Prompt playbook for AI coding

Reusable prompts that keep AI coding aligned with FamilyCo docs.

These prompts keep generated code consistent with FamilyCo’s architecture and policies. Copy, fill in the blanks, and use them with Cursor, Claude Code, Windsurf, or any other agent IDE.

Master prompt for implementation

You are implementing FamilyCo.
Read docs in this order:
1. /concepts/architecture
2. /concepts/domain-model
3. relevant feature docs only
4. /guides/coding-rules

Task:
[DESCRIBE THE IMPLEMENTATION TASK]

Constraints:
- Tech stack: Electron, Vue 3, PNPM monorepo, Prisma, SQLite.
- Do not change unrelated files.
- Reuse existing patterns before inventing new ones.
- If requirements are unclear, state assumptions explicitly.

Output format:
1. Files to change
2. Implementation plan
3. Code
4. Edge cases handled
5. Tests

New module screen

Implement the [MODULE] screen for FamilyCo.
Read:
- /modules/[module]
- /guides/coding-rules

Goal:
[GOAL]

Deliver:
- page component
- child components
- composable / service layer if needed
- typed mock data or real data adapter
- loading, empty, error states

Server use-case

Implement the server use-case: [USE_CASE].
Read:
- /concepts/architecture
- /concepts/domain-model
- /concepts/agent-operating-model
- /concepts/approval-policy
- /guides/coding-rules

Requirements:
- emit audit log,
- enforce approval policy,
- track budget impact if a provider call exists,
- return a typed result.

Prisma schema change

Update the Prisma schema for [FEATURE].
Read:
- /concepts/domain-model
- /guides/coding-rules

Deliver:
- schema changes,
- migration,
- seed adjustment if required,
- a note about affected services.

Refactor without drift

Refactor [TARGET] in FamilyCo.
Do not change behavior.
Read only the minimum relevant docs and current implementation.
Keep public contracts stable unless explicitly requested.
Return a short drift-risk checklist before code.

Token-saving tips

• Quote only the exact requirement lines needed.
• Reference page paths instead of pasting whole docs.
• Work one feature at a time.
• Summarize prior context every 3–5 turns.
• Move long examples into files instead of the prompt body.