Appie Article Writer

Overview

Generate articles in a chosen author voice via Appie.run Content Writer.

• light — ~500 words, 100 credits
• deeply_researched — ~3000 words, 500 credits

Generation is asynchronous. Always poll status. Never invent article content — only return what the API provides.

Prerequisites (always run first)

This skill needs the Appie.run MCP server connected to your agent.

1. Check whether the Appie tools (e.g. create_article, list_authors) are already available. If they are, continue.
2. If not, ask the user to connect the Appie MCP server:
   • URL: https://appie.run/mcp (transport: streamable HTTP)
   • Header: Authorization: Bearer <esk_...> (your Appie.run project API key, starting with esk_)
   • Setup docs for each client: https://appie.run/docs/mcp
3. New Appie.run accounts get 500 free credits. Never ask the user to paste their key into chat, and never invent one. Stop and wait until the tools are available before continuing.

Tool Reference

Action	Tool name	Cost	Notes
List authors	list_authors	free	Check readiness
Create author	create_author	free	bio ≥50 chars
Add author sample	add_author_sample	free	content ≥100 chars or blog_url
Create article	create_article	100 / 500	charged on generation
Article status	article_status	free	poll until ready
Get brain	get_brain	free	brand context (auto-used)
Update brain	update_brain	free	only if user asks

create_article parameters

Note: create_article fails with author_profile_required until an author exists. Create one with create_author first (or check with list_authors).

• topic (required) — clear description of the article subject
• article_type — "light" (default) or "deeply_researched"
• author_id (optional integer) — omit to use the project's active author

create_author parameters

• author_name (required)
• bio (required, ≥50 characters)
• is_active (optional boolean, default true)
• sample_title / sample_content (optional, sample_content ≥100 chars)

Full Workflow

1. Clarify type and topic

• If the user says only "write an article", ask whether they want light (~500 words, 100 credits) or deeply_researched (~3000 words, 500 credits).
• Confirm the exact topic if ambiguous.

2. Ensure a usable author

1. Call list_authors.
2. Prefer an author marked ready / active.
3. If none exist or the desired voice is missing:
   • Create one with create_author (bio ≥50 chars).
   • Optionally add a writing sample (≥100 chars) via add_author_sample so the voice is stronger.
4. If an author exists but is not ready, add a sample and re-check.

3. Start generation

1. Call create_article with topic, chosen article_type, and optional author_id.
2. Immediately tell the user:
   • The article id
   • Expected cost
   • That generation is running in the background
3. Capture the returned id for polling.

4. Poll until ready

1. Call article_status with the id.
2. Repeat every few turns or when the user asks "is it ready?" until content appears or status is failed/complete.
3. On success, return the full article text, title (if provided), id, and type.
4. On failure, surface the exact error message and stop.

5. Optional Brain context

• Call get_brain if the user wants to inspect brand voice, ICP, competitors, or key messages.
• Only call update_brain when the user explicitly requests brand or ICP changes.
• Content Writer already uses Brain context automatically; you do not need to inject it.

Error Handling & Edge Cases

• create_article fails with author_profile_required until the project has an author → check with list_authors, create one with create_author (bio ≥50 chars), then retry.
• Author not ready → add sample (≥100 chars) then retry create_article.
• Insufficient credits → report the exact cost (100 or 500) and stop. Do not start generation.
• Ambiguous topic → ask one clarifying question before calling create_article.
• Multiple authors → list them and let the user choose, or use the active one if the user does not care.
• Generation stuck / failed → show the status payload and offer to start a new article with a refined topic.
• Never invent or paraphrase the article body. Quote or present the exact content returned by article_status.
• If the user later wants the same article fed into a chatbot, note the article_id for the appie-chatbot skill.

Response Style

• State cost and async nature up front.
• After creation, give the id and offer to check status.
• When ready, deliver the full content cleanly (preserve headings and structure).
• Keep follow-up questions minimal and focused.