Vision Document Workflow

Guide the user through creating planning/VISION.md.

Context

Skill to load: Load the katachi:framework-core skill for workflow principles.

Vision document: @planning/VISION.md - if it exists

General Guidance

Follow the collaborative workflow principles from the framework-core skill.

Vision-specific guidance:

Use a scratchpad - Track state in /tmp/vision-state.md:

• Current section being worked on
• Questions asked and answered
• Gaps identified
• Topics to revisit

Process

0. Check Existing State

If planning/VISION.md exists:

• Read current vision document
• Read project state (features, specs, implementation progress if any)
• Summarize current state to user
• Ask: "What aspects need refinement? Or should we review the whole vision?"
• Enter iteration mode as appropriate

If no vision exists: proceed with initial creation

1. Understand the Problem

• What problem are you trying to solve?
• Who experiences this problem?
• How do they currently deal with it?

2. Research Existing Solutions

Use Task tool (general-purpose agent) to research:

• Existing solutions (commercial and open source)
• Technical approaches (models, frameworks, libraries)
• Known issues with alternatives
• Best practices and patterns

Synthesize findings to inform questions.

3. Define Core Workflows

• What are the main things a user will do with this software?
• For each workflow:
  • What triggers this workflow?
  • What's the end result?
  • What are the key steps?

4. Set Scope Boundaries

• What MUST be in v1? (included)
• What is explicitly NOT in v1? (excluded)
• Challenge scope creep: "Is this really necessary for v1?"

5. Technical Context

• Where does this run? (platform)
• How do users interact with it?
• What external systems does it connect to?
• DETECT: Platform/language/storage choices → prompt for ADR

6. Write the Document

• Show draft to user
• Ask for feedback
• Iterate until approved

7. External Validation

Dispatch a general-purpose subagent to review the completed vision.

Request structured critique covering:

• Clarity: Is the problem, solution, and scope clearly articulated?
• Completeness: Are workflows fully specified? Is technical context sufficient?
• Internal consistency: Do all sections align? Do workflows solve the problem?
• Unstated assumptions: What's implied but not explicit?
• Scope boundaries: Is v1 scope realistic? Are exclusions clear?
• Gaps: Between problem/workflows, workflows/scope, scope/technical context
• Edge cases: What scenarios aren't addressed?

Review subagent findings with user. Ask: "Should we iterate on any section based on this feedback, or is the vision complete?"

Decision Detection

When user mentions hard-to-change choices, offer to create ADRs:

• Platform choice → "Should we create an ADR for platform?"
• Language choice → "Should we create an ADR for language?"
• Storage approach → "Should we create an ADR for storage?"
• Model/library choice → "Should we create an ADR for this choice?"

Workflow

This is a collaborative process:

• Ask one question at a time
• Agent proposes, user confirms
• User makes all decisions
• Provide research-backed alternatives
• Never fill gaps yourself
• Iterate until approved