What problem does it solve? Salesforce writes can silently fire Flows, Apex triggers, and validation rules that email customers, create orders, or lock records, and destructive actions like merges and mass deletes cannot be undone. This Skill gives an AI agent the judgment to classify every Salesforce action as read, reversible write, committing write, or destructive before executing it. ## Core Features & Use Cases - Risk classification of operations: Maps Salesforce operation families (stage changes, lead conversion, order activation, merges, mass updates, OWD changes) to read/write/destructive gates with human-approval rules. - Object and state model guidance: Covers leads, accounts, contacts, opportunities, quotes/CPQ, orders, cases, price books, approval processes, and the sharing model (OWD, role hierarchy, profiles, permission sets, FLS). - Edge states and recovery playbooks: Documents 25 gotchas, edge cases like Person Accounts and multi-currency orgs, and recovery paths such as the 15-day Recycle Bin window and amendment orders. - Use Case: Before converting a lead or setting an opportunity to Closed Won, the agent checks wired automation, confirms the org (sandbox vs production), and pauses for human approval because the action fires committing downstream automation. ## Quick Start Ask the agent to update an opportunity stage in Salesforce and it will first read the record, check for wired automation, and classify the change before acting.
