requirements-discovery

Requirements Discovery

Transform a vague feature description into a precise, machine-executable requirements document. Every section of the output maps 1:1 to a code artifact.

When to Use

• Starting a new microservice or handler
• Adding new database tables or message events
• Any feature that touches multiple services
• When someone says "I want to build [X]"

Primary Goal

Zero ambiguity in the output. The requirements doc should be so precise that autonomous development can execute it without asking questions.

Step 0: Load Context

Before asking any questions:

1. Read the project's CLAUDE.md for architecture patterns and conventions
2. Read docs/service-map.md (or equivalent) for existing services and responsibilities
3. Read REQUIREMENTS-TEMPLATE.md in this skill's directory for the output format
4. Understand existing EVENT_TYPES, database tables, and service boundaries

Discovery Rounds

Round 1: Domain & Scope

Ask ALL of these. Do NOT proceed until answered:

1. What problem does this solve? (1-2 sentences, business context)
2. Who or what triggers this? (User action / scheduled / event from another service / external webhook)
3. What is the expected output/effect? (Data created / event published / state changed / notification sent)
4. Which existing service is this most similar to? (This becomes the reference implementation)
5. Which repository does this belong in? (Suggest one based on the description, confirm with user)

After answers: Summarize your understanding. Ask "Is this accurate?"

Round 2: Data & Integration

1. What data does this need to READ? (Which tables, which fields)
2. What NEW data does this create? (New tables? New columns?)
3. Which existing events does it CONSUME? (From the event catalog)
4. What NEW events should it PUBLISH? (Name them, describe payload structure)
5. External API integrations? (Remind: one integration point per external service)
6. New database access methods needed? (Check what already exists)