}`
+
+## 详细度（verbosity）策略
+
+- `minimal`：超短通知
+- `session`：简洁的会话运营上下文（推荐）
+- `verbose`：更丰富的状态/动作/风险信息
+
+## 快速更新命令（jq）
+
+```bash
+CONFIG_FILE="$HOME/.codex/.rcs-config.json"
+
+jq '.notifications.verbosity = "verbose" |
+    .notifications.openclaw.hooks["session-start"].instruction = "[session-start|exec]\nproject={{projectName}} session={{sessionId}} tmux={{tmuxSession}}\n요약: 시작 맥락 1문장\n우선순위: 지금 할 일 1~2개\n주의사항: 리스크/의존성(없으면 없음)" |
+    .notifications.openclaw.hooks["session-idle"].instruction = "[session-idle|exec]\nsession={{sessionId}} tmux={{tmuxSession}}\n요약: idle 원인 1문장\n복구계획: 즉시 조치 1~2개\n의사결정: 사용자 입력 필요 여부" |
+    .notifications.openclaw.hooks["ask-user-question"].instruction = "[ask-user-question|exec]\nsession={{sessionId}} tmux={{tmuxSession}} question={{question}}\n핵심질문: 필요한 답변 1문장\n영향: 미응답 시 영향 1문장\n권장응답: 가장 빠른 답변 형태" |
+    .notifications.openclaw.hooks["stop"].instruction = "[session-stop|exec]\nsession={{sessionId}} tmux={{tmuxSession}}\n요약: 중단 사유\n현재상태: 저장/미완료 항목\n재개: 첫 액션 1개" |
+    .notifications.openclaw.hooks["session-end"].instruction = "[session-end|exec]\nproject={{projectName}} session={{sessionId}} tmux={{tmuxSession}} reason={{reason}}\n성과: 완료 결과 1~2문장\n검증: 확인/테스트 결과\n다음: 후속 액션 1~2개"'   "$CONFIG_FILE" > "$CONFIG_FILE.tmp" && mv "$CONFIG_FILE.tmp" "$CONFIG_FILE"
+```

package/docs/plugin-bundle-ssot.md

@@ -0,0 +1,52 @@
+# Plugin Bundle SSOT Contract
+
+The repository keeps one canonical authoring surface for each plugin/setup asset type and treats `plugins/roblox-ai-os-creator-skills` as generated-or-verified plugin output.
+
+## Canonical roots
+
+- **Plugin skills:** root `skills/<name>/SKILL.md` is canonical. The plugin mirror at `plugins/roblox-ai-os-creator-skills/skills/<name>/` is refreshed by `npm run sync:plugin` and verified by `npm run verify:plugin-bundle`.
+- **Plugin skill membership:** `templates/catalog-manifest.json` controls which catalog skills are installable. Active/internal skills, plus setup-only policy additions, must have canonical root skill directories and plugin mirrors.
+- **Plugin MCP metadata:** `src/config/rcs-first-party-mcp.ts` is canonical. `plugins/roblox-ai-os-creator-skills/.mcp.json` must match `buildRcsPluginMcpManifest()`.
+- **Plugin manifest version and paths:** `package.json` is canonical for the plugin version. The plugin manifest must point to `./skills/`, `./.mcp.json`, and `./.app.json`.
+- **Native agents and prompts:** root `prompts/` plus `src/agents/definitions.ts` are setup-owned canonical sources for legacy setup mode. The official plugin intentionally does not ship plugin-scoped `agents`, `prompts`, or hooks; plugin setup archives/removes legacy RCS-managed prompt and native-agent files so users do not keep stale mixed surfaces.
+
+## Commands
+
+```bash
+npm run sync:plugin           # mutate plugin mirror/metadata from canonical roots
+npm run verify:plugin-bundle  # non-mutating SSOT check for CI/review
+npm run sync:plugin:check     # compatibility alias for the same non-mutating check
+```
+
+`prepack` runs sync and verification before packaging, but contributors should run the non-mutating verification before review so release-time sync does not hide stale plugin artifacts.
+
+## Adding or changing a skill
+
+1. Edit or add the root skill under `skills/<name>/SKILL.md`.
+2. Add/update the skill entry in `templates/catalog-manifest.json` and `src/catalog/manifest.json`.
+3. Run `npm run build && npm run sync:plugin`.
+4. Run `npm run verify:plugin-bundle`.
+
+Root skill directories that should not appear in the plugin must be represented in the catalog as `alias` or `merged`; otherwise the plugin bundle verifier fails because the root directory is neither installable nor explicitly excluded.
+
+## Native agent SSOT
+
+Native agents are setup-owned assets, not plugin-scoped bundle assets. Generated native-agent TOMLs and prompt files intentionally use different setup policies:
+
+1. `templates/catalog-manifest.json` and `src/catalog/manifest.json` select installable native agent TOMLs with `active` or `internal` status.
+2. `src/agents/definitions.ts` defines each native agent's metadata, model lane, posture, routing role, and tooling posture.
+3. `prompts/<name>.md` supplies prompt guidance. Cataloged prompt files remain setup-owned prompt assets even when their agent status is `merged`, `alias`, or `deprecated`; explicit non-native prompt assets such as harness/orchestrator prompts are also setup-owned.
+4. `src/agents/native-config.ts` generates native Codex TOML from the definition plus prompt for active/internal native agents only.
+5. `rcs setup` writes generated TOML to `.codex/agents/<name>.toml` for the selected scope and installs setup-owned prompts to `.codex/prompts/<name>.md`.
+
+Run the non-mutating native-agent verifier before review or release:
+
+```bash
+npm run verify:native-agents
+```
+
+The verifier fails when installable catalog agents are missing definitions or prompts, when definitions are missing catalog rows, when merged/alias canonical targets do not resolve directly to installable agents, when prompt files are neither cataloged native agents nor explicit setup prompt assets, or when generated TOML loses required metadata.
+
+`rcs setup` converges generated native-agent TOMLs safely: normal setup removes stale non-installable TOMLs only when they carry the exact `# roblox-ai-os-creator-skills agent: <name>` generated marker for the same cataloged agent name. User-authored or ambiguous TOMLs are preserved during normal setup; `--force` remains the explicit destructive cleanup path for stale non-installable native-agent files. Prompt cleanup uses the prompt asset policy, so cataloged prompt files and explicit setup prompt assets are preserved even when they are not installable native-agent TOMLs.
+
+The official plugin manifest must continue to omit `agents`, `prompts`, and `hooks`; legacy setup mode installs prompts/native agents, while plugin setup mode removes archived legacy copies instead of refreshing plugin-scoped prompt or agent assets.

package/docs/prompt-guidance-contract.md

@@ -0,0 +1,233 @@
+# GPT-5.5 Prompt Guidance Contract
+
+Status: contributor-facing contract for RCS prompt and orchestration surfaces.
+
+## Purpose
+
+This document explains the active **behavioral prompt contract** for RCS after Issue [#2007](https://github.com/Yeachan-Heo/roblox-ai-os-creator-skills/issues/2007): align prompt and instruction surfaces with OpenAI's official [GPT-5.5 prompt guidance](https://developers.openai.com/api/docs/guides/prompt-guidance) while preserving RCS product contracts.
+
+Use it when you edit any of these surfaces:
+
+- `AGENTS.md` when a repo chooses to track a project-root copy
+- `templates/AGENTS.md`
+- canonical XML-tagged role prompt surfaces in `prompts/*.md`
+- workflow skill instructions in `skills/*/SKILL.md`
+- hook/generator prompt guidance and regression tests under `src/hooks/` and `src/config/`
+
+## Scope and current source of truth
+
+The current prompt sources in this repository live in **`prompts/*.md`** and **`skills/*/SKILL.md`**, then get installed to `~/.codex/prompts/`, `~/.codex/skills/`, and native agent wrappers. Treat the source markdown body as canonical; launcher-specific TOML or runtime wrappers should preserve the same behavior.
+
+The GPT-5.5 contract is distributed across:
+
+- orchestration surfaces: `templates/AGENTS.md` and any tracked project-root `AGENTS.md`
+- shared fragments: `docs/prompt-guidance-fragments/*`
+- canonical XML-tagged subagent role prompts: `prompts/*.md`
+- workflow skills: `skills/*/SKILL.md`
+- generated top-level Codex config guidance: `src/config/generator.ts`
+- regression tests: `src/hooks/__tests__/prompt-guidance-*.test.ts`
+
+## Workflow skill guidance dedupe
+
+Workflow skills may use a compact reference to the shared workflow guidance pattern instead of repeating every GPT-5.5 bullet. That pattern must preserve: outcome-first framing, concise visible updates for multi-step work, scoped task-update overrides, evidence-backed validation, and explicit stop/escalation rules. Workflow-specific invariants such as state transitions, gates, cleanup, cancellation, and verification commands remain explicit in the owning skill.
+
+## Exact-model mini adaptation seam
+
+RCS also has a narrow **instruction-composition seam** for subagents/workers whose **final resolved model** is exactly `gpt-5.4-mini`.
+That seam is part of prompt delivery, but it is intentionally narrower than the general GPT-5.5 behavioral contract described below.
+
+Contributor rules for that seam:
+
+- Key mini-specific instruction adaptation off the **final resolved model string**, not off role name, lane, or default tier membership.
+- Use **exact string equality** for `gpt-5.4-mini`; do not widen behavior to `gpt-5.5`, `gpt-5.4-mini-tuned`, or other variants.
+- Keep one shared **inner role-instruction composition helper** as the source of truth for model-gated prompt adaptation.
+- Keep `src/team/worker-bootstrap.ts` limited to **outer AGENTS/runtime wrapping**. It should wrap already-composed instructions, not own model-specific adaptation logic.
+- Keep `src/team/role-router.ts` as a raw role-prompt loader unless a minimal plumbing change is unavoidable.
+
+Primary implementation surfaces for this seam:
+
+| Responsibility | Primary sources |
+|---|---|
+| shared inner prompt composition | `src/agents/native-config.ts`, `src/agents/__tests__/native-config.test.ts` |
+| team runtime/scaling plumbing | `src/team/runtime.ts`, `src/team/scaling.ts`, associated runtime/scaling tests |
+| outer wrapper boundary | `src/team/worker-bootstrap.ts`, `src/team/__tests__/worker-bootstrap.test.ts` |
+
+## What this contract is — and is not
+
+This contract is about **how RCS prompts should behave**. It is not the same thing as RCS's routing metadata.
+
+- **Behavioral contract:** outcome-first defaults, concise collaboration style, low-risk follow-through, localized task updates, evidence-backed validation, and explicit stop rules.
+- **Adjacent but separate routing layer:** role/tier/posture metadata such as `frontier-orchestrator`, `deep-worker`, and `fast-lane` in `src/agents/native-config.ts` and `docs/shared/agent-tiers.md`.
+
+If you are changing prompt prose, use this document first. If you are changing routing metadata or native config overlays, use the routing docs/tests first.
+
+## The 5 core GPT-5.5 patterns RCS should enforce
+
+### 1. Outcome-first, success-criteria-led prompts
+
+Contributors should describe the target result, success criteria, constraints, available evidence, expected output, and stop condition before adding process detail. Avoid process-heavy stacks unless the process is itself a product contract.
+
+Representative locations:
+
+| Surface | Evidence |
+|---|---|
+| shared fragments | `docs/prompt-guidance-fragments/core-operating-principles.md` |
+| root orchestration | `templates/AGENTS.md` |
+| core roles | `prompts/executor.md`, `prompts/planner.md`, `prompts/verifier.md` |
+| contract tests | `src/hooks/prompt-guidance-contract.ts` and `src/hooks/__tests__/prompt-guidance-*.test.ts` |
+
+Example prompt text:
+
+> Default to outcome-first, quality-focused responses: identify the user's target result, success criteria, constraints, available evidence, expected output, and stop condition before adding process detail.
+
+### 2. Concise personality/collaboration style with preambles for longer work
+
+Keep tone and collaboration steering short. For multi-step or tool-heavy tasks, start with a brief visible preamble that acknowledges the request and names the first step; keep later updates brief and evidence-based.
+
+Representative locations:
+
+| Surface | Evidence |
+|---|---|
+| root orchestration | `templates/AGENTS.md` |
+| executor/planner/verifier fragments | `docs/prompt-guidance-fragments/*-constraints.md` |
+| core prompts | `prompts/executor.md`, `prompts/planner.md`, `prompts/verifier.md` |
+
+### 3. Automatic follow-through on clear, low-risk, reversible next steps
+
+Contributors should preserve the bias toward continuing useful work automatically instead of asking avoidable confirmation questions.
+
+Example prompt text:
+
+> Proceed automatically on clear, low-risk, reversible next steps; ask only for irreversible, credential-gated, external-production, destructive, or materially scope-changing actions.
+
+Also preserve agent-owned safe runtime work:
+
+> Do not ask or instruct humans to perform ordinary non-destructive, reversible actions; execute those safe reversible RCS/runtime operations and ordinary commands yourself.
+
+### 4. Localized task-update overrides that preserve earlier non-conflicting instructions
+
+Contributors should treat user updates as **scoped overrides**, not full prompt resets.
+
+Example prompt text:
+
+> Treat newer user task updates as local overrides for the active task while preserving earlier non-conflicting instructions.
+
+Scenario examples for `continue`, `make a PR`, and `merge if CI green` reinforce this behavior in `prompts/executor.md`, `prompts/planner.md`, `prompts/verifier.md`, and related tests.
+
+### 5. Evidence budgets, validation, and explicit stop rules
+
+GPT-5.5 guidance favors enough retrieval/validation to answer correctly, then stopping. RCS prompts should continue tool use while correctness depends on repository inspection, official docs, diagnostics, tests, citations, or verification, but avoid extra loops that only improve phrasing or gather nonessential evidence.
+
+For coding work, prompts should ask for concrete validation:
+
+- targeted tests for changed behavior
+- typecheck/lint/build checks when applicable
+- a minimal smoke test when full validation is too expensive
+- an explicit reason and next-best check when validation cannot run
+
+Implementation plans should stay traceable: requirements, named files/resources/APIs, state transitions or data flow where relevant, validation commands, failure behavior, privacy/security considerations, and material open questions.
+
+## Absolute-language rule
+
+Use `MUST`, `NEVER`, `ALWAYS`, `only`, and similar absolute wording for true invariants: safety/security boundaries, side-effect constraints, required output fields, workflow state transitions, team/ralph gates, and product contracts. For judgment calls such as whether to search again, ask for clarification, or keep iterating, prefer decision rules and stop conditions.
+
+## Active workflow terminal handoff contract
+
+Prompt surfaces that control active workflows should describe terminal user-facing replies as explicit handoffs, not casual optional follow-ups.
+
+Contributor rules:
+
+1. Terminal active-workflow replies should name an explicit outcome such as `finished`, `blocked`, `failed`, `userinterlude`, or `askuserQuestion`.
+2. Terminal replies should include the evidence or blocking reason that justifies that outcome.
+3. Terminal replies should identify the handoff clearly: completed artifact, blocking dependency, failure recovery owner, or the single required question.
+4. Terminal replies should not end in permission-seeking softeners such as `If you want, I can ...`, `If you'd like, I can ...`, or `Would you like me to continue?`.
+
+This rule is specific to active workflow handoffs. Normal explanatory conversation outside an active workflow may still be conversational, but workflow-owned terminal replies must make the lifecycle state explicit.
+
+## Orchestration sharpness rules for root AGENTS surfaces
+
+When editing `templates/AGENTS.md`, any tracked root `AGENTS.md`, or other root orchestration guidance, keep the orchestration contract mode-driven and terse:
+
+1. **Mode selection comes first.** Distinguish between `$deep-interview`, `$ralplan`, `$team`, and direct solo execution instead of blending them into one generic flow.
+2. **Leader and worker responsibilities stay separate.** Leaders choose the mode, own verification, and integrate work; workers execute assigned slices and report blockers upward.
+3. **Stop/escalate rules are explicit.** The prompt should say when to stop, when to escalate to the user, and when workers must escalate back to the leader.
+4. **Output contract stays tight.** Default progress/final updates should be compact: current mode, action/result, and evidence or blocker/next step. Avoid repeating full-plan rationale unless the risk or decision changed.
+
+## Relationship to the guidance schema
+
+`docs/guidance-schema.md` defines the **section layout contract** for AGENTS and worker surfaces. This document defines the **behavioral wording contract** that should appear within those sections after the GPT-5.5 rollout.
+
+Use both documents together:
+
+- `docs/guidance-schema.md` for structure
+- `docs/prompt-guidance-contract.md` for behavior
+
+## Relationship to posture-aware routing
+
+Posture-aware routing is real, but it is not the same contract as the GPT-5.5 behavior rollout. Keep these separate when editing docs and prompts:
+
+| Topic | Primary sources |
+|---|---|
+| GPT-5.5 prompt behavior contract | `templates/AGENTS.md`, any tracked `AGENTS.md`, canonical XML-tagged role prompt surfaces in `prompts/*.md`, workflow skills in `skills/*/SKILL.md`, `src/config/generator.ts`, `src/hooks/__tests__/prompt-guidance-*.test.ts` |
+| exact-model mini composition seam | `src/agents/native-config.ts`, `src/team/runtime.ts`, `src/team/scaling.ts`, `src/team/worker-bootstrap.ts`, targeted native/runtime/scaling/bootstrap tests |
+| role/tier/posture routing | `README.md:133-179`, `docs/shared/agent-tiers.md`, `src/agents/native-config.ts` |
+
+If a change only affects posture overlays or native agent metadata, document it in the routing docs rather than expanding this contract unnecessarily.
+
+## Canonical role prompts vs specialized behavior prompts
+
+The main role catalog is the installable specialized-agent set used by native agent generation and internal role prompt composition.
+
+- Files like `prompts/executor.md`, `prompts/planner.md`, and `prompts/architect.md` are canonical XML-tagged role prompt surfaces.
+- `prompts/sisyphus-lite.md` should be treated as a specialized worker-behavior prompt, not as a first-class main catalog role.
+- Worker/runtime overlays may compose that behavior under worker protocol constraints without promoting it to the primary public role catalog.
+
+## Contributor checklist for prompt changes
+
+Before opening a PR that changes prompt text, confirm all of the following:
+
+1. **Preserve the five core behaviors.** Your change should keep or strengthen outcome-first framing, concise collaboration/preambles, low-risk follow-through, scoped overrides, and evidence-backed validation/stop rules.
+2. **Keep role-specific wording role-specific.** The phrasing can differ by role, but the behavior should stay semantically aligned.
+3. **Update scenario examples when behavior changes.** If you change how prompts handle `continue`, `make a PR`, or `merge if CI green`, update the prompt examples and related tests.
+4. **Keep the mini-only seam exact and centralized.** If you touch mini adaptation, gate it on the final resolved model with exact `gpt-5.4-mini` equality, keep the shared inner helper as the source of truth, and keep `worker-bootstrap.ts` wrapper-only.
+5. **Do not confuse routing metadata with prompt behavior.** Posture/tier updates belong in routing docs/tests unless they also change prompt prose.
+6. **Update regression coverage when the contract changes.** Start with `src/hooks/__tests__/prompt-guidance-contract.test.ts`, `prompt-guidance-wave-two.test.ts`, `prompt-guidance-scenarios.test.ts`, `prompt-guidance-catalog.test.ts`, `skill-guidance-contract.test.ts`, and `prompt-guidance-fragments.test.ts`; add native/runtime/scaling/bootstrap coverage when the mini-only seam changes.
+
+## Validation workflow for contributors
+
+For prompt-guidance edits, run at least:
+
+```bash
+npm run build
+node --test \
+  dist/hooks/__tests__/prompt-guidance-contract.test.js \
+  dist/hooks/__tests__/prompt-guidance-wave-two.test.js \
+  dist/hooks/__tests__/prompt-guidance-scenarios.test.js \
+  dist/hooks/__tests__/prompt-guidance-catalog.test.js \
+  dist/hooks/__tests__/skill-guidance-contract.test.js \
+  dist/hooks/__tests__/prompt-guidance-fragments.test.js \
+  dist/hooks/__tests__/explicit-terminal-stop-docs-contract.test.js
+```
+
+If you touch the exact-model `gpt-5.4-mini` composition seam, also run:
+
+```bash
+node --test \
+  dist/agents/__tests__/native-config.test.js \
+  dist/team/__tests__/runtime.test.js \
+  dist/team/__tests__/scaling.test.js \
+  dist/team/__tests__/worker-bootstrap.test.js
+```
+
+For broader prompt or skill changes, prefer the full suite:
+
+```bash
+npm test
+```
+
+## References
+
+- Implementation issue: [#2007](https://github.com/Yeachan-Heo/roblox-ai-os-creator-skills/issues/2007)
+- Official source: [OpenAI GPT-5.5 prompt guidance](https://developers.openai.com/api/docs/guides/prompt-guidance)
+- Prior rollout history: [#608](https://github.com/Yeachan-Heo/roblox-ai-os-creator-skills/issues/608), [#611](https://github.com/Yeachan-Heo/roblox-ai-os-creator-skills/pull/611), [#612](https://github.com/Yeachan-Heo/roblox-ai-os-creator-skills/pull/612)
+- Guidance schema: `docs/guidance-schema.md`

package/docs/prompt-guidance-fragments/core-operating-principles.md

@@ -0,0 +1,16 @@
+- Default to outcome-first, quality-focused responses: identify the user's target result, success criteria, constraints, available evidence, expected output, and stop condition before adding process detail.
+- Keep collaboration style short and direct. Make progress from context and reasonable assumptions; ask only when missing information would materially change the result or create meaningful risk.
+- Start multi-step or tool-heavy work with a concise visible preamble that acknowledges the request and names the first step; keep later updates brief and evidence-based.
+- Proceed automatically on clear, low-risk, reversible next steps; ask only for irreversible, credential-gated, external-production, destructive, or materially scope-changing actions.
+- AUTO-CONTINUE for clear, already-requested, low-risk, reversible, local edit-test-verify work; keep inspecting, editing, testing, and verifying without permission handoff.
+- ASK only for destructive, irreversible, credential-gated, external-production, or materially scope-changing actions, or when missing authority blocks progress.
+- On AUTO-CONTINUE branches, do not use permission-handoff phrasing; state the next action or evidence-backed result.
+- Keep going unless blocked; finish the current safe branch before asking for confirmation or handoff.
+- Ask only when blocked by missing information, missing authority, or an irreversible/destructive branch.
+- Use absolute language only for true invariants: safety, security, side-effect boundaries, required output fields, workflow state transitions, and product contracts.
+- Do not ask or instruct humans to perform ordinary non-destructive, reversible actions; execute those safe reversible RCS/runtime operations and ordinary commands yourself.
+- Treat RCS runtime manipulation, state transitions, and ordinary command execution as agent responsibilities when they are safe and reversible.
+- Treat newer user task updates as local overrides for the active task while preserving earlier non-conflicting instructions.
+- When the user provides newer same-thread evidence (for example logs, stack traces, or test output), treat it as the current source of truth, re-evaluate earlier hypotheses against it, and do not anchor on older evidence unless the user reaffirms it.
+- Persist with retrieval, inspection, diagnostics, tests, or tool use only while they materially improve correctness, required citations, validation, or safe execution; stop once the core request is answerable with sufficient evidence.
+- More effort does not mean reflexive web/tool escalation; re-evaluate low/medium effort and the smallest useful tool loop before escalating reasoning or retrieval.

package/docs/prompt-guidance-fragments/core-verification-and-sequencing.md

@@ -0,0 +1,6 @@
+Verification loop: define the claim and success criteria, run the smallest validation that can prove it, read the output, then report with evidence. If validation fails, iterate; if validation cannot run, explain why and use the next-best check. Keep evidence summaries concise but sufficient.
+
+- Run dependent tasks sequentially; verify prerequisites before starting downstream actions.
+- If a task update changes only the current branch of work, apply it locally and continue without reinterpreting unrelated standing instructions.
+- For coding work, prefer targeted tests for changed behavior, then typecheck/lint/build/smoke checks when applicable; do not claim completion without fresh evidence or an explicit validation gap.
+- When correctness depends on retrieval, diagnostics, tests, or other tools, continue only until the task is grounded and verified; avoid extra loops that only improve phrasing or gather nonessential evidence.

package/docs/prompt-guidance-fragments/executor-constraints.md

@@ -0,0 +1,13 @@
+- Default to outcome-first, quality-focused execution: clarify the target result, constraints, success criteria, validation path, and stop condition before adding process detail.
+- Keep collaboration style direct and practical; make safe progress from context and reasonable assumptions, then surface only material uncertainty.
+- Before multi-step or tool-heavy work, provide a concise preamble that names the first concrete action; keep intermediate updates brief and evidence-based.
+- Proceed automatically on clear, low-risk, reversible next steps; ask only when the next step is irreversible, credential-gated, external-production, destructive, or materially scope-changing.
+- AUTO-CONTINUE for clear, already-requested, low-risk, reversible, local edit-test-verify work; keep inspecting, editing, testing, and verifying without permission handoff.
+- ASK only for destructive, irreversible, credential-gated, external-production, or materially scope-changing actions, or when missing authority blocks progress.
+- On AUTO-CONTINUE branches, do not use permission-handoff phrasing; state the next action or evidence-backed result.
+- Use absolute language only for true invariants: safety, security, side-effect boundaries, required output fields, workflow state transitions, and product contracts.
+- Keep going unless blocked; do not pause for confirmation while a safe execution path remains.
+- Ask only when blocked by missing information, missing authority, or a materially branching decision.
+- Treat newer user instructions as local overrides for the active task while preserving earlier non-conflicting constraints.
+- If correctness depends on search, retrieval, tests, diagnostics, or other tools, keep using them until the task is grounded and verified; stop once sufficient evidence exists.
+- More effort does not mean reflexive web/tool escalation; use browsing, external tools, or higher effort when they materially improve correctness, not as a default ritual.

package/docs/prompt-guidance-fragments/executor-output.md

@@ -0,0 +1 @@
+Default final-output shape: outcome-first and evidence-dense; state what changed, what validation proves it, known gaps or risks, and the stop condition reached without padding.

package/docs/prompt-guidance-fragments/executor-shared.md

@@ -0,0 +1,7 @@
+- Default to outcome-first, quality-focused execution: clarify the target result, constraints, success criteria, validation path, and stop condition before adding process detail.
+- Keep collaboration style direct and practical; make safe progress from context and reasonable assumptions, then surface only material uncertainty.
+- Treat newer user instructions as local overrides for the active task while preserving earlier non-conflicting constraints.
+- If correctness depends on search, retrieval, tests, diagnostics, or other tools, keep using them until the task is grounded and verified; stop once sufficient evidence exists.
+- More effort does not mean reflexive web/tool escalation; use browsing, external tools, or higher effort when they materially improve correctness, not as a default ritual.
+
+Default final-output shape: outcome-first and evidence-dense; state what changed, what validation proves it, known gaps or risks, and the stop condition reached without padding.