| name | spec-writing |
| description | Guide for creating high-quality specifications for sandbox. Use when starting a new SPEC or reviewing specs against architectural standards. |
Spec Writing & Review
Design and review specifications (SPECs) against Open Mercato architecture and quality rules.
Workflow
- Load Context: Read
AGENTS.md for module conventions and .ai/specs/ for existing specs.
- Initialize: Create
{date}-{title}.md in .ai/specs/.
- Start Minimal: Write a Skeleton Spec (TLDR + 2-3 key sections). Do NOT write the full spec in one pass.
- Scan for critical unknowns — decisions that block data model, scope, or architecture.
- If unknowns exist, add a numbered Open Questions block (
Q1, Q2, …) after the TLDR.
- STOP after presenting the skeleton. Do not proceed until the user answers all questions.
- Iterate: Apply answers, remove Open Questions block. Repeat if new unknowns surface.
- Research: Challenge requirements against open-source market leaders.
- Design: Create architecture, data models, API contracts.
- Implementation Breakdown: Break into Phases (stories) and Steps (testable tasks).
- Review: Apply the Spec Checklist.
- Output: Finalize the specification file.
Output Formats
1. New Specification
Use the Specification Template. Adapt if needed, but ensure core concerns are addressed.
Required sections: TLDR, Problem Statement, Proposed Solution, Data Models, API Contracts, Risks, Changelog.
2. Architectural Review
# Architectural Review: {SPEC-0XX: Title}
## Summary
{1-3 sentences: what the spec proposes and overall health}
## Findings
### Critical
{Cross-module ORM, tenant isolation leaks, missing auth guards}
### High
{Missing undo logic, incorrect module placement, missing phase strategy}
### Medium
{Missing failure scenarios, inconsistent terminology}
### Low
{Style suggestions, nits}
Review Heuristics
- Command Graph vs. Independent Ops: Graph Save (coupled calculation) or Compound Command (independent steps)?
- Architectural Diff: Cut standard CRUD noise. Focus on what's unique.
- Singularity Law: Singular naming for entities, commands, events, feature IDs.
- Undo Contract: Is the "Undo" logic as detailed as the "Execute"?
- Module Isolation: Using Event Bus for side effects or cheating with direct imports?
- Canonical Mechanisms: Does the spec reach for the framework primitives (
makeCrudRoute, <CrudForm>, <DataTable>, apiCall / useGuardedMutation, DI-resolved cache, createModuleEvents) or invent its own substitute? See AGENTS.md → Mandatory Module Mechanisms for the full canon and links. No raw fetch, no raw <form>, no new Redis(...), no manual cross-module ORM joins.
- Sensitive Data: For every PII / GDPR / address / contact / free-text-about-people / integration-credential column the spec proposes, does it declare an
encryption.ts defaultEncryptionMaps entry and route reads through findWithDecryption? See AGENTS.md → CRITICAL Rule #11 (Encryption maps) + the "Encryption maps for sensitive data" row of the Mandatory Module Mechanisms table and .ai/skills/data-model-design/SKILL.md § Sensitive Data and Encryption Maps. No hand-rolled AES, no crypto.subtle, no "TODO encrypt later".
- Design System: Does every UI mock / className snippet in the spec match the DS canon — semantic status tokens (no
text-red-* / bg-green-*), Tailwind text scale (no text-[11px] / text-[13px]), shared primitives (StatusBadge, Alert, FormField, SectionHeader, CollapsibleSection, LoadingMessage / Spinner / DataLoader, EmptyState), lucide-react icons in page body (never inline <svg>), dialog Cmd/Ctrl+Enter submit + Escape cancel, aria-label on every icon-only button? See AGENTS.md → CRITICAL Rule #10 (Strict Design System alignment for every UI change) and .ai/skills/backend-ui-design/SKILL.md. Specs that touch existing pages MUST honour the Boy Scout rule.
Quick Rule Reference
- Singular naming for entities, commands, events, feature IDs.
- FK IDs only for cross-module links — no ORM relationships.
organization_id is mandatory for all tenant-scoped entities.
- Undoability is the default for state changes.
- Zod validation for all API inputs.
- Encryption maps for every sensitive / GDPR-relevant column (declare in
<module>/encryption.ts, read via findWithDecryption) — see AGENTS.md → Data Encryption.
- Canonical primitives for CRUD APIs (
makeCrudRoute), backend forms (CrudForm), tables (DataTable), HTTP (apiCall — never raw fetch), non-CrudForm writes (useGuardedMutation), cache (DI-resolved @open-mercato/cache), events (createModuleEvents) — see AGENTS.md → Mandatory Module Mechanisms.
- Design System tokens and shared UI primitives — no hardcoded status colors, no arbitrary text sizes, no inline
<svg> in page-body UI. See AGENTS.md → Design System.
Reference Materials