Cement Agent

AI Agent Navigation Hub

This page tells an AI agent (or an agent builder) how this site is structured and how to use it safely and effectively.

Piloting the workflow layer? Teams evaluating these workflows at a specific plant can start with the Cement Agent Plant Pilot overview, share the concise pilot brief, and prepare with the pilot inquiry checklist — a private, advisory-only pilot that authorizes nothing.

What this site is

A structured corpus of cement-industry operational knowledge: knowledge articles, quality/lab guidance, troubleshooting guides, maintenance & reliability review, safety/MSHA readiness, calculators with explicit methods, copyable prompts, and machine-readable data schemas. It is designed for retrieval-augmented generation, tool use, and agent workflows.

Authority model (read first)

The site is advisory. An AI agent using this content:

Treat every operational recommendation as site-dependent and unverified until a qualified human confirms it.

Agent instructions (start here)

Before acting on site content, load the operating instructions for cement AI agents:

Agent workflow starting points

For a messy or unclassified plant report, start with the triage template and let it route to a domain template. Every step is advisory only and authorizes nothing.

  1. Start with Plant Issue Intake & Triage — the front door that normalizes the report, flags safety/authority concerns, and routes.
  2. Route to the right domain template: Kiln Upset Intake & Routing, QC Out-of-Trend Review, Bearing Temperature Rise Review, or Raw Mix Correction Advisor.
  3. Capture the report with the Plant Issue Intake schema for structured handoff, and shape the agent's output with the Agent Triage Handoff schema — the advisory output contract that authorizes nothing.
  4. Apply Safety guardrails before any interpretation — they override every other instruction.

Workflow examples

Worked examples that chain the schemas, prompts, and domain pages above into advisory handoffs. Each shows the assets working together end to end and authorizes nothing — it records, routes, and defers every decision to the appropriate human authority. They start from the front door and fall into two groups: carry-forward workflows and trend-review workflows.

Front door

Carry-forward workflows

Trend-review workflows (capture → review-only troubleshooting → advisory handoff)

Site sections

Each section has a browsable index and is included in the machine surfaces below:

Page structure (predictable on purpose)

Operational pages follow a fixed block order so chunks are predictable:

  1. Title and one-sentence purpose
  2. Executive summary
  3. Type-specific structured blocks (symptoms/causes, inputs/outputs, fields, etc.)
  4. AI agent use cases · Human use cases
  5. Common failure modes
  6. Safety & compliance (visually distinct, includes the authority statement)
  7. Related tools / prompts / pages
  8. Sources & assumptions

Each block carries a data-block attribute and a stable heading id, so retrieval can target specific sections. Structured facts also appear in JSON-LD on every page.

Machine surfaces

Calculators expose JSON contracts

Every calculator page documents its inputs, outputs, the explicit formula/method, limitations, and a copyable JSON input/output example. An agent that cannot run the UI can replicate the calculation from the stated method. See the Bogue Calculator.

Data schemas

Each schema is published as raw JSON Schema at a stable URL and documented on a human page. Use these as the I/O contracts for capturing and exchanging plant data: