| name | project-vision |
| description | Captures project vision and philosophy through structured interview, producing comprehensive vision documentation and condensed summary. Use before project-planning for new projects or major architectural work requiring design alignment. Ensures blockers and autonomous development can align with project goals. |
Project Vision Skill
Overview
This skill conducts a structured interview to capture comprehensive project vision, philosophy, and design principles. It produces both detailed documentation and a condensed summary embedded in the project's CLAUDE.md file.
Purpose: Establish a "north star" for decision-making that makes architectural choices obvious and enables autonomous agents to resolve blockers without escalation.
Use this skill BEFORE project-planning for:
- New projects (before any planning or implementation)
- Major architectural changes (requires alignment with vision)
- Projects with recurring HUMAN-TASK escalations (missing decision context)
Output Artifacts
docs/VISION.md - Comprehensive vision document (~2-5k tokens)
CLAUDE.md update - Condensed summary (1 paragraph) with pointer to full doc
- Quality reports - Validation from three review passes
Workflow
Stage 1: Structured Interview
Goal: Extract complete vision through targeted questions across 6 dimensions.
Interview Structure:
1. Problem Space
- What problem does this project solve?
- Who experiences this problem?
- How do they currently solve it (or why can't they)?
- What makes this solution better/different?
2. Users & Use Cases
- Who are the primary users?
- What are the 3-5 core use cases?
- What user behaviors or workflows change?
- What should be easy? What should be hard/impossible?
3. Success Criteria
- What does "done" look like?
- How will we measure success?
- What quality level is required (prototype vs production)?
- What's the timeline/scope?
4. Design Philosophy
- What are the 3-5 guiding principles?
- What trade-offs have been pre-decided?
- When in doubt, optimize for what?
- Any non-negotiable constraints?
5. Anti-goals
- What are we explicitly NOT trying to do?
- What features/complexity should be avoided?
- What use cases are out of scope?
- What would make this project a failure?
6. Technical Constraints
- Platform/language/framework requirements (if any)
- Integration points with existing systems
- Performance/scale requirements
- Deployment constraints
- Security/compliance requirements
Interview Process:
- Ask all questions in each section
- Probe ambiguous answers with follow-ups
- Request examples when answers are abstract
- Clarify contradictions immediately
- Continue until you can articulate the vision independently
Output: Structured notes covering all 6 dimensions (temp file for next stage)
Stage 2: Draft Vision Document
Goal: Transform interview notes into comprehensive, well-structured vision document.
Deploy Vision Writer sub-agent:
Write comprehensive project vision document.
Context:
- Interview notes: /tmp/vision-[ts]/interview-notes.md
- Project directory: [path]
Output: /tmp/vision-[ts]/VISION-DRAFT.md
Instructions:
1. Read interview notes thoroughly
2. Produce vision document with this structure:
# Project Vision: [Project Name]
## Problem Statement
[1-2 paragraphs: problem, who has it, current solutions, why this is better]
## Users & Use Cases
### Primary Users
[Description of user personas]
### Core Use Cases
1. [Use case 1]: [Description]
2. [Use case 2]: [Description]
...
## Success Criteria
### Definition of Done
[What "complete" means for this project]
### Quality Bar
[Production-ready? Prototype? Research vehicle?]
### Metrics
[How we measure success]
## Design Philosophy
### Guiding Principles
1. **[Principle 1]**: [Why this matters, what it means in practice]
2. **[Principle 2]**: [Explanation]
...
### Key Trade-offs
[Pre-decided trade-offs that resolve common decision points]
### Optimization Target
[When in doubt, we optimize for: X over Y]
## Anti-goals
[Explicitly out of scope - what NOT to do]
- [Anti-goal 1]: [Why not]
- [Anti-goal 2]: [Why not]
## Technical Constraints
[Platform, integration, performance, deployment constraints]
## Decision Framework
[How to make decisions aligned with this vision]
- If [scenario], choose [approach] because [principle]
- Example decision patterns based on philosophy
3. Write clearly and specifically (not vague platitudes)
4. Include concrete examples for abstract principles
5. Make trade-offs explicit (not implicit)
6. Length: 2-5k tokens (comprehensive but not exhaustive)
Return: Summary + file path
Stage 3: Multi-Angle Review
Goal: Validate vision quality through three independent review passes.
Deploy 3 concurrent review sub-agents:
3.1 Self-Consistency Reviewer
Review vision for internal consistency.
Context:
- Vision draft: /tmp/vision-[ts]/VISION-DRAFT.md
Output: /tmp/vision-[ts]/REVIEW-CONSISTENCY.md
Instructions:
1. Read vision document thoroughly
2. Check for contradictions:
- Do principles conflict with each other?
- Do success criteria align with problem statement?
- Do anti-goals contradict stated use cases?
- Do technical constraints conflict with philosophy?
3. Check for completeness:
- Every section substantive (not placeholder)?
- Principles actionable (not vague)?
- Trade-offs explicit?
4. Flag ambiguities:
- Terms used inconsistently?
- Principles too abstract to apply?
- Success criteria not measurable?
Produce report with:
- Contradictions found (with line references)
- Completeness gaps
- Ambiguities requiring clarification
- Recommendation: APPROVED / NEEDS_REVISION
Return: Summary + file path + status
3.2 Gap Analyzer
Identify gaps in vision coverage.
Context:
- Vision draft: /tmp/vision-[ts]/VISION-DRAFT.md
Output: /tmp/vision-[ts]/REVIEW-GAPS.md
Instructions:
1. Read vision document
2. Generate follow-up questions for major gaps:
- Missing user personas or use cases?
- Unclear success metrics?
- Principles without examples?
- Important trade-offs not addressed?
- Technical constraints under-specified?
3. Classify gaps:
- CRITICAL: Blocks decision-making
- IMPORTANT: Would improve clarity
- NICE_TO_HAVE: Optional enhancement
Produce report with:
- Follow-up questions by category
- Rationale for each question
- Impact if left unresolved
- Recommendation: APPROVED / NEEDS_INTERVIEW
Return: Summary + file path + status
3.3 PRIME DIRECTIVE Reviewer
Review vision for adherence to PRIME DIRECTIVE.
Context:
- Vision draft: /tmp/vision-[ts]/VISION-DRAFT.md
- PRIME DIRECTIVE: Read the project CLAUDE.md (Maximal Simplicity Policy)
Output: /tmp/vision-[ts]/REVIEW-PRIME-DIRECTIVE.md
Instructions:
1. Read the project CLAUDE.md PRIME DIRECTIVE section
2. Read vision document
3. Evaluate alignment:
- Does philosophy emphasize simplicity over complexity?
- Are principles about separation of concerns vs complecting?
- Do trade-offs favor simple over clever?
- Does decision framework guide toward simplicity?
4. Check for red flags:
- Encouraging premature abstraction?
- Conflating concerns in design principles?
- Optimizing for "flexibility" over directness?
- Framework-style patterns over simple solutions?
Produce report with:
- Alignment score: STRONG / MODERATE / WEAK
- Areas of strong alignment
- Areas of concern
- Suggested principle additions/modifications
- Recommendation: APPROVED / NEEDS_REVISION
Return: Summary + file path + status
Parallelism: Deploy all reviewers in single message via multiple Task calls.
Gate Logic (based on sub-agent reports):
- All APPROVED → Proceed to Stage 4
- Any NEEDS_REVISION → Make revisions → Re-review
- NEEDS_INTERVIEW → Conduct follow-up interview → Update draft → Re-review
Stage 4: Finalize & Integrate
Goal: Write final vision document and embed summary in CLAUDE.md.
Process:
4.1 Write Final Vision Document
4.2 Generate Condensed Summary
Create 1-paragraph summary for CLAUDE.md:
- Problem (1 sentence)
- Solution approach (1 sentence)
- Key philosophy (1 sentence)
- Optimization target (phrase)
- Pointer to full doc
Example:
## Project Vision
my-project is a lightweight task runner emphasizing simplicity and composability. Core philosophy: prefer simple, separated concerns over clever abstractions; optimize for debuggability and comprehension over performance; make the straightforward path obvious. When in doubt, choose the approach with fewer moving parts. See [docs/VISION.md](docs/VISION.md) for complete vision and decision framework.
4.3 Update CLAUDE.md
If CLAUDE.md exists in project:
- Add
## Project Vision section after project description
- Include condensed summary
- Pointer to docs/VISION.md
If no CLAUDE.md:
- Create minimal CLAUDE.md with vision section
- Note this in completion report for user review
4.4 Commit Vision Documents
git add docs/VISION.md CLAUDE.md
git commit -m "docs: capture project vision and philosophy
- Comprehensive vision document (problem, users, philosophy, constraints)
- Decision framework for autonomous development
- Embedded summary in CLAUDE.md"
Stage 5: Completion Report
Report to user:
- Vision document location:
docs/VISION.md
- CLAUDE.md updated with summary
- Review outcomes (consistency, gaps, PRIME DIRECTIVE alignment)
- Commit hash
- Next step: Ready for project-planning with vision alignment
Integration with Other Skills
project-planning
Research sub-agents should:
- Read
CLAUDE.md (gets condensed vision automatically)
- Read
docs/VISION.md for detailed context
- Evaluate technical approaches against design philosophy
investigating-blocker
Prior art research synthesis should:
- Check industry approaches against project philosophy
- Recommend solutions aligned with optimization targets
- Flag when prior art conflicts with vision (escalate vs auto-resolve)
reviewing-plans
Plan auditors should:
- Verify features align with use cases
- Check approaches match design principles
- Validate decisions follow trade-off guidance
autonomous-development
Implementation agents should:
- Reference philosophy when choosing between approaches
- Flag implementation choices that violate principles
- Use decision framework for ambiguous cases
Key Principles
- Interview before drafting: Don't guess vision, extract it comprehensively
- Multi-angle review: Catch contradictions, gaps, and anti-patterns early
- Concrete over abstract: Principles must be actionable with examples
- PRIME DIRECTIVE alignment: Vision should amplify simplicity policy
- Dual output: Detailed doc for deep reference, summary for quick context
- Decision-enabling: Vision should resolve common architectural questions
Troubleshooting
Interview yields vague answers
- Ask for concrete examples
- Request counter-examples (what NOT to do)
- Probe trade-offs (if X vs Y, which and why?)
- Ground in specific use cases
Vision document too generic
- Add concrete examples for each principle
- Make trade-offs explicit with scenarios
- Include decision patterns based on actual choices
- Reference specific project constraints
PRIME DIRECTIVE conflict detected
- Re-interview on simplicity philosophy
- Add/strengthen simplicity principles
- Make "simple over clever" explicit in trade-offs
- Add anti-goals against premature abstraction
Reviewers find critical gaps
- Conduct targeted follow-up interview
- Don't guess - ask user directly
- Update draft comprehensively
- Re-run full review cycle
Context Management
Main agent:
- Conducts interview directly (only hands-on work)
- Dispatches sub-agents with file references
- Routes review feedback to revision
Sub-agents:
- Return summary + file path (not full content)
- Keep prompts under 500 tokens
- Use temp directory for coordination
Temp structure:
/tmp/vision-[timestamp]/
├── interview-notes.md # Structured interview output
├── VISION-DRAFT.md # Initial draft
├── REVIEW-*.md # Review reports
└── VISION-FINAL.md # Revised version (if needed)
Success Criteria
Vision document is complete when:
Example Vision Section for CLAUDE.md
## Project Vision
my-saas is a B2B analytics platform for small businesses, prioritizing rapid iteration and customer learning over scale. Core philosophy: ship early, learn from users, consolidate based on evidence; optimize for time-to-insight over performance; embrace manual processes until automation pays for itself. Pre-alpha risk tolerance allows breaking changes. When facing build-vs-buy, prefer battle-tested external services over custom code. See [docs/VISION.md](docs/VISION.md) for complete vision, use cases, and decision framework.
This gives autonomous agents immediate context while keeping CLAUDE.md scannable.