Using the answers

AI coding tools

Give Claude Code, Cursor and other MCP-capable assistants the business context behind the code.

AI coding assistants read code quickly and well. What they cannot read is the reason behind it. Asked to “extend refunds to 45 days”, an assistant changes the number correctly and breaks the reconciliation that depended on it.

om connects to assistants through the Model Context Protocol (MCP), so the reason behind a line arrives with the line. The assistant looks up a rule’s reason, owner and dependants before it edits the code.

Before it deletes a rule

A new joiner asks their assistant to clean up payouts/legacy_hold.ts. It looks dead. The assistant checks with om first:

assistant → om.why("payouts/legacy_hold.ts")
Still used: the settlement job imports it.
Holds COD payouts for 24 hours after delivery.
Reason: fraud is flagged within a day (INC-118).
Owner: Priya S. · approved Jun 2026
assistant → Keeping the file. Asking Priya first.

What the assistant can do

ToolWhat it returns
Look up a ruleThe rule, its reason, status and owner, for a file and line or a name
Check impactEverything that depends on the rule
Find the ownerWho to ask before changing it
Search rulesRules matching a question, with sources

What changes in practice

  • The assistant mentions the approved reason when it proposes a change to a rule.
  • It flags Disputed, Stale or Unknown rules instead of treating the code as the whole truth.
  • It names the owner to involve, so the change is reviewed by the person who decided.

If a change to an approved rule still reaches a pull request, om flags it there and asks the owner before merge. See Pull request guardrail.

Works with Claude Code, Cursor and other MCP-capable assistants.

Was this page helpful?
Esc