Update to Replai

{
  "description": "Set up and run a Replai AI customer support Agent through Replai's MCP server: connect, choose the Workspace, add knowledge, preview answers, connect LINE or the website widget, publish, and work the Inbox. Use when the user asks to do this in Replai.",
  "included_files": [],
  "name": "replai-setup",
  "skill_md_contents": "---\nname: replai-setup\ndescription: \"Set up and run a Replai AI customer support Agent through Replai's MCP server: connect, choose the Workspace, add knowledge, preview answers, connect LINE or the website widget, publish, and work the Inbox. Use when the user asks to do this in Replai.\"\n---\n\n# Set up and run a Replai support Agent\n\nReplai gives SaaS teams an AI customer support Agent grounded in their product knowledge.\nIt answers on a website chat widget or LINE and hands conversations to the team's Inbox when\na person is needed.\nThis skill covers doing that setup with Replai's remote MCP server.\nGuide: https://replai.tw/en/guide/connect-ai-agents\n\n## 1. Connect\n\nMCP server: `https://api.replai.tw/mcp` (Streamable HTTP, OAuth).\n\nUse the Replai MCP connection included with the assistant's installed package. For other clients,\nadd the server URL in the client's MCP settings with OAuth authentication. The connection guide\nabove has client-specific registration instructions.\n\nThe first connection opens the user's browser. The user signs in to Replai with Google, chooses\none Workspace and approves the requested permissions. This step belongs to the user; do not try\nto complete it for them, and never ask for passwords or API keys.\n\nUse only the connected Replai MCP tools. No local software installation or shell execution is needed.\n\n## 2. Confirm the Workspace\n\nCall `connection_context` first. It returns the one Workspace this connection can reach and the\ngranted operation scopes. Each connection reaches exactly one Workspace; to work in another one,\nthe user reconnects and chooses it. If a scope you need is missing, tell the user which\npermission to approve.\n\n## 3. Create or choose an Agent\n\nCall `agents_list` and reuse an existing Agent when one fits; refer to Agents by ID, because\nnames can repeat. Otherwise call `agent_create`. Read the draft with `agent_draft_read` and\nchange it with `agent_draft_edit`. Draft edits never reach customers until publication.\n\n## 4. Add knowledge\n\n- Website: `website_discover`, then `website_import` with the pages to import.\n- Text or a single URL: `source_add`.\n- Google Drive or GitHub: check `provider_status`; if not connected, `provider_connect_handoff`\n  returns a Replai link for the user to open. Follow it with the tool its `readback_tool` names\n  (`provider_connect_attempt_read` for GitHub) until it reports connected. Then\n  `drive_files_list` / `source_import_drive` or `source_import_repository`.\n- Files on the user's computer: `source_upload_handoff` returns an upload link for the user;\n  MCP cannot read local paths.\n- Imports run asynchronously: follow them with `source_operation_read` and `sources_list`.\n\n## 5. Preview answers\n\nSave representative customer questions with `test_case_create`, run them with\n`test_case_run` (it runs the draft and uses the Workspace's monthly test allowance; do not\nrerun automatically after a timeout) and read results with `test_result_read`. `knowledge_gaps_list` shows questions the\nAgent could not answer.\n\n## 6. Connect a channel\n\n- Website widget: `widget_enable`, then `widget_installation_read` for the install code. The\n  code goes before `</body>` on the user's site. `widget_settings_edit` changes its look and copy.\n- LINE: `provider_connect_handoff` with provider `line` returns the setup link where the user\n  enters the LINE channel credentials. Then `line_connection_verify`. Group behavior:\n  `line_behavior_read` / `line_behavior_edit`.\n\n## 7. Publish\n\nCall `publication_review`, show the user the candidate, and publish only when the user agrees:\n`publication_publish` with that `review_id`. Publishing changes what customers receive.\n`publication_history` lists versions; `publication_restore` puts an earlier version back into\nthe draft, which then needs publishing again.\n\n## 8. After launch\n\n- Inbox: `inbox_list`, `inbox_read`, `inbox_reply`, `inbox_drafts`, `inbox_draft_approve`,\n  `inbox_takeover`, `inbox_return_to_ai`, `inbox_resolve`.\n- Insights: `agent_analytics`, `agent_kpis_read`, `workspace_usage`.\n\n## Rules for every write\n\n- Tools that take `request_key` need a new UUID that you keep before calling. After a lost\n  response, repeat the identical request with the same key; it returns the original result\n  and does not act twice.\n- Errors carry `code` and `next_action`. Read current state before trying again.\n- Removal tools (`agent_remove`, `channel_remove`, `provider_disconnect`) each have an\n  impact read (`agent_removal_impact`, `channel_removal_impact`,\n  `provider_disconnect_impact`); show it to the user and remove only on their explicit request.\n- The user can revoke this access in Replai under Workspace settings → MCP & CLI.\n"
}