Near-Miss & Hazard Observation

The problem this kills

The injury that finally happens almost always had warnings: the spill nobody logged, the guard somebody propped open, the forklift that "almost" got someone. Those near-misses are the leading indicators of a real incident - and most plants lose them. They live in a head, a hallway conversation, or a clipboard nobody reads.

Why? Because reporting is a chore. A paper form bolted to a wall, a clunky portal that needs a desktop login, an email to "Safety" that disappears. So people shrug and move on - until the warning becomes a recordable.

This tool flips that. Reporting takes 30 seconds from a phone: what you saw, where, a photo, submit. Then every observation gets triaged, gets a corrective action with a named owner, gets closed out, and gets counted - so your hot areas light up and the fixes actually happen.

What you'll build

A simple, mobile-first web app, tailored to your plant, that does this:

• Frictionless submit. Any worker opens a link on their phone, picks an area, taps the hazard type, snaps a photo, types a sentence, and submits. No app store, no training.
• Triage queue. Supervisors and EHS leads see new observations, sort by area or severity, and review each one.
• Corrective action + owner. The lead approves a fix and assigns an owner with a due date. Nothing gets "written to the record" until a human approves it.
• Closure confirmation. The owner does the fix; the lead confirms it's actually closed.
• Trends. A dashboard shows observations by area, by hazard type, and open vs. closed over time - so you can see where the heat is.
• Weekly digest. A clean email summary goes to your EHS leads via Resend.
• Export. Download the full observation log as a CSV anytime.

What's inside the Implementation Plan

A start-to-finish runbook you paste into an AI coding agent (Claude Code). You don't write code - you answer questions and approve what it builds.

• It starts by interviewing you about your plant. This is the key difference from a generic template: before it builds anything, the plan makes the AI ask about your areas, your hazard types, your severity scale, who triages, your closure rules, and your messy edge cases - then it shapes the data model and the whole tool around your answers and reads a short spec back to you for a thumbs-up.
• Step-by-step build prompts. Each step ends with a ready-to-copy prompt. You paste, it builds, you check.
• Governance baked in (see below).
• A "No API yet?" fallback so you can launch today even if you have no system to connect to: import your area list from a Google Sheet or CSV, and export a clean observation-log CSV in the exact columns your EHS records expect.

The governance it includes (this is the point)

This isn't a toy form. The plan builds in the controls a real safety program needs:

• Login so only your team can use the tool.
• Row-level security so people only ever see their own plant's / organization's data.
• A complete audit trail - who submitted, who triaged, who approved the fix, who confirmed closure, and exactly when.
• A human-in-the-loop approval gate - the form captures the report and the AI can suggest a category, but a supervisor or EHS lead reviews and approves the corrective action and the closure. Nothing is committed to the record automatically.
• Duplicate guards - the same observation can't be processed twice (a unique observation ID is the dedupe key).

Who it's for

• Every employee on the floor - the reporters. They need it dead simple.
• Supervisors - who triage and own fixes in their area.
• EHS leads / safety managers - who set the rules, confirm closures, and watch the trends.

If you run safety for a plant, a warehouse, or a production line and you've ever wished near-misses just showed up in one place that people actually use - this is for you. No developer required.

You've got this - paste the first prompt and let's build it.