Free tool · Kris McCabe
Before an AI agent touches production Salesforce data, dry-run it here. Drop a Describe payload, dial in the action, and get a Blast Radius score, an Irreversibility Index, and a guardrail policy you can hand straight to Agentforce, LangChain, or a system prompt — every calculation stays in this browser tab.
Drop a Salesforce SObject Describe JSON (or paste one below). Parsed entirely in your browser — nothing is uploaded.
Drop a Describe JSON file here, or click to browse.
What is the agent about to do, and how?
Records touched in a single execution.
Module D — pulling Describe metadata directly from a Salesforce org via OAuth 2.0 PKCE — needs a registered Connected App (a real Client ID and callback URL) before it can run for real. Rather than fake it, it's disabled until that's set up. Everything above works fully offline with a Describe payload you already have.
Load a payload and set the action above — this updates instantly as you change anything.
Agent-Impact is a static, client-side simulator. There is no backend for it to call.
The optional live Salesforce connector (not yet enabled — see the panel on the page) would request only the id and api OAuth scopes and only ever call read-only Describe/Tooling endpoints. It would never write to your org.
Blast Radius, Irreversibility, and the resulting risk tier are heuristic estimates built from your inputs and generic, publicly documented Salesforce metadata concepts (cascade-delete relationships, record-triggered flows, bulk API batching). They're a starting point for a guardrail conversation, not a certified risk assessment, and not derived from any employer's internal systems or proprietary data.
This is an independent tool built by Kris McCabe on his own time. It isn't affiliated with, endorsed by, or built using confidential material from Salesforce, Anthropic, or any employer, past or present.