Map Systems

Translates the concept doc into a graph of systems with dependencies and a recommended design order. The systems index that this skill writes is the authoritative list /design-system reads to figure out what to write next.

Output: design/systems-index.md

Purpose / When to Run

Run when:

• design/concept.md is complete and approved
• The team needs a single source of truth for what systems exist
• New systems were discovered mid-design and the index needs updating

Argument next is a shortcut: re-reads the index and shows the next system in the priority order without rebuilding the whole list.

Inputs

• design/concept.md (required)
• design/design-bible.md if present
• .claude/docs/technical-preferences.md
• Existing design/systems-index.md if updating

Outputs

• design/systems-index.md

Phase 1: Argument Handling

If argument is next:

1. Read design/systems-index.md
2. Find the highest-priority system with status Not started
3. Print: "Next system to design: <name> (priority <N>, layer <X>). Run /design-system <name> to start."
4. Done — do not rebuild the index.

Otherwise, full mapping flow:

• If design/systems-index.md exists, ask whether to refresh it, append new systems only, or rebuild from scratch.

Phase 2: Read Concept

Read design/concept.md. Extract:

• The one-liner
• The MVP scope (in v1) and cuts (out of v1)
• The primary metric
• The riskiest assumption

If concept doc is missing, stop:

"No design/concept.md found. Run /brainstorm first."

If concept is fragmentary (sections empty), warn: "Concept doc has [N] empty sections. Mapping with incomplete concept will produce a brittle systems list — recommend completing the concept first."

Phase 3: Propose System List

For a typical mobile app, propose a candidate list. Adapt to what the concept says — never copy this list verbatim. Use it as a checklist to make sure no obvious system is forgotten.

Standard candidate categories:

Foundation layer

• App shell / navigation chrome
• Authentication / session
• Persistence (local DB / cache / secure storage)
• Network client / API layer
• Error reporting / crash handling
• Logging
• Analytics
• Feature flags / remote config
• Theme / dark mode
• Localization

Core layer (concept-specific)

• The 1-3 systems that deliver the core JTBD from the concept doc — name them based on the concept

Supporting features

• Profile / account
• Settings
• Notifications (push + in-app)
• Sharing / deep links
• Onboarding
• Help / support
• Search
• Empty / error / offline screens

Monetization

• Paywall (if monetized)
• IAP / subscription management
• Receipt validation
• Restore purchases

Platform integrations

• Camera / media (if used)
• Maps / location (if used)
• Biometrics (if used)
• Background tasks (if used)
• Widgets (if used)
• App Clips / Instant Apps (if used)

Present the proposed list to the user as a draft. Ask:

• Prompt: "Which of these systems are in scope for v1? Mark each."
• Use multi-select if your interface supports it; otherwise iterate quickly with one yes/no per system.

For each in-scope system, also confirm: 1-3 sentence description.

Phase 4: Capture Dependencies

For each in-scope system, ask the user (or infer from the concept):

• Which other systems must exist for this to function (hard dependencies)?
• Which systems would enrich it but are not required (soft dependencies)?

Build a dependency table. Validate:

• No cycles. If a cycle is detected, surface it: "Auth depends on Profile, Profile depends on Auth — break this with a contract or merge them."
• Foundation systems should depend only on other Foundation systems.
• Core systems may depend on Foundation.
• Supporting features may depend on Foundation and Core.

Phase 5: Prioritize Design Order

Topological sort: the design order respects dependencies (upstream first).

Within each layer, prioritize by:

1. Risk — is this on the riskiest-assumption path? Higher priority.
2. MVP weight — how central to the v1 promise.
3. Specialist availability — if a system needs a hard-to-source specialist, schedule it earlier so blockers surface early.

Assign each system a priority bucket: P0 (must design first), P1, P2. P0 should be ≤ 5 systems.

Phase 6: Build the Index

Write design/systems-index.md:

# Systems Index

> **Generated**: <date>
> **Source concept**: `design/concept.md`
> **Total systems**: <N>

## Progress Tracker
- Total: <N>
- Designed: 0
- In review: 0
- Approved: 0

## Design Order

The order below reflects upstream-first decomposition. Run `/design-system <name>` for each, top to bottom.

### Foundation (P0)
| # | System | Layer | Priority | Status | Doc |
|---|--------|-------|----------|--------|-----|
| 1 | navigation-shell | Foundation | P0 | Not started | — |
| 2 | auth | Foundation | P0 | Not started | — |
| 3 | persistence | Foundation | P0 | Not started | — |
| ... |

### Core (P0–P1)
| ... |

### Supporting (P1–P2)
| ... |

### Monetization (P1)
| ... |

## Dependency Graph

auth ├─ persistence ├─ network-client └─ error-reporting

profile → auth, persistence

[etc — render as ASCII tree or simple text]

## Out of v1 Scope (deferred)
- <system>: <reason — typically from concept's cut list>

## Open Mapping Questions
- <any unresolved item from the conversation>

Ask: "May I write this to design/systems-index.md?"

Phase 7: Detect Anti-Patterns

Before finishing, scan the proposed map for common issues:

• Foundation explosion: more than ~10 Foundation systems means the team is over-architecting.
• Core orphans: a Core system with no Foundation dependencies — likely missing a lower-layer dependency.
• Notifications without auth: push systems that don't depend on Auth.
• Paywall without auth: monetization needs identity.
• Single P0 with 6+ dependents: bottleneck — flag as a critical-path risk.
• No analytics in MVP: if the concept's primary metric is engagement-based, analytics must be P0.

Surface each as a callout in the index under "Open Mapping Questions" or as a warning before write.

Phase 8: Hand Off

After writing:

• Print: "Systems index written. Next: /design-system <first P0 system> — author the first PRD."
• Update production/session-state/active.md with the next-system pointer.

Use AskUserQuestion:

• Prompt: "Want to start designing the first system now?"
• Options:
  • Yes — /design-system <name>``
  • Run /design-bible first (recommended if not yet done)
  • Stop here for now

Edge Cases

• Concept implies too many systems for v1: surface the cut list. Suggest deferring 30%+ to v1.1.
• Concept does not describe a core mechanic: stop. The concept is too vague — recommend re-running /brainstorm.
• Updating an existing index: only add new systems and update statuses; never silently delete a row. If a system is being killed, mark it Killed and keep the row.

Quality Gates

• Every in-scope system has: name, layer, priority, dependency list, status.
• The dependency graph is acyclic.
• Foundation systems depend only on other Foundation systems.
• The next-priority system pointer is set.
• Out-of-scope list is non-empty (the concept's cut list must be reflected here).

Examples

For a "bouldering project tracker":

• Foundation: navigation, auth, local persistence (SwiftData), error reporting
• Core: project list, attempt logger, session timer
• Supporting: profile, settings, notifications (session reminders), share-image
• Monetization: deferred to v1.1 (per concept doc cut list)
• P0 design order: auth → persistence → project list → attempt logger