You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
# Navigate to projectcd /path/to/project
# Start Claude Code
claude --dangerously-skip-permissions
# Then prompt:
Read docs/phases/PHASE_INDEX.md and execute the next incomplete phase
Architecture Reference
[Paste or summarize the architecture from PHASE_PLAN.md]
[Diagram if applicable]
Phase Dependencies
Phase 01
↓
Phase 02
↓
Phase 03 ──→ Phase 04
Session Management
Finding Session UUIDs
During a session:
ls ~/.claude/tasks/ # Active sessions have folders here
After a session:
ls -lt ~/.claude/projects/[PROJECT]/ # Sort by date, find .jsonl
Resuming Work
# Resume specific session
claude --resume {UUID}
# Or start fresh - Claude Code sees existing code and continues
claude --dangerously-skip-permissions
Discoveries
Capture learnings, surprises, and decisions made during execution.
This skill enables Claude Code to execute SAPA Phase Briefs systematically. It bridges the gap between planning (in Claude Web) and building (in Claude Code).
When this skill is available, Claude Code can:
Read and expand Phase Briefs into detailed execution steps
Track progress in PHASE_INDEX.md
Validate phase completion before proceeding
Capture discoveries that inform future phases
When to Use
Activate this skill when:
User has a docs/PHASE_PLAN.md or docs/phases/ directory from SAPA
User says "execute phase X" or "start phase X"
User asks to "generate execution prompts from the phase plan"
User wants to check phase status or update progress
Core Commands
generate phases
Reads docs/PHASE_PLAN.md (SAPA output) and generates detailed execution prompts in docs/phases/:
Appends a note to the current phase's Notes section in PHASE_INDEX.md. Use for:
Unexpected findings
Architecture decisions made during execution
Corrections to future phases
Phase Expansion Logic
When expanding a SAPA Phase Brief into an execution prompt, include:
1. Header Block
# Phase XX: [Name]> **Status**: ⬜ Not Started> **Estimated effort**: X-Y hours> **Dependencies**: Phase XX (must be complete)
2. Context Section
## Context### What Exists[List files, modules, capabilities from prior phases]### What We're Building[Clear statement of this phase's contribution]### Architecture Reference[Relevant diagram or description from PHASE_PLAN.md]
3. Deliverables Section
## Deliverables### 1. [First Deliverable]**File**: `path/to/file.ext`**Purpose**: [Why this exists]**Implementation**:
[Detailed steps, code scaffolds, or pseudocode]**Validation**:
-[ ][How to verify this works]
4. Technical Constraints
## Technical Constraints-[Stack requirement]-[Pattern to follow]-[Convention from prior phases]
5. Validation Checklist
## Validation Checklist
Before marking this phase complete:
-[ ] All deliverables implemented
-[ ] Code compiles without warnings
-[ ][Feature-specific check]-[ ][Feature-specific check]-[ ] Works on target platform(s)
This document is a methodology for Claude to guide a human from raw idea through structured planning to Claude Code development. It is designed to be dropped into any Claude.ai conversation to invoke a consistent, repeatable process.
When this document is provided, Claude should follow the SAPA methodology before producing any planning artifacts or recommending development.
The SAPA Methodology
S - Spectrum
Before recommending any document or artifact, Claude presents the full spectrum of planning options and explains when each is appropriate:
Document
When to Use
Audience
Napkin sketch
Solo weekend build
Just you
One-pager / Brief
Gut-checking with trusted people
Advisors, co-founders
Lean Canvas
Validating business model
Investors, yourself
Product Brief
Aligning a small team
2-5 person team
PRD
Cross-functional coordination
Engineering, design, leadership
BRD
Enterprise/contractual contexts
Executives, legal, partners
Technical Spec
Detailed implementation guidance
Engineers, Claude Code
Claude does not assume which document is needed. Claude presents the spectrum and helps the human locate themselves on it.
A - Assessment
Claude asks clarifying questions to understand the human's current state:
Validation Status
Is the idea validated? Have real users expressed interest?
Has a prototype been built? Has anyone used it?
Is there evidence of willingness to pay?
Resource Reality
Who is building this? Solo? Small team? Organization?
What's the timeline? Weekend project? Quarter-long initiative?
Is budget/approval required from others?
Risk Identification
What's the biggest unknown? Technical feasibility? Market demand? Business model?
What would make this a waste of time?
What needs to be true for this to succeed?
Intent Clarity
Is this exploration, validation, or committed execution?
Side project, internal pitch, or startup?
Learning exercise or shipping product?
Claude synthesizes these answers into a situational summary before recommending next steps.
P - Prescription
Based on the assessment, Claude recommends the minimum viable documentation needed. Claude should bias toward lighter documentation when:
The builder is solo or small team
The idea is unvalidated
Speed matters more than coordination
The human will be doing the building themselves
Claude should recommend heavier documentation when:
Multiple stakeholders need alignment
Resources must be secured from others
The project has dependencies or handoffs
There's organizational or contractual complexity
Claude then produces ONLY the recommended artifact(s), not the full spectrum.
A - Architecture for Claude Code
Once the human has the planning artifact they need, Claude helps prepare for Claude Code development by producing:
1. Phase Plan
A numbered sequence of development phases, where each phase:
Delivers working, testable functionality
Builds on the previous phase
Can be completed in a single Claude Code session (roughly 1-4 hours of work)
Has clear acceptance criteria
2. Phase Briefs
For each phase, a self-contained brief that can be given to Claude Code containing:
Context: What exists from prior phases
Objective: What this phase accomplishes
Deliverables: Specific files, features, or capabilities
Constraints: Technical requirements, patterns to follow
Acceptance: How to verify the phase is complete
3. Technical Spine
A minimal technical document covering:
Stack decisions (with rationale)
Architecture overview (diagram if helpful)
Key data structures or schemas
External dependencies
Deployment target
The goal is to give Claude Code everything it needs to execute each phase without requiring the human to re-explain context.
Invocation
When a human provides this document along with an idea, Claude should:
Acknowledge the SAPA framework is active
Begin with Spectrum - show the options
Proceed to Assessment - ask the clarifying questions
Deliver Prescription - produce the right artifact
Prepare Architecture - create the Claude Code handoff materials
Claude should not skip steps. The value is in the process surfacing clarity, not in rushing to output.
Artifact Templates
When Claude produces planning artifacts, use these structures:
One-Pager Template
# [Project Name]## Problem[2-3 sentences: what pain exists]## Solution[2-3 sentences: what we're building]## Why Now[Why this is possible/necessary now]## Why Us[Unfair advantage or unique insight]## First Users[Specific description of initial target]## Success Metric[One measurable outcome for v1]## Open Questions[Top 3 unknowns to resolve]
Lean Canvas Template
# Lean Canvas: [Project Name]## Problem[Top 3 problems]## Customer Segments[Target users]## Unique Value Proposition[Single clear message]## Solution[Top 3 features]## Channels[Path to customers]## Revenue Streams[How you make money]## Cost Structure[Fixed and variable costs]## Key Metrics[Numbers that matter]## Unfair Advantage[What can't be copied]
Product Brief Template
# Product Brief: [Project Name]## Overview[What is this and why are we building it]## Target User[Specific persona with context]## User Problem[The pain, in their words]## Proposed Solution[How we solve it]## Key Features (v1)[Bulleted list, prioritized]## Out of Scope (v1)[What we're explicitly not building yet]## Success Criteria[How we know v1 worked]## Risks & Mitigations[What could go wrong, what we'll do]## Open Questions[Decisions still needed]
Phase Plan Template
# Phase Plan: [Project Name]## Technical Stack[Key technology choices]## Architecture Overview[Brief description or diagram]## Phases### Phase 1: [Name]**Objective**: [What this accomplishes]**Deliverables**: [Specific outputs]**Acceptance**: [How to verify complete]### Phase 2: [Name]
...
### Phase N: [Name]
...
## Cross-Phase Concerns[Patterns, conventions, or constraints that apply throughout]
Phase Brief Template (for Claude Code)
# Phase Brief: [Phase Name]## Context[What exists from prior phases - or "greenfield" if Phase 1]## Objective[Single clear goal for this phase]## Deliverables-[ ][Specific file or feature]-[ ][Specific file or feature]-[ ][Specific file or feature]## Technical Constraints[Stack, patterns, conventions to follow]## Dependencies[External services, APIs, packages needed]## Acceptance Criteria[How to verify this phase is complete]## Handoff Notes[Anything the next phase needs to know]
Example Flow
Human: [Provides SAPA document + idea for "Infinite Intern" product]
Claude:
"I see you've invoked the SAPA framework. Let me walk through this properly."
Shows document spectrum, explains each
Asks: "Before I recommend a planning approach, help me understand..."
Are you building this solo or with others?
Is this validated? Have you talked to potential users?
What's your timeline and resource situation?
Is this a side project, potential startup, or something else?
Based on answers, recommends specific artifact (e.g., "Given you're solo and unvalidated, I recommend a one-pager to crystallize the idea, then straight to Phase Plan")
Produces the recommended artifacts
Produces Phase Plan + Phase Briefs ready for Claude Code
Notes for Claude
This framework exists because humans often jump to "write me a PRD" when they need something lighter, or skip planning entirely when they need alignment
The goal is right-sized documentation, not maximum documentation
The Phase Briefs are the critical handoff - they must be self-contained enough that Claude Code can execute without the human re-explaining
Bias toward building over planning when the human is the builder
The framework should feel like a conversation, not a form being filled out
