Seed Spec - Initial Specification Generator

You are creating the initial specification for feature: $ARGUMENTS

Step 1: Understand the Feature

Use AskUserQuestion to gather essential information. Ask up to 4 questions per batch, grouped by topic.

Batch 1: Core Purpose

• What is the primary goal of this feature?
• Who are the target users?
• What problem does this solve?

Batch 2: Scope & Boundaries

• What should this feature explicitly NOT do?
• Are there existing features this interacts with?
• What are the key success criteria?

Batch 3: Technical Context (if applicable)

• Are there specific technologies or patterns to use?
• Are there performance or scale requirements?
• Are there security considerations?

Adapt questions based on initial answers - skip irrelevant follow-ups.

Step 2: Quick Codebase Scan

Before generating the spec, do a brief exploration:

1. Check for similar features:

   Glob: **/*{feature_keyword}*
   Grep: related terms in likely directories
   

2. Identify relevant areas:

   • Entry points that might need modification
   • Data models that might be affected
   • Existing patterns to follow

3. Note findings for inclusion in spec

Step 3: Create Directory Structure

mkdir -p docs/specs/{feature_name}

Step 4: Generate Initial Spec

Read the template from: templates/spec-template.md

Generate docs/specs/{feature_name}/spec.md with:

YAML Frontmatter

---
feature: {feature_name}
phase: what-auto
iteration: 0
last_updated: {current ISO timestamp}
convergence:
  questions_stable_count: 0
  open_questions_count: {count from initial questions}
  high_confidence_ratio: 0.0
---

Content Sections

1. Overview: Synthesize user's description into clear problem statement

2. High Confidence: Items directly stated by user or verified in code

   • User-stated requirements
   • Discovered existing patterns to follow

3. Medium Confidence: Inferred from user answers or codebase

   • Likely integrations based on codebase scan
   • Assumed patterns based on project conventions

4. Open Questions: Everything that needs research

   • Technical questions surfaced during questionnaire
   • Integration points that need verification
   • Edge cases mentioned but not detailed

5. Iteration Log: Empty, will be populated by refinement loops

Step 5: Report Completion

Output summary:

• Feature name
• Number of high/medium confidence items
• Number of open questions
• File location

Do NOT proceed to refinement - return control to orchestrator.

Output Format

Return a structured summary:

SEED COMPLETE
Feature: {feature_name}
Location: docs/specs/{feature_name}/spec.md
High Confidence Items: {n}
Medium Confidence Items: {n}
Open Questions: {n}
Ready for: what-auto phase