--- title: "Workplace Safety Incident Review With AI Agents | Kylon" description: "How a construction safety desk dispositions incidents, spot-checks the draft report and applies narrow notices without side effects in one Kylon room." url: "https://kylon.io/use-cases/workplace-safety-incidents" language: "en" --- [HOME](https://kylon.io/)\>[Use Cases](https://kylon.io/use-cases)\>Workplace Safety Incidents: A Quarter of Dispositions That Change Only for a Reason Use Case8 min read # Workplace Safety Incidents: A Quarter of Dispositions That Change Only for a Reason Incident dispositions, the draft report spot-check, class upgrades, site corrections, hotline duplicates and corrective action closures run through one Kylon room. Agents work the SharePoint incident list, and the safety manager approves each change against its scope. [Get started](https://app.kylon.io/sign-up) Kylon TeamProduct November 6, 2026 ## Every incident classified once, and reclassified only for a reason A quarter of safety incidents has to end with one disposition per incident. A first aid case stays off the recordable count, a medical treatment or lost time case goes on it, an injury to a subcontractor's worker is routed to that contractor's own report, and a second report of the same event is marked as a duplicate of the first. Lost days follow the class. The order of the rules matters, because one incident can match more than one. Incidents live in a SharePoint list, corrective actions are ClickUp items, incident forms and photos are filed in Box, and the policy and every notice for the quarter sit in a Drive folder. Site leads ask questions by email and the field posts in Slack. The first pass is not the hard part. The hard part is the stream of notices that each change one field and should leave every other field exactly where it was. ## Who works in the room People - Safety manager Approves every disposition and field change, sends replies to site leads and signs the quarter - Site safety leads Report incidents from their sites and ask about counts and classifications - Corrective action owners Carry out corrective actions in the field and confirm when each one is done Agents - AdaIncident classifier Applies the precedence list to every incident using its reported class, contractor flag and duplicate reference, and proposes disposition and lost days SharePointBoxGoogle Drive - MycoReports and site follow-up Spot-checks the draft incident report, keeps corrective actions in ClickUp in step and drafts replies on site lead threads ClickUpGoogle Drive[Gmail](https://kylon.io/integrations/gmail) [Slack](https://kylon.io/integrations/slack) Anything about an incident's class, disposition or lost days goes to Ada, the safety room's default agent. Myco owns the draft report, the corrective actions and the site lead threads. When Ada finishes re-running a batch, it hands any incident a site lead has asked about to Myco in the same thread, and Myco drafts the reply on that lead's existing email thread rather than starting a new one. ## From first disposition to quarter close The precedence list and recording rules are in the run folder and in the room guidelines, so Ada and Myco apply the same order. Every write to SharePoint or ClickUp waits for the safety manager. ```mermaid flowchart TD disp["Ada applies the precedence list,
proposes disposition and lost days"] ok1["Safety manager approves
before SharePoint changes"] spot["Myco spot-checks the named
incidents in the draft report"] okspot["Safety manager accepts
the corrections"] med([A medical update or a hotline
cross-reference arrives]) boundary([A site boundary
correction arrives]) rerun["Ada re-runs each named incident
against the full precedence list"] site["Ada corrects only the site
field on the named incidents"] reply["Myco drafts the reply on the
site lead's own thread"] oksite["Safety manager approves"] send["Safety manager approves
and sends the reply"] ca["Myco proposes the ClickUp closures
and one report item per site"] final["Safety manager approves each
closure and signs the quarter"] disp --> ok1 --> spot --> okspot okspot --> rerun & site med -.- rerun boundary -.- site rerun --> reply --> send --> ca site --> oksite --> ca ca --> final class disp,spot,rerun,site,reply,ca agent class ok1,okspot,oksite,send,final person class med,boundary event classDef agent fill:#FFFFFF,stroke:#DCD3C8,color:#1B1712 classDef person fill:#EEF0E3,stroke:#6E7447,color:#1B1712 classDef event fill:#F4ECDD,stroke:#F4ECDD,color:#564C42 ``` | Stage | What happens | Who acts | Who decides | | --- | --- | --- | --- | | Disposition | Each incident is classified from its reported class, contractor flag and duplicate reference, with lost days and corrective action status | Ada, from SharePoint and Box | The safety manager approves before SharePoint changes | | Draft report spot-check | Nine named incidents get a verdict line by line, other lines come off and missing ones are listed | Myco | The safety manager accepts the corrections | | Medical updates | New classes and lost days are recorded, and the named incidents are re-run against the full precedence list | Ada, then Myco for the site lead reply | The safety manager approves and sends the reply | | Site boundary correction | The site field is corrected on the named incidents and nothing else moves | Ada | The safety manager approves | | Hotline cross-reference | Hotline references are recorded and the named incidents become duplicates | Ada, then Myco for the reply | The safety manager approves and sends the reply | | Corrective actions and close-out | Named ClickUp items close with their incidents' action status, then one report item per site is created | Myco | The safety manager approves each closure and signs the quarter | Stages of a quarterly incident review. Incident numbers, sites and lost days in this example are illustrative. ## Notices that change one field and leave the rest alone Each notice in the quarter has a narrow scope, and most of the care goes into keeping it narrow. A site boundary correction moves eleven incidents from one site to its neighbor, and that is all it does: no disposition changes because of it. A corrective action closure changes the status of each named ClickUp item and the matching status on the incident, and nothing about the incident's disposition, lost days or site. Ada states the scope in the thread before proposing any change, and the safety manager approves against it. The medical update and the hotline cross-reference work the other way. They change what an incident is, so each named incident is re-run against the full precedence list rather than patched in place. A site lead who asks why an incident dropped off the recordable count gets one reply on the thread they opened, with the incident number and the hotline reference now on file. Every agent turn that wrote to SharePoint keeps its run history, which is what the safety manager opens when a count is questioned. ## What the checks catch - A disposition changed during a site boundary correction, when the notice only moves the site field - A corrective action closure that also edited lost days or the disposition of the linked incident - A spot-check report line for an incident outside the nine named, which comes off whatever it says - A class upgrade patched on the class field alone, without re-running the incident against the full precedence list - A reply to a site lead started as a new email instead of on the thread the lead opened ## What stays with people - Approving every disposition and field change before SharePoint or ClickUp is updated - Deciding classification questions the precedence list does not settle - The reply to each site lead, which Myco drafts on the lead's thread and the safety manager sends - Signing the quarter and its per-site reports ## Integrations in this workflow Kylon is not an EHS management system, and it does not take the place of one. It sits across the safety desk's tools and holds the coordination between them. - Box - ClickUp - [Gmail](https://kylon.io/integrations/gmail) - Google Drive - SharePoint - [Slack](https://kylon.io/integrations/slack) - **SharePoint** holds the incident list with class, disposition, lost days, site and duplicate reference - **ClickUp** tracks corrective actions and the per-site report items created at close-out - **Box** holds incident forms, photos and witness statements - **Gmail** carries replies on site lead threads, each one sent by the safety manager - **Google Drive** keeps the precedence list, the quarter's notices and the draft incident report - **Slack** is where superintendents and foremen outside the room post updates from site ## Where to start One quarter is the natural first cycle. Connect the SharePoint list that holds your incidents, and write one rule into the room guidelines: every notice states which fields it may change, and the safety manager approves each change against that scope. > “Each notice changes what it says it changes, and the thread shows that nothing else moved.” ## Set it up with your agent Open a new Room, add your agent and paste this prompt. It connects the tools, builds the App with you and asks for your rules one step at a time, and nothing goes out until a person approves it. Set-up prompt Set up our workplace safety incidents desk in this Room. Take it one step at a time and wait for my OK before you move on. 1. Connect Box, ClickUp, Gmail, Google Drive, SharePoint and Slack. Tell me which ones still need connecting. 2. Ask me who works this desk: the safety manager, the site safety leads and the corrective action owners. Help me invite them to this Room. 3. Build a Safety Incidents App with Incident, Site, Disposition and Status, one record per incident. Pull our current incidents from the SharePoint list. 4. Classify each incident from its reported class, contractor flag and duplicate reference using our rules from the Drive folder. Propose a disposition and lost days for each. Show me any that need a second look. 5. For each notice, state which fields it may change before proposing anything. Check that no other field moves. Flag any change outside the stated scope and hold it. 6. Draft replies to site leads on the thread each lead opened. Draft corrective action closures in ClickUp. Hold every SharePoint write and every ClickUp update until the safety manager approves each one and sends each reply. 7. Write our rules, the scope rule for notices and who approves what into this Room's guideline so every new notice follows the same path. 8. After our period closes cleanly, offer to turn the notice review into an Automation. Before you start, ask me for our rules one at a time. [Get started](https://app.kylon.io/sign-up) Your first company harness. Where humans and agents run your business together. [Get started](https://app.kylon.io/sign-up) ## More Use Cases [View all](https://kylon.io/use-cases) [ ### Construction Submittals: A Log, Follow-Ups and Architect Replies That Stay in Step Submittal dispositions, PM follow-ups, draft log reviews, new drawing sets and architect response batches run through one Kylon room. Agents work the SharePoint log against Box and ClickUp, and the project manager approves every change before it is written. Construction & engineeringApprovals & reviews ](https://kylon.io/use-cases/construction-submittals)[ ### Permit Application Review: Permits Issued When Documents, Fees and Inspections Agree A permit issues only when documents, fee receipts and inspection confirmations agree. In one Kylon room, agents check all three and recompute fees when a valuation changes, and the intake supervisor approves every disposition and fee. Construction & engineeringApprovals & reviewsCompliance & risk ](https://kylon.io/use-cases/permit-application-review)