| name | new-stack-search |
| description | Create a new stack search workspace from the Claude-Stack-Research-Workspace template. Clones the template, creates a GitHub repo, and launches the structured intake interview. Use when the user wants to research software options, compare tools, evaluate vendors, or answer 'does X exist?'. Triggers on phrases like 'new stack search', 'find me a tool for', 'compare options for', 'what tools exist for', 'stack search', 'evaluate tools for'. |
New Stack Search
Spawn a new stack search workspace from the danielrosehill/Claude-Stack-Research-Workspace template and immediately start the structured requirements intake.
Steps
Step 1: Get the topic
Ask the user: "What are you trying to find or evaluate? Give me a short name for this search."
Examples:
- "vector database for my app"
- "email API service"
- "self-hosted note app"
- "alternative to Notion"
Step 2: Derive the repo name
Convert the topic into a title-case repo name with the pattern: Stack-Search-{Topic}
Examples:
- "vector database" ->
Stack-Search-Vector-Database
- "email API service" ->
Stack-Search-Email-API
- "alternative to Notion" ->
Stack-Search-Notion-Alternative
Confirm with the user: "I'll create the repo as {name}. Good?"
Step 3: Ask visibility
Ask: "Public or private repo?" Default to private if the user doesn't care.
Step 4: Clone the template
cd ~/repos/github/my-repos
gh repo create danielrosehill/{repo-name} --template danielrosehill/Claude-Stack-Research-Workspace --clone --{public|private}
cd {repo-name}
If gh repo create --template fails (e.g., template not marked as GitHub template), fall back to:
cd ~/repos/github/my-repos
git clone https://github.com/danielrosehill/Claude-Stack-Research-Workspace.git {repo-name}
cd {repo-name}
rm -rf .git
git init && git branch -m main
gh repo create danielrosehill/{repo-name} --source . --push --{public|private}
Step 5: Clean up template placeholders
- Replace the placeholder text in
CONTEXT.md with the user's topic name (fill in "What is being evaluated" with what they told you in Step 1).
- Delete the
.gitkeep from context/from-human/ since we'll be writing a real file there shortly.
Step 6: Use Tavily for initial landscape scan
Before starting the intake interview, do a quick landscape scan using Tavily MCP (if available) to prime the conversation:
tavily_search: "{topic} best tools comparison {current year}"
tavily_search: "{topic} open source vs SaaS options"
Save any useful findings to context/from-internet/initial-landscape-scan.md with dated source links. This gives the intake interview a head start — the user can reference tools that came up in the scan.
If Tavily MCP is not available, skip this step silently and proceed.
Step 7: Run the intake interview
Now run through the /define-stack interview questions inline. Do NOT tell the user to run /define-stack themselves — you are the launcher, so run the interview here. Ask the 27 questions from the define-stack command, one section at a time:
Section 1: The Problem (questions 1-4)
- "What are you trying to do? Describe the problem or need in a sentence or two."
- "Is this a new need, or are you replacing something that isn't working?"
- "Have you tried any existing solutions for this? If so, which ones?"
- (If they tried something) "What specifically didn't work about them? What was the gap or frustration?"
Section 2: Deployment Model (questions 5-7)
5. "Are you open to SaaS (vendor-hosted), self-hosted, or either?"
6. (If self-hosted in scope) "Preferred deployment method — Docker, native package, or no preference?"
7. (If self-hosted in scope) "Where would you host it — home server, VPS, cloud VM?"
Section 3: Budget (questions 8-10)
8. "What's your budget band?" — Free/OSS only, Low (up to $20/mo), Mid ($20-200/mo), High ($200+/mo), Enterprise/negotiated
9. "Is there a one-time setup budget, or just ongoing costs?"
10. "Is transparent pricing important, or are you okay with 'contact sales' vendors?"
Section 4: Who and Scale (questions 11-13)
11. "Who is this for — personal use, your business, a client, or a side project?"
12. "Any scale context? Number of users, data volume, requests per second?"
13. (If client) "Is the client's budget different from your usual range?"
Section 5: Hard Requirements (questions 14-16)
14. "What are the absolute must-haves? If a tool is missing any one of these, it's out."
15. "Let me push on specifics — can you make each must-have concrete?" (e.g., "Must have a REST API" not "good integrations")
16. "Any technical constraints that would rule things out — specific language, framework, protocol, platform?"
Section 6: Nice-to-Haves (questions 17-18)
17. "What would be nice but isn't a deal-breaker?"
18. "If two tools tie on must-haves, what would make you pick one over the other?"
Section 7: Integrations (questions 19-20)
19. "Does this need to integrate with anything specific? Google Workspace, GitHub, Slack, APIs, webhooks?"
20. "Is API access (REST/GraphQL) important for building your own integrations?"
Section 8: Data Portability (question 21)
21. "How important is data portability? Must have automated backup via API / periodic manual export is fine / not important here."
Section 9: Prior Research (questions 22-25)
22. "Have you done any prior research? Blog posts, Reddit threads, recommendations?"
If you did a Tavily landscape scan in Step 6, share what you found here: "I did a quick scan and found these names coming up: [list]. Any of these ring a bell or worth including?"
- "Any specific tools you already want me to evaluate?"
- "Any tools to exclude — ones you've tried, ruled out, or just don't want?"
- "Do you know of any category leaders in this space?"
Section 10: Wrap-Up (questions 26-27)
26. "Anything else — constraints, preferences, timeline, context?"
27. "How urgent is this? Browsing vs need to pick something this week?"
Step 8: Write the requirements file
Write context/from-human/stack-requirements.md from the interview answers, using the format from context/from-human/stack-requirements-template.md.
Update CONTEXT.md with a summary.
Step 9: Commit and push
git add -A
git commit -m "Initialize stack search: {topic}"
git push
Step 10: Report
Tell the user:
- Repo created at
~/repos/github/my-repos/{repo-name}
- GitHub URL
- Requirements captured
- "Open a Claude session in that repo and run
/start-research to begin, or I can start researching right now if you'd like."
Style
- This is a launcher — keep it moving. Don't over-explain between steps.
- If the user gives a big freeform dump at any point, parse it into the structured fields and skip answered questions.
- The Tavily landscape scan is a bonus, not a blocker. If it fails or the MCP isn't available, proceed without it.