Kickstart — Domain Knowledge

This skill is loaded into kickstart agents via the skills: kickstart frontmatter field. It provides domain expertise — not execution instructions. Agents read workflow files in workflows/ for step-by-step process via progressive disclosure.

Overview

Takes a raw project idea through guided discovery, optional market research & monetization analysis, then generates design artifacts and hands off to /bootstrap to wire up the full .claude/ ecosystem.

Phase Map

/kickstart → ideate? → intake → compose → stack → research? → monetize? → brand? → design → handoff → /cks:new → discover

Each phase is independently resumable. The command reads .kickstart/state.md to determine where to resume and dispatches the appropriate agent.

Phase	Agent	Required?	Output
0 — Ideate	kickstart-ideator	Optional	.kickstart/ideation.md
1 — Intake	kickstart-intake	Yes	.kickstart/context.md
1b — Compose	kickstart-intake	Yes	.kickstart/manifest.md
1c — Stack	kickstart-intake	Yes	.kickstart/stack.md
2 — Research	deep-researcher	Optional	.kickstart/research.md
3 — Monetize	monetize-discoverer	Optional	.monetize/
4 — Brand	kickstart-brand	Optional	.kickstart/brand.md
5 — Design	kickstart-designer	Yes	.kickstart/artifacts/
6 — Handoff	kickstart-handoff	Yes	CLAUDE.md, .prd/, scaffold

MANDATORY GATES — READ THIS FIRST

Phase 1b (Compose) is ALWAYS run after Intake. Phases 2 (Research), 3 (Monetize), and 4 (Brand) are OPTIONAL for the USER but the QUESTION is MANDATORY for YOU.

You MUST call AskUserQuestion at each gate. You MUST NOT:

• Decide on the user's behalf whether to skip a phase
• Skip the question because you predict what they will say
• Skip the question because an API key is missing (WebSearch fallback exists)
• Skip the question because you think it's not relevant
• Skip the question because other artifacts already exist (e.g., .prd/ from bootstrap)

The user's explicit response is what drives skip/proceed — never your inference.

Failure mode to avoid: Claude sees no API key or sees .prd/ already exists and decides "they probably don't want this" and skips the AskUserQuestion call entirely. This is WRONG. Always ask. Every time. No exceptions.

STATE FILE ENFORCEMENT

STOP RULE: After completing any phase (including marking it skipped), you MUST:

1. Write/update .kickstart/state.md BEFORE displaying the completion banner
2. Write/update .kickstart/state.md BEFORE starting the next phase
3. Never skip a state file write — the resume system depends on it

If you do not update the state file, the resume system breaks and phases get re-run or skipped incorrectly on the next invocation.

State File Format

For the full state file template and validation rules per phase, read references/validation-and-state.md.

Progress Banners

Display a progress banner at the start of every phase showing all phases and their status:

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 KICKSTART ► {PHASE_NAME}
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

 [0]  Ideate          {✅ done | ▶ current | ○ pending | ⊘ skipped}
 [1]  Intake          {✅ done | ▶ current | ○ pending | ⊘ skipped}
 [1b] Compose        {✅ done | ▶ current | ○ pending | ⊘ skipped}
 [1c] Stack          {✅ done | ▶ current | ○ pending | ⊘ skipped}
 [2]  Research        {✅ done | ▶ current | ○ pending | ⊘ skipped}
 [3]  Monetize        {✅ done | ▶ current | ○ pending | ⊘ skipped}
 [4]  Brand           {✅ done | ▶ current | ○ pending | ⊘ skipped}
 [5]  Design          {✅ done | ▶ current | ○ pending | ⊘ skipped}
 [6]  Handoff         {✅ done | ▶ current | ○ pending | ⊘ skipped}

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

After each phase completes, display:

  [N] {Phase Name}   ✅ done
      Output: {artifact file path(s)}

Multi Sub-Project Rules

Read .kickstart/manifest.md to determine mode:

• Single sub-project: Artifacts written flat to .kickstart/artifacts/
• Multiple sub-projects:
  • Shared artifacts → .kickstart/artifacts/shared/
  • Per-SP artifacts → .kickstart/artifacts/sp-{NN}-{name}/
  • Build order follows dependency graph from manifest

Phase Prerequisites

Phase	Requires
Ideate	Nothing (entry point)
Intake	Nothing (or .kickstart/ideation.md if ideation was run)
Compose	.kickstart/context.md
Stack	.kickstart/context.md + .kickstart/manifest.md
Research	.kickstart/context.md + PERPLEXITY_API_KEY (or deep-research sources)
Monetize	.kickstart/context.md (Perplexity optional — falls back to WebSearch)
Brand	.kickstart/context.md (uses Canva MCP, WebFetch, or manual Q&A)
Design	.kickstart/context.md + .kickstart/manifest.md (research/monetize/brand optional but consumed if present)
Handoff	.kickstart/manifest.md + design artifacts per sub-project

Output Artifacts

File	Purpose
.kickstart/ideation.md	Refined idea pitch from brainstorming (if ideation was run)
.kickstart/state.md	Progress tracker — resume point on interruption
.kickstart/context.md	Full idea context from intake Q&A
.kickstart/manifest.md	Project composition — sub-projects, dependencies, build order
.kickstart/stack.md	Technology stack decisions — frameworks, database, hosting, CI/CD
.kickstart/research.md	Market research with citations (if opted in)
.kickstart/brand.md	Brand guidelines — colors, typography, voice, UI prefs (if opted in)
.kickstart/artifacts/	Design artifacts (ERD, schema, PRD, API, Architecture, Feature Roadmap)
.monetize/*	Monetization artifacts (if opted in)
.prd/PROJECT-MANIFEST.md	Manifest copy for feature lifecycle (created during handoff)

Environment Variables

Variable	Required By	Phase	Fallback
PERPLEXITY_API_KEY	Research (standard), Monetize	2, 3	WebSearch (Claude built-in)

If the key is missing, research and monetize phases automatically fall back to WebSearch.

Error Handling

Failure	Behavior
PERPLEXITY_API_KEY missing	Fall back to WebSearch, note source in output
Perplexity rate limit	Retry once, then save partial results and continue
User abandons mid-intake	Save partial context + state.md, next run offers to resume
/bootstrap not available	Generate artifacts only, print manual bootstrap instructions
Phase validation fails	Retry once, then ask user what to do
Scaffold build fails	Report error, save what was created, continue to next sub-step

Educational Mode

During Q&A, when the user mentions AI/tech concepts, briefly explain using the glossary in references/ai-glossary.md:

> **{Term}** — {one-line definition from glossary}

Only surface when natural — don't force explanations.

Reference Files

File	When to Read
workflows/intake.md	Phase 1 — intake Q&A steps
workflows/compose.md	Phase 1b — sub-project identification
workflows/stack-selection.md	Phase 1c — technology stack & platform selection
workflows/brand.md	Phase 4 — brand extraction steps
workflows/design.md	Phase 5 — artifact generation steps
workflows/handoff.md	Phase 6 — scaffolding steps
workflows/auto-chain.md	After Phase 6 — feature lifecycle handoff
references/validation-and-state.md	Validation rules + state file format
references/phase-banners.md	Sub-step validation banners for Phase 5/6
references/ai-glossary.md	During intake Q&A — surface relevant definitions

Customization

This skill ships with opinionated defaults. Review and adapt to your needs:

• Optional phases: Which phases users are offered (research/monetize/brand) — edit gate questions in agents
• Validation rules: What each phase must produce — edit references/validation-and-state.md
• Design artifacts: Which artifacts are generated per sub-project — edit workflows/design.md
• Auto-chain behavior: What happens after kickstart completes — edit workflows/auto-chain.md
• Educational mode: AI glossary references during Q&A — edit references/ai-glossary.md

Common Rationalizations

Rationalization	Reality
"Let me just start coding"	30 minutes of planning saves 3 hours of rework. Kickstart produces the design artifacts coding needs.
"I'll figure out the architecture as I go"	Architecture decisions made under implementation pressure are consistently worse than planned ones.
"We don't need a design system yet"	Inconsistent UI from day one creates technical debt that compounds with every new screen.
"Research is a waste of time"	Building on wrong assumptions is the real waste. Research validates before you invest implementation time.

Rules

1. Always show the progress banner before starting each phase
2. Always validate output before marking a phase as done
3. Always update state.md after every phase transition
4. Never skip validation — if an artifact is missing, the phase is not done
5. Show completion details — not just "done" but what was produced (counts, summaries)
6. Respect user choices — if they skip research/monetize, mark as skipped, not pending
7. Resume gracefully — on re-run, show full progress and offer to continue