project-discovery

Project discovery

Interrogate every answer. "PostgreSQL" is the start of a question, not the end of one: self-hosted or managed, what volume, what compliance, what is the backup story. The second question is where the real constraint surfaces.

Every recommendation ships with its trade-off. Say what they gain and what they give up. When a choice fights their constraints, say so and name the cheaper path.

Discovery categories

1. Project overview

Question	Why It Matters
What does this project do? Who is it for?	Frames all technical decisions
What problem are we solving? What's painful now?	Ensures we're solving real problems
How will we know if it's successful?	Defines done, prevents scope creep
What's the timeline? MVP vs full launch?	Determines build vs buy, complexity tolerance
Who's building this? Solo, small team, org?	Affects architecture, tooling choices
What's the budget for infrastructure/services?	Managed vs self-hosted decisions

2-10. Every other layer

When the interview reaches a layer the user has not already decided, read ${CLAUDE_SKILL_DIR}/references/stack-menu.md and use that layer's option list and separating questions. This covers framework, database, ORM, auth, API style, styling, hosting, integrations, and testing. Do not name options from memory. The file exists so the trade-off you quote is the one you quote next time.

Conducting discovery

Opening

"Before we write any code, I want to understand what we're building and make sure we make good foundational decisions. This might take 15-30 minutes but will save hours later. Let's start with the big picture. What are we building and who is it for?"

During discovery

• Group related decisions: "So for the data layer, we're going with Postgres on Supabase with Drizzle ORM. That gives us..."
• Flag decisions that need more thought: "Let's come back to the auth flow once we understand the user types better"
• Challenge weak reasoning: "You said 'because everyone uses it'. Is that the right reason for your specific case?"

Closing discovery

"Let me summarize what we've decided:

Project: [One sentence]
Stack: [Frontend] + [Backend] + [Database]
Auth: [Approach]
Hosting: [Where]
Key patterns: [2-3 important conventions]

Features for MVP:
1. [Feature]
2. [Feature]
...

Does this capture it? Anything we should revisit before we start building?"

What to produce

1. Read ${CLAUDE_SKILL_DIR}/references/output-templates.md and write CLAUDE.md, the folder skeleton, and the config files it lists.

2. The rule corpus is ${CLAUDE_SKILL_DIR}/references/coding-standards.md. You need it once, when writing PATTERNS.md. Read it then, copy across only the rules the chosen stack actually uses, and delete the rest. A rule for a library the project does not have is a rule the agent will misapply.

3. Copy all four files in ${CLAUDE_SKILL_DIR}/references/lib-files.md into lib/ verbatim. Every downstream rule in the rule corpus assumes these exact signatures.

The run is done when CLAUDE.md names a decision for every category in ${CLAUDE_SKILL_DIR}/references/stack-menu.md, including categories decided as "not needed", with the reason. PATTERNS.md exists and contains only rules the chosen stack uses. All four files from ${CLAUDE_SKILL_DIR}/references/lib-files.md are in lib/. An unnamed category is an undiscovered one. Go back and ask.