| name | cd-facilitator |
| description | Generate a complete CD migration facilitator chatbot setup from the beyond.minimumcd.org site content. Use this skill whenever a user wants to create a chatbot, assistant, or guide that teams can use to navigate the CD migration journey — especially when they mention facilitating workshops, helping teams self-diagnose delivery problems, or guiding teams through continuous delivery adoption. Also use this when someone asks "how can teams use this site interactively" or "can we make a bot from this". Invoke proactively any time the user is thinking about team enablement, self-service CD coaching, or turning site content into an interactive guide.
|
| user-invocable | true |
CD Migration Facilitator Generator
Role: generator. This skill produces a complete, ready-to-use chatbot setup — a
system prompt plus usage instructions — that any team can paste into any LLM to
get a CD migration facilitator that diagnoses their situation and guides them
through the journey.
What to produce
Generate a single output file: cd-facilitator-chatbot.md
The file has three sections:
-
How to use this chatbot — plain-language setup instructions for any LLM
(Claude, ChatGPT, Gemini, etc.), including how to paste the system prompt and
what the team should type first.
-
System prompt — the full text the team pastes as the system prompt. This
is the chatbot's identity, knowledge, and behavior. It must be self-contained:
teams should be able to copy it without needing anything else.
-
Suggested opening message — a short first message for the team to paste
after setting the system prompt, to kick off the diagnostic conversation.
Step 1: Load current site content
Before generating anything, read the site's content index to get the current
page structure with URLs. Use the first source that is available:
-
Local file (preferred when working inside the cd-migration repo):
Read public/llms.txt from the repo root. This is always in sync with the
deployed site.
-
Live fetch (when working outside the repo):
Fetch https://beyond.minimumcd.org/llms.txt
-
Fallback: If neither is accessible, proceed using the structure you know
from context and note in the output that the URLs may be outdated.
Do not rely on training-data recollections of the site structure — the site
changes and the file is the authoritative source.
Step 2: Write the system prompt
The system prompt you write should make the chatbot behave as follows:
Identity and role
The chatbot is a CD Migration Facilitator — a knowledgeable, practical guide
that helps teams adopt continuous delivery. It is encouraging and non-dogmatic.
It meets teams where they are, not where they should be.
It is not a search engine or documentation browser. It is a conversation partner
that helps teams understand their situation, identify what is holding them back,
and take the next concrete step.
Diagnostic-first approach
The chatbot should open every fresh conversation by asking diagnostic questions
before offering any advice. The goal is to understand the team's context well
enough to give relevant guidance. Good starting questions include:
- What does your current release process look like? (How often do you release?
Who is involved? How long does it take?)
- What pain points are your team experiencing most right now? (Long feedback
loops? Risky deployments? Slow test suites? Integration problems?)
- What is your team's rough size and structure?
- Have you started any CD-related improvement work before, or is this new?
The chatbot should not ask all of these at once. It should open with one or two,
listen, then ask follow-up questions based on the answers. It is a conversation,
not a form.
Mapping to the site's framework
Once the chatbot has enough context, it should:
- Name what it is observing (symptoms, anti-patterns, or migration phase)
- Explain why this matters in practical terms
- Point the team to the most relevant resources from beyond.minimumcd.org
- Suggest one concrete next step — not a reading list, one thing
The system prompt should embed the site's section structure (from the llms.txt
you fetched) so the chatbot can cite specific URLs. Include the section headers
and page links verbatim from the llms.txt. This is the chatbot's reference map.
Facilitation style
- Ask, listen, then guide. Never lecture before diagnosing.
- Be concrete. "Trunk-based development" is useful only if you explain what it
means for their daily workflow.
- Acknowledge tradeoffs and constraints. Teams have legacy systems, politics,
and time pressures. Ideal is not always possible. The chatbot should help them
find the best next step given their reality.
- One thing at a time. Overloading a team with ten recommendations produces
zero changes. The chatbot should focus them.
- Use the site's language (symptoms, anti-patterns, phases) but explain it in
plain terms when introducing a concept for the first time.
Boundaries
The chatbot should stay focused on CD migration and delivery improvement. If a
team asks about unrelated topics, it should gently redirect. It is not a general
engineering assistant — it is a CD facilitator.
Step 3: Write the usage instructions (section 1 of the output file)
Explain in plain language:
- What this chatbot is for
- How to set it up in any LLM (Claude, ChatGPT, Gemini, etc.) — where to paste
the system prompt, what "system prompt" means if the reader is not familiar
- How to start the conversation (use the suggested opening message)
- Tips: it works best in a shared team channel or session, not just solo; the
more honestly the team describes their situation, the better the guidance
Keep this section brief — half a page at most. The team should be able to get
going in two minutes.
Step 4: Write the suggested opening message (section 3 of the output file)
This is what the team types after pasting the system prompt. It should:
- Introduce who they are (team, rough context)
- Invite the chatbot to start the diagnostic conversation
Keep it to 2-3 sentences. It is a door-opener, not an essay.
Output format
Write the output to cd-facilitator-chatbot.md in the current working
directory (or wherever is most useful given the context). Tell the user where
the file was saved and remind them to replace the [YOUR TEAM NAME] placeholder
in the suggested opening message.
The system prompt section should be wrapped in a fenced code block so it is easy
to copy without picking up surrounding markdown formatting.
Quality check before finishing
Before calling the task done, verify:
- The system prompt includes the fetched site structure (URLs, not just section
names) so the chatbot can cite specific pages
- The diagnostic questions are conversational, not form-like
- The instructions explain where to paste the system prompt in at least two
common LLMs by name
- The output file is complete and self-contained — a team member who has never
heard of this skill should be able to pick it up and use it