IMS — Claude Code Instructions

Project

HSE Incident Management System for Setia Corporation (3PL warehouse). Replaces WhatsApp/Excel safety reporting. Client: Ms. Agnes, Mr. Terence, Mr. Yap, Mr. Jensen.

Stack

Frontend + API routes: Next.js (App Router)
DB + Auth + Storage: Supabase (Postgres + RLS + pgvector)
Hosting: Vercel (frontend) + Supabase cloud
AI: Claude API — server-side only, never client-side
Email: Resend
WhatsApp: Meta WhatsApp Business Cloud API
PDF: pdf-lib (JKKP form fill)
QR: qrcode npm package

Current Phase

Phase 0 — Foundation (not started) See [[phases]] memory for full phase breakdown and deliverables.

Dev Conventions

Never commit .env or API keys. All secrets in env vars only.
RLS enforced at DB level for every table — never rely on UI-only hiding.
Claude API calls: server-side API routes only (/app/api/...).
Incident reference format: SITE-YYYYMM-####
Every DB mutation writes to audit_log (table, record_id, action, changed_by, changed_at, old/new value).
Evidence files: never hard-delete; retain minimum 5 years (DOSH JKKP 8 requirement).
AI suggestions logged to audit_log — capture what Claude suggested + what human chose.

File Structure (target)

IMS/
├── docs/                    # PRD, tech spec, roadmap, specs — source of truth
├── app/                     # Next.js App Router pages + API routes
├── components/              # Shared UI components
├── lib/
│   ├── supabase/            # Supabase client + server helpers
│   ├── claude/              # Claude API wrappers (server-side)
│   └── notifications/       # Resend + WhatsApp senders
├── supabase/
│   └── migrations/          # SQL migration files
└── public/

Deep Context (memory files — load instead of re-reading raw docs)

[[project_overview]] — objectives, roles, success metrics
[[architecture]] — module diagram, stack rationale, deployment
[[db_schema]] — all 10 tables, fields, RLS, pgvector
[[phases]] — 4 phases + deliverables, current phase
[[compliance]] — NADOPOD 2004, JKKP 6/7/8 rules

graphify

When graphify-out/graph.json exists:

Run graphify query "<question>" before browsing source.
Use graphify path "<A>" "<B>" for relationships.
Use graphify explain "<concept>" for focused concepts.
After code changes: graphify update .

2.4 KiB Raw Blame History