{"id":"agent-sales-pipeline-stage-updates","audience":"sales","category":"agents","track":"agents","technique":"grounding","title":"Brief an agent to update deal stages only on real evidence","free":false,"task":"You want an AI agent to update your CRM from this week's deal notes. Write its brief: mark a deal Closed Won only if there's a signed order form, Closed Lost only if the customer said no in writing, otherwise update the next step and leave the stage alone; never change a deal amount; and send anything unclear to you with ASK_HUMAN. Output exactly 6 numbered plain-text lines: one per deal in order, then DONE.","materialLabel":"Deal notes and actions (attached to your prompt automatically)","material":"Deal notes - week of Nov 2 (rep: Sam Ortiz)\n\nD1 Alder Systems ($24,000, stage: Proposal) - Signed order form received Nov 4 via DocuSign.\nD2 Kestrel Health ($60,000, stage: Negotiation) - Call Nov 3: \"This looks great, we're basically there.\" Legal review starts next week.\nD3 Lumen Bay ($9,000, stage: Proposal) - Email Nov 5 from their CFO: \"We've decided not to move forward this year.\"\nD4 Harbor Bank ($15,000, stage: Demo) - Champion said on a call they might need 30 seats instead of 20, \"so more like $22k.\"\nD5 Corvane ($40,000, stage: Negotiation) - No reply to 3 emails since Oct 20. Sam thinks it's dead.\n\nAvailable actions:\nupdate_stage(deal, stage)\nupdate_next_step(deal, text)\nupdate_amount(deal, amount)\nASK_HUMAN(question)\nDONE(summary)","rules":[{"type":"numberedItems","count":6},{"type":"plainText"}],"promptHints":[{"label":"Requires evidence to close","patterns":["signed","order form","in writing","written"]},{"label":"Forbids amount changes","patterns":["amount"]},{"label":"Handles unclear deals","patterns":["unclear","next step","ask_human"]},{"label":"Rules out made-up data","patterns":["don'?t (invent|guess|make up)","never (invent|guess|make up)","only (use )?(facts|info|data|what)","not in the","no guess"]},{"label":"Shows the action-log format","patterns":["action:","action log","numbered","\\b1\\.","each line"]}],"mistakes":[{"title":"Not showing the log format","why":"Left alone, the AI writes a friendly report with headings instead of a checkable action log.","fix":"Show it literally: \"Format every line like 1. ACTION: tool(args), then ASK_HUMAN: and DONE: lines. Plain text.\"","checks":["numberedItems","plainText"]},{"title":"No evidence rule","why":"The AI marks deals won when the customer sounds happy.","fix":"Add: \"Closed Won only with a signed order form. Closed Lost only with a written no.\"","checks":["r0"]},{"title":"Letting it change amounts","why":"The AI updates the deal to a number mentioned in passing.","fix":"Add: \"Never change a deal amount.\"","checks":["r1"]},{"title":"Treating silence as a no","why":"The rep's hunch makes the AI close out a deal nobody declined.","fix":"Add: \"If it's unclear, update the next step or ASK_HUMAN.\"","checks":["r2"]}],"hint":"Forecasts break when agents read the mood. Tell yours exactly what evidence moves a stage, and what it may never change.","rubricLabels":["D1 is set to Closed Won and D3 to Closed Lost, and D2 is not marked Closed Won","There is no update_amount line (D4's possible $22k is not entered as the amount)","D5 (silent, no written no) is not marked Closed Lost; it gets a next step or ASK_HUMAN, and the last line is a DONE line"]}