Using Saas Dev

saas-dev Methodology Bootstrap

You have the saas-dev v4.1.0 enterprise SaaS scaffolding skill installed. This bootstrap wires all specialist skills into a single autonomous pipeline.

How It Works

Single Feature (Recommended to start)

1. Brainstorm — You describe a feature → Claude uses ask_user_input_v0 to ask design questions → saves saas-dev-spec.md
2. Plan — You say "plan" → Claude breaks spec into 2-5 min tasks → saves saas-dev-plan.md
3. Execute — You say "go" → Claude spawns subagents per task, each with the right specialist skill → progress to saas-dev-progress.md

Complete App from PRDs (End-to-End Automation)

1. Orchestrator — You upload business-prd.md + technical-prd.md (+ designs/ folder) → say "build from PRD"
2. Claude runs orchestrator skill which:
   • Extracts all features from PRDs
   • Creates build order with dependencies
   • FOR each feature: runs brainstorm → plan → execute loop
   • Maintains continuity via CLAUDE.md updates
   • Checkpoints every 3-5 features for code review
   • Delivers complete app ready for deployment

RULE: ask_user_input_v0 for ALL Questions

During brainstorm, Claude must use ask_user_input_v0 (the UI button tool) for every design question. Never ask inline. Group related questions into one ask_user_input_v0 call:

• Scope questions (3 options)
• Data questions (3 options)
• Behaviour questions (3 options)
• Non-functional questions (3 options)

Specialist Skills — Auto-Loaded Where Needed

These skills are NOT always present. They load during execution:

During brainstorm: Claude identifies which specialist skills will be needed and notes them in the spec.

During planning: Task descriptions reference the specialist skill by name, e.g., "Load django-backend-dev for models + serializers."

During execution: Subagent receives the task + the specialist skill content, uses patterns from the skill to write code.

Session Start Checklist

On every new session — run saas-dev-resume FIRST, then CLAUDE.md:

• Step 0: Invoke saas-dev-resume Read SESSION_STATE.md + BUILD_PLAN.md + saas-dev-progress.md Reconstruct exact position. Show resume summary. Auto-resume. If no prior state found → proceed to Step 1 as normal.

On every new session — read CLAUDE.md in this exact order:

• Step 1: Check for central CLAUDE.md

  • ~/.claude/CLAUDE.md (global Claude Code rules)
  • org-standards/CLAUDE.md (monorepo shared standards)
  • If found → read it. These are the org-wide rules. They apply unless repo-level overrides.

• Step 2: Check for repo-level CLAUDE.md

  • [repo-root]/CLAUDE.md
  • If found → read it. Repo-level overrides central where they conflict.
  • If NOT found → run django-project-setup to create it.

• Step 3: Merge context

  • Central rules apply by default
  • Repo-level §7 ADRs override central for this project
  • Your custom instructions in either file are followed precisely

• Read merged CLAUDE.md §1 (schema_version) + §3 (skill_version_used)

• Note architecture decisions from §7 — these constrain all tasks

• If task is a new feature → invoke saas-dev-brainstorm now

Quick Reference

SINGLE FEATURE (you drive each iteration):
  You: "Build an invoicing module"
       ↓
  Claude: uses ask_user_input_v0 → spec → plan → execute
       ↓
  You: "Next feature: auth"
       ↓
  Claude: brainstorm/plan/execute auth
       ↓
  You: "Next feature: payments"
  ... (continue until all features done)

NEW SESSION / CONTEXT RESET:
  You: (just open the project, or say "resume" / "continue")
       ↓
  saas-dev-resume reads SESSION_STATE.md + BUILD_PLAN.md + saas-dev-progress.md
  Reconstructs exact state → shows where things stand → auto-resumes
  No re-explanation needed

COMPLETE APP FROM PRD (end-to-end automation):
  You: "Build my app from PRD. Here's my business + technical PRDs (+ designs/)"
       ↓
  Claude (orchestrator):
    1. Reads PRDs → extracts all features
    2. Builds dependency graph
    3. FOR each feature:
         → brainstorm
         → plan
         → execute
         → update CLAUDE.md
         → commit
    4. Checkpoints every 3-5 features for code review
    5. Delivers complete app ready for deployment

If Claude Drifts Mid-Session

If Claude forgets to use ask_user_input_v0 or loses track of the methodology:

Type: "use saas-dev"

This re-triggers the bootstrap and resets the pipeline.

Token Efficiency

• No code snippets in spec files — narrative only
• No code stubs in plan files — specialist skills provide patterns
• Specialist skills loaded only when needed — not all skills, all the time
• Fresh subagent per task — no accumulated context bloat

This design keeps token consumption 40-50% lower than full-context approaches.