💬Text & WhatsApp Agents
The 7 text and WhatsApp agent skills: create chat agents, search and rank them, calibrate behavior, manage the website widget and WhatsApp lines, and read or rewrite prompts.
These skills build and tune your text and chat agents, and manage the channels they run on (website widget, WhatsApp, API).
Writes use a preview, then commit flow: Claude shows the exact change and asks for your confirmation before applying it.
create-text-agent
What it does: Walks you through creating a new text or chat agent in a few quick steps: pick a template, choose a language, and add your company info (auto-pulled from your website URL or entered manually), then builds the agent with your real company details baked in.
Use it when: You want to spin up a new chat agent for customer service, sales or ecommerce, reception or scheduling, or a custom use case.
Try saying:
"Create a customer service chat agent for my company."
"Set up a sales chat agent, here's our website: example.com"
"I want a receptionist chat agent in Spanish for my dental clinic."
You get back: A ready-to-use chat agent with your company information already in it, a link to open it in the platform, and suggested starter questions your widget can offer visitors. After creation you are invited to test it and ask for tweaks.
Writes data? Yes, via preview then commit. Claude shows the agent details and asks "Create this agent? Yes or No" before creating it.
Behind the scenes: get_workspace_company_context (reuse info on file), tavily_extract (scrape your website), preview_create_text_agent and commit_create_text_agent (the create), generate_text_agent_example_questions (starter questions). Fine-tuning afterward hands off to text-agent-calibrator.
text-agent-searcher
What it does: Searches, lists, counts, and ranks your text and chat agents, including which channels (website widget, WhatsApp, API) they are connected to and how busy they are.
Use it when: You want to find a specific chat agent, see all your agents, check which ones handle the most messages, or see what channels and WhatsApp lines are connected.
Try saying:
"What chat agents do I have?"
"Which of my agents handles the most messages?"
"What WhatsApp lines are connected to my agents?"
You get back: A summary and table of your agents (name, type, status, model, created date), rankings by message volume, status breakdowns, full details on a specific agent, or your connected WhatsApp lines with usage stats.
Writes data? No, read-only.
Behind the scenes: list_all_text_agents, search_text_agent and search_text_agents_by_name, get_text_agent_details and get_text_agent_details_query, count_summary, top_agents_by_messages, get_workspace_whatsapp_lines. Handles both older (v1) and current (v2) agents.
text-agent-calibrator
What it does: Helps you understand and fine-tune how an existing chat agent behaves. It reads the agent's setup, reviews recent real conversations to spot mistakes or gaps, and proposes targeted prompt tweaks you approve one at a time.
Use it when: Your chat agent is answering something wrong, using the wrong tone, missing handovers, or you want to tune it based on how recent chats actually went.
Try saying:
"Why is my agent answering pricing questions like that?"
"Let's calibrate Diana based on this week's conversations."
"Adjust how my chat agent handles complaints."
You get back: A diagnosis grounded in real recent conversations, a proposed targeted fix in plain language, a clear before-and-after preview, and, once approved, the updated agent plus a link. It loops with you for up to 5 rounds of test-and-tune per session.
Writes data? Yes, via preview then commit. Shows the exact before and after and asks "Apply this update? Yes or No".
Behind the scenes: reads with get_text_agent_details_query, get_text_agent_company_context, get_text_agent_context, get_text_agent_conversations; resolves names with search_text_agents_by_name; applies via preview_update_text_agent and commit_update_text_agent. Full rewrites hand off to text-prompt-improver.
text-agent-widget
What it does: Connects a voice agent to your website chat widget so visitors can escalate from a chat to a phone call, and lets you check, swap, or remove that voice agent link. Creates the widget automatically if one does not exist yet.
Use it when: You want to add, change, or remove the voice agent that powers the "call" option in your text agent's website widget.
Try saying:
"Attach my Sofia voice agent to my chat widget."
"What voice agent is connected to my widget right now?"
"Remove the voice agent from my website widget."
You get back: The current widget state in plain English, a list of voice agents you can attach (when relevant), and, after you approve, confirmation plus a link to the agent's settings.
Writes data? Yes, via preview then commit. Shows the change and asks "Apply this change? Yes or No". Detaching when no widget exists is a safe no-op.
Behind the scenes: get_widget_config, list_voice_agents_for_widget to read state; preview_update_widget_voice_agent and commit_update_widget_voice_agent to attach or replace; preview_remove_widget_voice_agent and commit_remove_widget_voice_agent to detach.
connect-whatsapp-text-agent
What it does: Shows you which WhatsApp business lines are registered in your workspace and which text agents they are connected to, and walks you through what is needed before you finish hooking a new WhatsApp number up to an agent. The final connect step happens in the Dapta admin panel.
Use it when: You want to see your WhatsApp lines, check whether an agent already has WhatsApp connected, find which agent owns a number, or get ready to connect a new number.
Try saying:
"Which WhatsApp lines does my workspace have?"
"Is my Soporte agent connected to WhatsApp, and to which number?"
"I want to hook a new WhatsApp number up to my Ventas agent, what do I need?"
You get back: A table of your WhatsApp lines (number, profile name, linked agent, conversation counts) plus channel usage, the agent's current channels, the prerequisites for connecting (Meta Business Suite access, a fresh phone number, a verified Meta account), and a direct link to the agent's channels page to finish setup.
Writes data? Mostly read-only. It cannot register or detach WhatsApp lines (done in the admin panel), but it can edit the agent itself (for example, rename) via preview then commit.
Behind the scenes: get_workspace_whatsapp_lines, search_text_agents_by_name, get_text_agent_details, and preview_update_text_agent and commit_update_text_agent for any agent edits.
text-prompts
What it does: Lets you read and understand your text agent's current setup: its system prompt, the company information it knows, and its configured persona and language, without changing anything.
Use it when: You want to inspect what an agent currently does before deciding whether to change it.
Try saying:
"Show me my agent's system prompt."
"What does my chat agent know about my company?"
"What language and persona is my support agent set up with?"
You get back: The agent's system prompt in a copy-friendly block, its company name and description, and its persona fields (identity name, purpose, language, key instructions). If there is no linked company or persona record, it tells you plainly.
Writes data? No, read-only. To change the prompt, Claude routes you to text-prompt-improver or text-agent-calibrator.
Behind the scenes: get_text_agent_details, get_text_agent_company_context, get_text_agent_context.
text-prompt-improver
What it does: Rewrites or refines your text agent's system prompt to address what you want improved, then applies it after you approve.
Use it when: You want to improve, rewrite, or update the instructions a chat or text agent runs on, for example fix its tone, tighten objectives, or add new rules.
Try saying:
"Rewrite my support agent's prompt to be more concise and friendly."
"Improve my chat agent so it always asks for an order number first."
"Update my agent's instructions to mention our new return policy."
You get back: A proposed new prompt shown as a before-and-after diff (with length change and preview), and after you confirm, the updated prompt applied plus a link to its settings page.
Writes data? Yes, via preview then commit. Shows the diff and waits for your yes. Older v1 legacy agents cannot be updated through this tool.
Behind the scenes: get_text_agent_details, get_text_agent_context, then preview_update_text_agent and commit_update_text_agent.
Last updated